Frameworks

Frameworks I build with

Three sets. AI-native methods from products I built, the product management frameworks I rely on, and the organizational and operating frameworks behind how teams decide and run, each shown through a real organization that used it well. Then how I measure KPIs. Every framework ships with a template you can copy into a doc and use today.

MeasurementHow I measure KPIsMy metric tree, review rhythm and formulas, with the companies whose practice it follows.
AI-native

AI-native frameworks

Methods from products I built, for teams where agents do the work and people keep the judgment.

The problem

Without a gate, one bad draft reaches a customer before anyone sees it, and trust in the whole system goes with it. With a gate that never loosens, the reviewer becomes the bottleneck and people start approving without reading.

How to run it

  1. Split reads from writes

    List every action each agent can take. Reads run freely, and every write (send, post, pay, change a record) becomes a proposal that waits in a queue.

  2. Put the evidence on the card

    Each proposal shows the draft, the facts it used with links to their source, and approve, edit and reject in one click. Nothing is written until the person confirms.

  3. Commit exactly once

    Claim the approval inside a database transaction, so a double click or a retried request fires the effect one time. A rejected proposal never changes state.

  4. Start every write on approve to send

    Give each action type a rung: draft only, approve to send, auto after 24 hours unless cancelled, or autonomous. No write agent starts autonomous.

  5. Demote fast, promote on purpose

    Two consecutive rejects step that action type down one rung. Approvals never promote on their own; moving up is a decision the owner makes and logs.

  6. Keep some actions off the ladder

    Name the actions that always need a person however well the agent does, such as clearing a conflict of interest or giving advice that needs a license.

Done when

  • no code path lets an agent write, send or pay without a committed proposal.
  • a double click and a replayed request on one approval each produce exactly one effect, proven by a test.
  • every action type has a recorded rung, and a test shows two consecutive rejects demote it.
  • each proposal card links every figure in the draft to the record it came from.

How teams get it wrong

  • Teams gate the whole agent instead of each action type, so one risky action keeps every safe one stuck in review; give each action type its own rung.
  • The approve click and the send run as two separate steps, so a retry sends twice; claim the decision and fire the effect in one transaction.
  • Approvals quietly promote the agent until it runs unattended; make every promotion a logged decision by the owner.
Template Markdown, ready to paste into a doc
# Agent approval register

Agent: ____
Approver (named person): ____
Review queue lives in: ____
Date set up: ____

## Action types
| Action | Read or write | Starting rung | Always needs a person | Sources shown on the card |
|---|---|---|---|---|
| ____ | ____ | approve to send | yes / no | ____ |
| ____ | ____ | approve to send | yes / no | ____ |
| ____ | ____ | draft only | yes / no | ____ |

Rungs: draft only, approve to send, auto after 24 hours unless cancelled, autonomous.

## Rules
- Demote one rung after ____ consecutive rejects (default 2).
- Promote only when the unedited approval rate holds above ____% for ____ days. Log who decided.
- Actions that never leave approve to send: ____

## The proposal card shows
- [ ] The full draft, editable
- [ ] Every figure linked to its source record
- [ ] Approve, edit and reject in one click
- [ ] One line on why the agent proposed it

## Tests before launch
- [ ] A double click on approve fires the effect once
- [ ] A replayed approval request fires the effect once
- [ ] A rejected proposal changes nothing
- [ ] Two rejects in a row demote the rung

## Promotion log
| Date | Action | From rung | To rung | Decided by | Evidence |
|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ | ____ |

The problem

A model asked to decide gives different answers to the same input, and nobody can explain why a lead was rejected or a number moved. A rule written only in a prompt is a request, and an odd input or a model update can talk past it.

How to run it

  1. Split the feature into words and decisions

    Write two lists. Language work covers understanding the caller, extracting fields and drafting prose. Decisions cover who qualifies, what it costs, which week runs short and what gets sent.

  2. Move every decision into a pure function

    Each decision becomes code with no I/O and a test table of hand-calculated cases, edge cases included. The server action only reads, calls the function and writes.

  3. Check the model at the boundary

    Give the model a schema and validate its output before use. Malformed output is retried or dropped, and never saved.

  4. Hand the model finished numbers

    The model explains figures the code already computed. Scan its output for any figure that is not in its inputs and reject the draft if one appears.

  5. Write hard rules as code

    Rules such as never giving legal advice or never clearing a conflict live in code paths with tests, so no prompt or input can switch them off.

  6. Ship a fallback with every model call

    Each call has a timeout and a deterministic path, such as a template, so the feature still does its core job with no API key or during a provider outage.

Done when

  • the same inputs produce the same decisions on ten repeat runs.
  • every decision function has a fixture test with hand-calculated expected values.
  • a test shows malformed model output is never written to the database.
  • the feature completes its core job with the model API key removed.

How teams get it wrong

  • The prompt says to qualify only certain callers and the team treats that as enforcement; move the rule into code and test it.
  • The model writes fresh test questions or criteria on every run, so results drift for no real reason; fix the inputs with rules and run extraction at zero temperature with a fixed seed.
  • The fallback is written last and never exercised; force a provider outage in CI and assert the output.
Template Markdown, ready to paste into a doc
# Words and decisions split: ____ (feature)

## Language work the model does
- ____ (for example: understand the caller)
- ____ (for example: extract the intake fields)
- ____ (for example: draft the reply)

## Decisions code makes
| Decision | Function | Hand-calculated fixtures | Owner |
|---|---|---|---|
| ____ | ____ | ____ cases | ____ |
| ____ | ____ | ____ cases | ____ |
| ____ | ____ | ____ cases | ____ |

## Model boundary
Output schema: ____
On malformed output: retry ____ times, then ____ (never save it)
Figures the model may mention: only ____
Temperature and seed for extraction: ____

## Hard rules in code
- [ ] ____ (for example: never gives legal advice)
- [ ] ____ (for example: flags a conflict, never clears it)

## Fallbacks
| Model call | Timeout | Deterministic fallback | Tested with key removed |
|---|---|---|---|
| ____ | ____ s | ____ | yes / no |
| ____ | ____ s | ____ | yes / no |

## Done checks
- [ ] Same inputs, ten runs, same decisions
- [ ] Malformed output never written, proven by a test
- [ ] Core job works with the API key removed

The problem

Models write fluent claims that no record supports, and a reader cannot tell the invented line from the true one. One wrong dollar figure in a customer email costs more trust than a hundred correct ones earn.

How to run it

  1. Define what counts as a source

    Name the record types a claim may cite, such as a commit, an invoice, a collected signal or a document section, and give each a stable id.

  2. Store source ids with every claim

    Each drafted claim carries the ids it rests on. A claim with an empty list is dropped before review.

  3. Rebind after drafting

    Look up every cited id again in the set you actually collected. Remove any claim whose id does not resolve, and recompute scores from the surviving evidence only.

  4. Check every figure

    Pull every number and amount out of the draft and confirm each one appears in the source facts, ignoring formatting. If one does not, reject the draft and fall back to a plain template.

  5. Label how strong the evidence is

    Show a tier beside each claim: verified from a primary record, corroborated by a second source, or inferred. Never upgrade an unknown.

  6. Show the receipt

    Link each claim to its source in the interface, and let every total open into the line items that add up to it.

Done when

  • a test that injects a fake source id shows the claim removed before review.
  • a test shows a draft with an invented dollar figure is rejected, including figures written in words.
  • every published claim shows its tier and opens its source in one click.
  • each headline total reconciles exactly to its listed line items.

How teams get it wrong

  • Teams ask the model to include sources and trust the links it writes; resolve every id against records you collected yourself.
  • The figure check only catches amounts with a dollar sign, and the model writes 500 dollars; test every way a figure can be written.
  • Thin evidence is shown with the same confidence as strong evidence; show the tier and pull thin evidence toward the base rate.
Template Markdown, ready to paste into a doc
# Source register: ____ (feature)

## What counts as a source
| Source type | Id format | Collected by | Default tier |
|---|---|---|---|
| ____ | ____ | ____ | verified / corroborated / inferred |
| ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ |

## Claim rules
- Every claim stores its text, source ids, tier and drafted time.
- A claim with no source id is dropped before review.
- Cited ids are looked up again in: ____ (the store you collected)
- Scores are recomputed from surviving evidence only.

## Figure check
Figures a draft may contain come from: ____
Formats the check must catch:
- [ ] $1,234.56 and $ 1234
- [ ] 500 dollars and USD 2,000
- [ ] Percentages and dates: ____
On a failed check, reject the draft and use: ____ (template)

## Reader view
- [ ] Each claim links to its source
- [ ] Each claim shows its tier
- [ ] Each total opens into its line items and reconciles exactly

## Tests
- [ ] A fake source id is removed
- [ ] An invented figure rejects the draft
- [ ] Grounding rate on the eval set is 100%

The problem

Without a fixed eval set, a prompt tweak that fixes one case quietly breaks five others, and nobody notices until a user does. Teams also test only the cases where the model should produce something, so it learns to always produce something.

How to run it

  1. Write the eval set in week one

    Hand-label cases before you tune a prompt. Cover both sides of the boundary, including cases where the correct output is nothing.

  2. Mark each metric as gate, floor or report

    Gates hold exactly, such as 100% grounding and zero banned phrases. Floors hold a minimum, such as an 85% clean-draft rate. Everything else is reported.

  3. Stand in for the human metric

    Your real target is something like drafts approved without edits. Write a deterministic proxy CI can check: a length band, a concrete token from the evidence, no banned phrase and no near-duplicates.

  4. Make it run anywhere

    Run the rails against a deterministic path so the eval runs offline and identically in CI, and run the live model path on a schedule.

  5. Fail the build

    The eval exits non-zero on any gate breach, any floor miss, or any case that behaves differently from its label, and CI blocks the merge.

  6. Grow it from real misses

    Every bad output a user reports becomes a labeled case. Raise the floors as the set grows.

Done when

  • the eval set includes labeled cases where the correct output is nothing.
  • a planted regression, such as a banned phrase in a template, fails CI.
  • the eval runs in CI with no model API key.
  • each metric is written down as a gate, a floor or a report, with its number.

How teams get it wrong

  • The eval set holds only happy cases, so the model passes by always answering; label silent cases on purpose.
  • The eval prints a score but never fails the build; make it exit non-zero and wire it into CI.
  • Teams tune the prompt until the set passes and then stop adding cases; add every production miss as a new case.
Template Markdown, ready to paste into a doc
# Eval gate: ____ (AI feature)

Owner: ____
Runs on every pull request, plus a live model run every ____

## Cases
| Group | Count | Expected behavior |
|---|---|---|
| Must produce output | ____ | grounded, above the threshold |
| Must stay silent | ____ | no output |
| Edge and abuse cases | ____ | ____ |
| From production misses | ____ | ____ |

## Metrics
| Metric | Type | Bar |
|---|---|---|
| Grounding rate | gate | 100% |
| Banned-phrase leaks | gate | 0 |
| Cases matching their label | gate | all |
| Clean-draft rate | floor | ____% |
| ____ | report | ____ |

## Clean-draft checks (proxy for approved unedited)
- [ ] Length between ____ and ____ characters
- [ ] Contains a concrete token from its evidence
- [ ] No banned phrase
- [ ] Not a near-duplicate of another draft

## Wiring
- [ ] Runs with no API key on a deterministic path
- [ ] Exits non-zero on any gate or floor miss
- [ ] CI blocks the merge on failure
- [ ] A planted regression proved it fails

## Human target
Drafts approved without edits: ____%, measured from ____

The problem

Many flows send the email or push to the CRM first and save later, so when a provider is down the lead disappears and nobody knows it existed. Retries then send the same confirmation twice or create a duplicate record.

How to run it

  1. Map what happens after capture

    List every downstream step: notification email, CRM or case-system export, calendar booking, text message and model call. Treat each one as best effort.

  2. Persist before anything else

    Save the record, even a partial one, to one ledger with status new before the first downstream call. A hang-up halfway through still leaves a record.

  3. Key every side effect

    Each email, booking, export and payment carries an idempotency key, so a replayed webhook or retried job does nothing the second time.

  4. Give each integration a fallback

    If the export fails, email the record; if that fails too, park it in a retry queue that shows on the dashboard.

  5. Test with each dependency down

    Switch off each provider in a test and assert that the record is saved and surfaced and that nothing was sent twice.

  6. Refuse instead of pretending

    If the record cannot be stored, tell the person it did not go through. Never show success for a submission you did not keep.

Done when

  • a test with every downstream provider failing still finds the record saved and visible.
  • replaying the same webhook twice produces one record and one message.
  • every integration has a written fallback and a test that takes it.
  • the form shows an error if the save fails.

How teams get it wrong

  • The record is written when the call or form ends, so a hang-up or timeout loses it; save the partial record first and update it as you go.
  • Email is the only copy of the lead; keep the database as the source of truth and treat email as a notice.
  • Retries go in without idempotency keys, so an outage turns into duplicate texts; key every side effect before you add retries.
Template Markdown, ready to paste into a doc
# Capture flow audit: ____ (form, call or webhook)

## Capture
Record saved to: ____ (table or ledger)
Saved at step: ____ (before any downstream call)
Partial records kept: yes / no
If the save fails, the person sees: ____

## Downstream steps
| Step | Provider | Idempotency key | Fallback | Retry queue |
|---|---|---|---|---|
| Notification email | ____ | ____ | ____ | ____ |
| CRM or case export | ____ | ____ | email with the record | ____ |
| Booking | ____ | ____ | ____ | ____ |
| Confirmation text | ____ | ____ | ____ | ____ |
| Model call | ____ | ____ | ____ | ____ |

## Failure tests
- [ ] Each provider down alone: record saved and visible
- [ ] All providers down: record saved and visible
- [ ] Same webhook replayed twice: one record, one message
- [ ] Save fails: the person sees an error, and no success screen
- [ ] Failed items show on the dashboard with a retry action

## Weekly check
Inbound events in provider logs: ____
Records saved: ____
Gap: ____ (target 0)

The problem

An answer engine often names two or three companies and sends no click to anyone else, and a brand that gets skipped never hears about it. Without a measure per question, teams publish more content at random and cannot tell which question they lost, or to whom.

How to run it

  1. Write the buyer question set

    List the questions buyers ask before they choose, in their words, built by fixed rules from your own pages and category. Keep the set versioned so runs compare.

  2. Ask the real engines

    Send each question to each engine you track and store the full answer. Never let one model play the part of another engine.

  3. Record three things per answer

    Whether you were named by brand or domain, who was named instead, and which pages the engine cited.

  4. Score it honestly

    Divide citations by the question and engine pairs that returned an answer, times 100. If an engine is down, shrink the denominator rather than counting a miss.

  5. Ship one fix per lost question

    Add a short direct answer to the question on your own page, earn mentions on the third-party pages engines already cite, or refresh dates on pages that go stale, like pricing.

  6. Prove the lift

    Re-run the same questions after a fix ships and log the before and after for that question next to the edit behind it.

Done when

  • the question set is versioned and repeat runs on unchanged inputs give the same score.
  • every result row stores the engine's raw answer and the pages it cited.
  • an engine outage lowers the denominator and leaves the score honest.
  • each shipped fix is linked to the question it targets and a before and after reading.

How teams get it wrong

  • A model writes fresh test questions on every run, so the score moves for no reason; build questions by fixed rules and cache results for the day.
  • Teams simulate engine answers to save money and get a score that swings from run to run; query the real engines and cap the spend instead.
  • Teams chase the overall number; work the single biggest lost question each week.
Template Markdown, ready to paste into a doc
# Share of Answer tracker: ____ (brand)

Domain: ____
Engines tracked: ____
Run time each day: ____

## Buyer question set (version ____)
| # | Question in the buyer's words | Source page or rule | Stage (compare, choose, price) |
|---|---|---|---|
| 1 | ____ | ____ | ____ |
| 2 | ____ | ____ | ____ |
| 3 | ____ | ____ | ____ |

## Daily result per question and engine
| Question | Engine | Named (yes / no) | Named instead | Pages cited |
|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ |

Share of Answer = citations / pairs that returned an answer x 100 = ____
Engines unavailable today (removed from the denominator): ____

## This week's fix
Lost question: ____
Named instead: ____
Fix type:
- [ ] Direct answer passage on our page: ____ (URL)
- [ ] Mention on a page the engine already cites: ____
- [ ] Refresh a stale page: ____
Shipped on: ____
Before: ____  After: ____

The problem

Point-based scoring adds up activity, so a company with eight duplicate posts outranks one with two real buying triggers, and reps spend their week on noise. Scores also hide what they cannot see, so a confident number built on one signal reads like one built on ten.

How to run it

  1. Keep fit and intent apart

    Score fit (whether the account could buy) and intent (whether it is moving now) as two numbers. Decide with both in hand and never add them into one total first.

  2. Weight by cost to the buyer

    For each signal type, rate what it cost to produce: a senior person's time counts more than an anonymous visit, and a paid evaluation more than a blog read. Cheap traces get close to zero weight.

  3. Saturate repeats and decay with time

    The first instance of a signal type moves the score most, and each repeat adds less. Give each type a half-life, so an account that goes quiet drifts back to the base rate.

  4. Let bad news subtract

    Count disconfirming signals, such as a closed evaluation, a competitor adopted or a hiring freeze, as negative evidence.

  5. Qualify before a rep sees it

    Grade each account on five questions: right buyer, fresh trigger, able to pay, reachable, and a concrete need you meet. Hard disqualifiers drop it, and only A and B grades reach the list.

  6. Ship the blind spots with the score

    Show the evidence tier, the age of the newest signal, and a flag when the score rests on one person. Openers cite public signals only, never tracked behavior.

Done when

  • a test shows eight duplicate signals of one type scoring below two different strong triggers.
  • fit and intent are stored and shown as separate fields.
  • every score on the list links to the signals behind it, with source and date.
  • a test shows an account with no new signal for one half-life scoring lower than before.

How teams get it wrong

  • Teams count signals, so volume beats quality; saturate each type and weight by cost.
  • Low-fit, high-intent accounts get thrown away; route them to a lighter play instead of deleting them.
  • Outreach mentions what the scorer saw, like repeat pricing-page visits; reference only public signals the buyer chose to publish.
Template Markdown, ready to paste into a doc
# Signal weight sheet: ____ (ICP)

Fit in one sentence: ____

## Signal types
| Signal | Public source | Cost to the buyer (low / medium / high) | Half-life (days) | Confirms or disconfirms |
|---|---|---|---|---|
| Funding filing | ____ | ____ | ____ | confirms |
| Role posted that needs what we sell | ____ | ____ | ____ | confirms |
| Public request for a provider | ____ | high | ____ | confirms |
| Competitor adopted | ____ | ____ | ____ | disconfirms |
| ____ | ____ | ____ | ____ | ____ |

Repeats of one type: the first ____ count fully, then each adds less.

## Five-question qualification
- [ ] Right buyer: ____
- [ ] Fresh trigger (a change, not a state): ____
- [ ] Able to pay: ____
- [ ] Reachable (verified domain and a named buyer): ____
- [ ] Concrete need we meet: ____
Hard disqualifiers: ____
Grade: A / B / C / D (only A and B reach the weekly list)

## Shown with every score
- [ ] Evidence tier: verified, corroborated or inferred
- [ ] Date of the newest signal
- [ ] Flag if one person or one signal carries the score
- [ ] Opener cites a public signal only

The problem

Repeat work hides inside everyone's week, so nobody sees its total cost and it never gets fixed. Teams that automate too early build workflows for tasks that never come back, and teams that wait too long spend months on chores a tool could draft.

How to run it

  1. Log repeats for two weeks

    Each time someone does a task they have done before, add a line with the task, its trigger, the minutes spent and the tools touched.

  2. Count to three

    A task that shows up three times qualifies. Rank the qualifiers by minutes per month.

  3. Write the steps as done today

    Record the inputs, the outputs, and the one or two points where judgment is needed. Those points stay with a person.

  4. Use the lightest tool that fits

    Use a text expander such as Text Blaze for typed replies, Zapier or n8n for handoffs between apps, and Claude for drafting and summarizing.

  5. Keep a person on the output

    The workflow drafts or prepares, and a named person confirms before anything is sent or changed. Nothing is deleted automatically.

  6. Review monthly

    Compare minutes saved with time spent fixing the workflow, and retire any that cost more than they return.

Done when

  • the repeat log exists and every entry has a trigger and minutes spent.
  • each automated task has a named owner who confirms its output.
  • each workflow records how often it ran and how often a person had to fix it.
  • every task that hit three repeats has a recorded decision to automate it or keep it manual.

How teams get it wrong

  • Teams automate the judgment step along with the busywork; automate the preparation and keep the decision with a person.
  • A workflow breaks quietly and nobody owns it; give every workflow a named owner and an alert on failure.
  • Teams buy a new platform for one chore; start with the tools already in use.
Template Markdown, ready to paste into a doc
# Repeat work log: ____ (team)

Log period: ____ to ____

| Task | Trigger | Who | Minutes | Tools touched | Times seen |
|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |

## Qualified (seen three times or more)
| Task | Minutes per month | Judgment step that stays with a person | Tool | Owner |
|---|---|---|---|---|
| ____ | ____ | ____ | Text Blaze / Zapier / n8n / Claude | ____ |
| ____ | ____ | ____ | ____ | ____ |

## Workflow spec: ____
Trigger: ____
Inputs: ____
Steps the workflow does: ____
Output it drafts: ____
Person who confirms: ____
On failure, alert: ____

## Monthly review
- [ ] Runs this month: ____
- [ ] Times a person had to fix it: ____
- [ ] Minutes saved: ____
- [ ] Keep, change or retire: ____
Product management

Product management frameworks

The classics I rely on, each shown through a real company that used it well, with sources.

Case study · Superhuman

Two years of building, and only 22% of users would be very disappointed to lose it.

The problem
Superhuman started coding its email client in 2015, and in summer 2017 founder Rahul Vohra went looking for a way to measure product-market fit. He asked users how they would feel if they could no longer use Superhuman, and 22% said very disappointed. That was well short of the 40% bar Sean Ellis had found by benchmarking nearly 100 startups.
What they did
Vohra found the very disappointed group was mostly founders, managers, executives and business development people, and focusing on them alone lifted the score to 33%. The team described that customer as one persona who handles 100 to 200 emails a day, politely set aside users who would not be disappointed, and split the roadmap between doubling down on speed and removing blockers such as the missing mobile app, integrations and calendaring.
What happened
Three quarters later the score had reached 58%, and the team tracked it weekly and quarterly as its primary OKR.

Lesson for your teamMeasure fit inside the segment that loves you most, and spend half your roadmap on the people who almost do.

Sources: First Round Review: How Superhuman built an engine to find product/market fit Superhuman blog: the same essay by Rahul Vohra

How to run it

  1. Survey people who have used it for real

    Send four questions to users who have used the product recently and more than once: how they would feel without it, who would benefit most, the main benefit they get, and how to improve it.

  2. Score the very disappointed share

    Divide very disappointed answers by all answers to the first question. Sean Ellis set the bar at 40%, and below it you should treat fit as unproven.

  3. Find the segment that loves it

    Tag each respondent by role or use case and recompute the score for each group. Describe the highest-scoring group as one specific person, your high-expectation customer.

  4. Listen to the right near-fans

    Set aside people who would not be disappointed. Among the somewhat disappointed, keep those whose main benefit matches what your fans love, and read what holds them back.

  5. Split the roadmap in half

    Spend half the work deepening the main benefit fans named and half removing what blocks the near-fans. Write both lists down before you start.

  6. Track it as the main goal

    Rerun the survey on a fixed cadence with fresh respondents, and report the score for the target segment every time.

Done when

  • the very disappointed share is computed for all respondents and for each tagged segment, with the response count beside it.
  • the high-expectation customer is written as one paragraph the whole team can quote.
  • every roadmap item is labeled as deepening the main benefit or removing a blocker for near-fans.
  • the next survey run has a date and an owner.
Template Markdown, ready to paste into a doc
# Product-market fit survey: ____ (product)

Owner: ____
Sent to: users active in the last ____ days who have used it at least ____ times
Responses collected: ____
Next run: ____

## The four questions
1. How would you feel if you could no longer use ____? (Very disappointed / Somewhat disappointed / Not disappointed)
2. What type of people do you think would most benefit from ____?
3. What is the main benefit you receive from ____?
4. How can we improve ____ for you?

## Scores (bar: 40% very disappointed)
| Segment | Responses | Very disappointed | Score |
|---|---|---|---|
| All respondents | ____ | ____ | ____% |
| ____ | ____ | ____ | ____% |
| ____ | ____ | ____ | ____% |

## High-expectation customer
One paragraph, in plain words: ____

## Main benefit fans named (top three, in their words)
1. ____
2. ____
3. ____

## Roadmap split
| Deepen the main benefit (for fans) | Remove a blocker (for near-fans with the same main benefit) |
|---|---|
| ____ | ____ |
| ____ | ____ |

Set aside: answers from people who would not be disappointed.
Case study · Amazon (Kindle)

Ebooks in 2004 meant reading on a PC, with a thin catalog and high prices.

The problem
As Amazon veterans Colin Bryar and Bill Carr tell it, ebooks were close to a nothing business in 2004. You could read one only on a PC or a Mac, prices were high, and about 15,000 titles were available against hundreds of thousands in print, so there was no good reason to buy one.
What they did
Bezos had teams write the press release at the start of a project instead of the end, and the Kindle team worked backwards from the reading experience. That fixed details such as a device that arrives already linked to your account with past purchases downloaded, and an always-on connection so you could shop, download and read on one device.
What happened
Kindle launched in November 2007 after more than three years of work, with more than 90,000 titles, no computer needed, no monthly wireless fee, and books that downloaded in under 60 seconds.

Lesson for your teamWrite down the customer experience you would be proud to announce, then hold engineering to it.

Sources: First Round Review: lessons from Working Backwards by Colin Bryar and Bill Carr Amazon press release: Introducing Amazon Kindle (2007) About Amazon: an insider look at Working Backwards

How to run it

  1. Name the customer and the problem

    In one sentence each, say who the customer is and what is hard for them today, in words they would use.

  2. Write the press release

    Keep it under one page: headline, problem, solution, a customer quote and how to get started. If it does not describe something meaningfully faster, easier or cheaper than what exists, stop here.

  3. Write the FAQ

    Answer the questions a customer would ask, then the internal ones on cost, dependencies, risks and what has to be true. Keep it to five pages or less.

  4. Read it with skeptics

    Have everyone read the document in the meeting before anyone speaks, then challenge it. Rewrite it until the hard questions have plain answers.

  5. Turn the release into requirements

    Every promise in the press release becomes a requirement the build must meet. Anything the release does not promise waits.

  6. Shelve most of them

    Expect most PR/FAQs never to launch. Keep the ones that hold up under questioning and fund the best of them.

Done when

  • the press release fits on one page and names a specific customer and problem.
  • the FAQ answers price, the biggest risk and the dependencies in five pages or less.
  • a reader outside the team can say in one sentence why a customer would switch.
  • every requirement in the build plan traces to a line in the press release.
Template Markdown, ready to paste into a doc
# PR/FAQ: ____ (product)

Author: ____
Review meeting: ____

## Press release (one page or less)
Headline: ____ launches ____ so that ____ can ____
Subhead (who it is for and the one benefit): ____
The problem today, in the customer's words: ____
How the product removes it: ____
Quote from us (why we built it): ____
Quote from a customer (what changed for them): ____
How to get started: ____

## Customer FAQ
- What does it cost? ____
- Why is this better than what I use now? ____
- What does it not do? ____
- ____

## Internal FAQ (whole FAQ five pages or less)
- Who exactly is the customer, and how many are there? ____
- What has to be true for this to work? ____
- What are the biggest risks, and how do we retire each one? ____
- What does it depend on that we do not control? ____
- What will it cost to build and to run? ____

## Decision
Faster, easier or cheaper than today, and how: ____
Build / rewrite / shelve: ____
Requirements taken from the press release:
- [ ] ____
- [ ] ____
Case study · Southern New Hampshire University

Working adults asked about a degree and got a form letter and a mailed packet.

The problem
Southern New Hampshire University treated online students much like the high school graduates it had recruited the same way for decades. A prospective student who asked for information got a boilerplate reply within 24 hours and, a week or so later, the same mailed packet everyone received. The average online student was 30, juggling work and family, and needed a credential to improve their career prospects.
What they did
SNHU moved its small online team to separate offices in a former mill yard in Manchester and mapped every hurdle from first inquiry to first class for that job. It took over fetching transcripts from students' previous colleges, settled the financial aid conversation in a few days, and gave every new online student a personal adviser who stays in constant contact.
What happened
By the end of fiscal 2016, SNHU was closing in on $535 million in revenue, a 34% compound annual growth rate over five years, and its online college served more than 75,000 students.

Lesson for your teamSegment by the job and the situation, then fix the process the customer actually goes through.

Sources: Boston Globe: excerpt from Competing Against Luck GovTech: how jobs to be done applies to online education

How to run it

  1. Interview recent switchers

    Talk to people who recently started or stopped using a product like yours. Ask about the day they decided, and leave features out of it.

  2. Rebuild the moment of decision

    Ask when the need came up, what they were doing, what they tried first and what nearly stopped them. Write the timeline in their words.

  3. Write the job statement

    Use the form: when I am in this situation, I want to make this progress, so I can reach this outcome. Add how they want to feel and how they want others to see them.

  4. Map every hurdle

    List each step from first contact to the job being done, and circle every place where your process slows, confuses or ignores the customer.

  5. Rebuild around the job

    Change the offer, the process and the support so each circled hurdle is gone for people with this job, even if other customers keep the old path.

Done when

  • the team has at least eight switch interviews, each with a written timeline of the decision.
  • the job statement names a situation, the progress wanted and the outcome, and every interviewee fits it or is listed as an exception.
  • the hurdle map covers first contact through first value, with an owner for each circled hurdle.
Template Markdown, ready to paste into a doc
# Jobs to Be Done: ____ (product)

## Switch interviews
| # | Person | Switched from | Switched to | When they decided |
|---|---|---|---|---|
| 1 | ____ | ____ | ____ | ____ |
| 2 | ____ | ____ | ____ | ____ |
| 3 | ____ | ____ | ____ | ____ |

## Timeline of one decision (repeat per interview)
- First thought: when did the need come up, and what were they doing? ____
- First look: what did they notice or try first? ____
- Comparing: what else did they consider? ____
- The push: what made them act that day? ____
- What almost stopped them: ____

## Job statement
When I am ____ (situation),
I want to ____ (progress),
so I can ____ (outcome).
How I want to feel: ____
How I want others to see me: ____

## Hurdle map: first contact to first value
| Step | What the customer does | Wait or confusion | Hurdle (yes / no) | Owner | Fix |
|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |

## Exceptions
People who did not fit this job, and the job they had instead: ____
Case study · Intercom

A backlog full of good ideas, and no consistent way to compare any two.

The problem
Intercom's product team found the prioritization methods it looked at did not compare different ideas in a consistent way. The usual pulls kept showing up: pet ideas the team would use themselves, clever ideas over ones tied to goals, new ideas over ones they were confident in, and extra effort that was easy to discount.
What they did
The team built RICE, which multiplies reach per period, impact on a five-step scale and confidence as a percentage, then divides by effort in person-months to get total impact per time worked. Scores live in a shared spreadsheet that sorts the backlog automatically.
What happened
Intercom published the method and its spreadsheet for other teams to copy, with one standing rule: a dependency or a strategic need can justify building a lower score first.

Lesson for your teamPut every estimate in the open, so the debate is about the inputs instead of whose idea it is.

Sources: Intercom blog: RICE, simple prioritization for product managers

How to run it

  1. Fix the goal and the period

    Pick one goal and one time window, such as a quarter, so every estimate counts the same thing.

  2. Estimate reach in real numbers

    Count the people or events each idea touches in the period, from your own data, such as customers per quarter or signups per month.

  3. Score impact on a fixed scale

    Use 3 for massive, 2 for high, 1 for medium, 0.5 for low and 0.25 for minimal, judged against the goal you picked.

  4. Discount for confidence

    Use 100% when data backs the estimates, 80% when some of it does and 50% when it is mostly a hunch. Treat anything lower as a moonshot and scope it separately.

  5. Estimate effort in person-months

    Count the total work across product, design and engineering, in whole numbers, or 0.5 for anything well under a month.

  6. Rank, then decide

    Sort by the score and use the order to start the conversation. When you overrule it, write down why.

Done when

  • every idea on the list has all four inputs filled and a score from the same formula.
  • each reach figure names its data source.
  • every lower-scoring idea built ahead of a higher one has a written reason, such as a dependency.
Template Markdown, ready to paste into a doc
# RICE scoring: ____ (team)

Goal: ____
Period: ____ (for example, next quarter)
Reach unit: ____ (for example, customers per quarter)
Scored by: ____   Date: ____

## Scores
Score = Reach x Impact x Confidence / Effort
Impact: 3 massive, 2 high, 1 medium, 0.5 low, 0.25 minimal
Confidence: 100% high, 80% medium, 50% low

| Idea | Reach | Reach source | Impact | Confidence | Effort (person-months) | Score |
|---|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____% | ____ | ____ |
| ____ | ____ | ____ | ____ | ____% | ____ | ____ |
| ____ | ____ | ____ | ____ | ____% | ____ | ____ |
| ____ | ____ | ____ | ____ | ____% | ____ | ____ |

## Below 50% confidence (moonshots, scoped separately)
- ____
- ____

## Overrides
| Built ahead of a higher score | Reason (dependency, strategy, commitment) | Decided by |
|---|---|---|
| ____ | ____ | ____ |

## Next rescore
Date: ____
Trigger for an early rescore: ____
Case study · Basecamp

Customers kept asking for a calendar that could take six months to build.

The problem
Soon after Basecamp 3 launched, customers started asking the team to add a calendar. A proper calendar can easily take six months or more, and past versions of Basecamp had calendars that only about 10% of customers used.
What they did
Basecamp set an appetite of one six-week cycle and called a customer who had asked, to learn when she wanted a calendar rather than what it should look like. She needed to see free spaces on a shared schedule, so the team shaped a Dot Grid: a two-month, read-only grid with a dot for each event, and no dragging, no multi-day spans and no color coding.
What happened
The team shaped the Dot Grid as a six-week project and launched it at the end of that cycle, in place of a calendar that could have taken six months or more.

Lesson for your teamFix the time and let the scope move, and you will find the smaller version that does the job.

Sources: Shape Up, chapter 2: Principles of Shaping Shape Up, chapter 3: Set Boundaries

How to run it

  1. Set the appetite

    Before any design, decide how much time the problem is worth, such as two weeks or six, for a small team of a designer and one or two programmers.

  2. Find the baseline

    Ask a customer when they wanted the feature and what they were doing at that moment. Narrow the request to the specific need behind it.

  3. Shape it roughly

    Sketch the main elements and flow, name the rabbit holes, and list what is out of bounds. Leave the details to the team that builds it.

  4. Bet in the cool-down

    Choose which shaped projects to bet on in the break between cycles, then give each team the whole cycle without interruptions.

  5. Use the circuit breaker

    Work that is not done when the cycle ends gets no extension by default. Reshape it and bet again, or drop it.

Done when

  • every project in the cycle has a written appetite, a shaped pitch and a list of what is out of bounds.
  • the pitch names the specific customer need the request was narrowed to.
  • the cycle end date is fixed and unfinished work goes back to shaping instead of rolling over.
Template Markdown, ready to paste into a doc
# Pitch: ____

Appetite: ____ weeks, for ____ designer and ____ programmers
Cycle: ____ to ____ (no extension by default)

## Problem
The customer story behind the request (when did they want it, and what were they doing): ____
What exists today (baseline): ____
Narrowed need, in one sentence: ____

## Solution sketch
Main elements: ____
Main flow: ____
Sketch link: ____

## Rabbit holes
- ____ and how we avoid it: ____
- ____ and how we avoid it: ____

## Out of bounds
- ____
- ____
- ____

## Betting table (cool-down)
Bet on it: yes / no
Decided on: ____ by ____

## At cycle end
Shipped: yes / no
If not: reshape and bet again / drop
Notes: ____
Case study · Dropbox

Years of hard engineering ahead, for a product people struggled to grasp from a description.

The problem
Dropbox had to sync files across Windows, Mac, iPhone and Android, backed by an online service that needed high reliability. People also had a hard time understanding the product when it was explained, and Drew Houston faced the risk of waking up after years of development with a product nobody wanted.
What they did
Houston made a three-minute screencast, narrated by him, showing the product working as it was meant to, to test whether people would try it if the experience was better. He posted a version to Digg with about a dozen Easter eggs aimed at that audience.
What happened
The video drove hundreds of thousands of people to the site, and the beta waiting list grew overnight to 75,000 people, up from 5,000.

Lesson for your teamWhen building is expensive, give people something to react to before you build it.

Sources: TechCrunch: Dropbox case excerpted from The Lean Startup TechCrunch: how Dropbox got its first 10 million users

How to run it

  1. Write the leap-of-faith assumption

    State the belief that would sink the product if it were wrong, as a sentence you can test. For example: if we offer this experience, people will sign up to try it.

  2. Pick the smallest test

    Choose the cheapest thing that makes people act, such as a demo video, a landing page with a waitlist, a request form, or a manual service behind a simple front end.

  3. Set the bar before launch

    Write down the number that means yes, the number that means no, and the date you will read them.

  4. Show it where early users gather

    Put the test in front of the people most likely to want it first, in the places they already spend time and in the way they talk.

  5. Read the result and decide

    Compare the result with the bar. Keep going, change direction or stop, and log the decision with the evidence.

Done when

  • the assumption, the test, the bar and the read date are written down before the test goes live.
  • the test measures an action, such as a signup, a request or a payment, rather than an opinion.
  • the decision to keep going, change direction or stop is logged with the numbers behind it.
Template Markdown, ready to paste into a doc
# MVP test card: ____

Leap-of-faith assumption: If we ____, then ____ will ____.
Why this is the riskiest assumption: ____

## Test
Type: demo video / landing page with waitlist / request form / manual service / ____
Audience, and where we reach them: ____
Build budget: ____ days
Launch date: ____
Read date: ____

## Bar (set before launch)
Action we count: ____ (signup, request, payment)
Yes if: ____ or more by the read date
No if: ____ or fewer by the read date

## Result
| Measure | Bar | Actual |
|---|---|---|
| ____ | ____ | ____ |
| ____ | ____ | ____ |

What surprised us: ____

## Decision
Keep going / change direction / stop: ____
Evidence: ____
Next assumption to test: ____
Decided by: ____ on ____
Case study · Google (Gmail)

Seven-day active users counted anyone who opened Gmail once, and missed who relied on it.

The problem
Google teams commonly tracked seven-day active users, which counts everyone who visited at least once in the last week. The Gmail team wanted to understand how engaged its users were, and that count could not tell a daily user from one who visited once.
What they did
Working through the HEART categories and the goals, signals and metrics process, the team reasoned that engaged users check email as part of their daily routine. It chose a new engagement metric: the share of active users who visited on five or more days in the last week.
What happened
The researchers found that metric strongly predictive of longer-term retention, so it could serve as an early read on it. By the time of the paper, the authors had applied the framework to more than 20 Google products and projects.

Lesson for your teamChoose the behavior that shows the product is part of someone's routine, and measure its share instead of its count.

Sources: Google Research: Measuring the user experience on a large scale (CHI 2010) The paper as a PDF

How to run it

  1. Write the goals first

    For the product or feature, write what the user experience should achieve, using Happiness, Engagement, Adoption, Retention and Task success as prompts. Settle disagreements about goals at this step.

  2. Leave categories out on purpose

    Mark each category in or out, with a reason. Engagement may mean little for a tool people must use at work, for example.

  3. Find a signal for each goal

    Name the user action or attitude that would show the goal was met, and where it is logged or surveyed. Prefer signals that move only when the experience gets better or worse.

  4. Turn signals into rates

    Convert raw counts into ratios, percentages or averages per user, so a growing user base does not pass for a better experience.

  5. Check what the metric predicts

    Before a new metric goes on the main dashboard, test whether it tracks a longer-term outcome such as retention.

Done when

  • each metric on the dashboard links to a written goal and a named signal.
  • every HEART category is marked in or out with a reason.
  • no metric on the dashboard is a raw count that rises only because the user base grows.
  • at least one engagement or happiness metric has been checked against later retention.
Template Markdown, ready to paste into a doc
# HEART metrics: ____ (product or feature)

Owner: ____
Dashboard: ____
Review cadence: ____

## Goals, signals, metrics
| Category | In or out (why) | Goal | Signal | Metric (a rate or average) | Source |
|---|---|---|---|---|---|
| Happiness | ____ | ____ | ____ | ____ | survey |
| Engagement | ____ | ____ | ____ | ____ | logs |
| Adoption | ____ | ____ | ____ | ____ | logs |
| Retention | ____ | ____ | ____ | ____ | logs |
| Task success | ____ | ____ | ____ | ____ | logs or study |

## Rules
- Every metric is a ratio, a percentage or an average per user.
- Every signal is logged today, or has a ticket to log it: ____
- Traffic from bots and crawlers is filtered out: yes / no

## Validation
Metric checked against later retention: ____
Window: ____ weeks
Finding: ____

## Change log
| Date | Metric added, changed or removed | Why |
|---|---|---|
| ____ | ____ | ____ |
Case study · Duolingo

By mid-2018, daily users were growing in single digits after years of explosive growth.

The problem
By mid-2018, Duolingo's daily active users were growing at a single-digit rate year over year, which was troubling after the explosive growth the company had seen. Jorge Mazal, Head of Product since late 2017, needed to find which lever would restart growth.
What they did
The team built a growth model that sorts users into states such as new, current, reactivated and dormant, linked by retention rates, and ran a sensitivity analysis on years of daily history. Current user retention rate, the chance that someone who came in each of the past two weeks comes back this week, had 5 times the impact on DAU of the next-best input. Duolingo created a Retention Team with it as its North Star metric, and that team shipped leaderboards, streak improvements and better push notifications.
What happened
Mazal reports that current user retention rose 21% and DAU grew 4.5 times over four years, and that the leaderboards launch raised overall learning time by 17%.

Lesson for your teamModel which input moves your top number most, then give that input a team of its own.

Sources: Lenny's Newsletter: How Duolingo reignited user growth, by Jorge Mazal Duolingo blog: how the growth model sharpened product teams Amplitude: the North Star Metric and its three to five inputs

How to run it

  1. Name the North Star

    Pick one number that rises when customers get the value they came for, such as daily learners or orders delivered on time. It should lead revenue rather than be revenue.

  2. Break it into three to five inputs

    Write the North Star as a result of inputs a team can move, such as new users, current user retention and reactivation of lapsed users.

  3. Model the system

    Use your history to estimate how users move between states, then change each input by the same amount in the model and see which one moves the North Star most.

  4. Give the top input its own team

    Staff a team with that input as its goal, and have it run experiments that must move the input before they ship.

  5. Check that the link holds

    Confirm with A/B tests that moving the input moves the North Star. Rerun the model each quarter, since the strongest input can change.

Done when

  • the North Star and each input have an exact definition and a single owner.
  • a model ranks the inputs by their effect on the North Star, using your own history.
  • at least one experiment shows a change in an input followed by a change in the North Star.
Template Markdown, ready to paste into a doc
# North Star tree: ____ (product)

North Star: ____
Exact definition: ____
Why it reflects the value customers get: ____
Owner: ____

## Inputs (three to five)
| Input | Exact definition | Current value | North Star change from a 2% lift (model) | Team |
|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ |

## Model
User states: new, current, reactivated, dormant, ____
History used: ____ months of daily data
Top input this quarter: ____
Its impact compared with the next-best input: ____ times

## Experiments on the top input
| Experiment | Input change | North Star change | Ship (yes / no) |
|---|---|---|---|
| ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ |

## Next model run
Date: ____
Run by: ____
Operating

Organizational and operating frameworks

How decisions get made, how authority is handed down and how the weekly rhythm runs, each shown through a real organization and mapped as a diagram.

Case study · Wyeth Pharmaceuticals

Competitors were chasing the same drug class while every big call waited on one senior group.

The problem
By late 2000, Wyeth's pharmaceutical division ran three businesses in biotech, vaccines and traditional drugs, yet most important decisions went up to one group of senior executives. The biotech team saw a short window to lead with Enbrel, a rheumatoid arthritis drug, and needed a new plant in Dublin that would be the largest capital investment Wyeth had ever made. Input reached the top slowly through overlapping committees.
What they did
Wyeth moved from a first look at the Grange Castle plant to implementation in six months. Along the way it handed many decisions to business unit managers and left senior executives a veto on the most sensitive plant questions. After the investment call, the D for many Enbrel decisions sat with Cavan Redmond, who ran the biotech division. He gathered input from manufacturing, marketing, forecasting, finance and R&D.
What happened
The Bain authors report Enbrel sales of $1.7 billion through the first half of 2005. When the FDA gave the antibiotic Tygacil priority review, the key calls were made one or two levels down in the biotech group, and the drug moved into volume production three days after its June 2005 approval.

Lesson for your teamPut the D where the operating knowledge is, and keep the veto narrow.

Sources: Bain & Company: Who has the D? How clear decision roles enhance organizational performance Harvard Business Review reprint (PDF): Who Has the D?, Rogers and Blenko, January 2006 Bain & Company: RAPID tool to clarify decision accountability

How it works

Who holds each role, decision by decisionRead each row as one decision: one role holds the D, a few hold a veto, and the rest give input or carry it out.
DecisionRRecommendAAgree (veto)PPerformIInputDDecide
Build a second manufacturing plantBusiness unit headCFOOperationsManufacturing, marketing, financeCEO
Set the launch price of a new productProduct marketing leadLegalSalesFinance, salesBusiness unit head
Change the standard features of next year's modelProduct developmentRegulatory affairsManufacturingMarketing, dealersProduct line head
Hire a regional sales directorHiring managerHRHiring managerPeer directorsVP of sales

How to run it

  1. List the decisions that matter

    Write down the recurring decisions that create the most value or cause the most friction, such as pricing, capacity and senior hires. Leave routine calls off the list.

  2. Give each one a single D

    One person decides and commits the organization to act. If two people think they hold the D, the decision turns into a tug of war, so settle it on paper first.

  3. Keep the agree list short

    Agree is a real veto, usually held by legal, compliance or the head of a unit whose operations will change. A long list of people who can say no usually means decisions sit too high in the company.

  4. Name one recommender and the inputs

    The recommender gathers the facts, consults the people who give input and brings a proposal with options. Input is not binding, but the people giving it often carry out the decision, so their views count.

  5. Name who performs

    Say who executes once the call is made, and bring them in early as input so the plan is one they can run.

  6. Move the D to where the knowledge is

    Hand decisions down to the unit that holds the operating knowledge. Keep at the top only the calls that need the senior team, such as large capital commitments.

Done when

  • every decision on the list has exactly one D, written down.
  • each person in the agree role can name the narrow grounds for a veto.
  • the people in each role can state their role on a live decision without checking the chart.
  • a vetoed proposal goes back to the recommender first and reaches the D only after a set number of days.
Template Markdown, ready to paste into a doc
# Decision rights register: ____ (team or unit)

Owner of this register: ____
Last reviewed: ____

## Decisions
| Decision | Recommend | Agree (veto) | Perform | Input | Decide |
|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |

## Rules
- One D per decision. If two people claim it, ____ settles it by ____.
- Agree is a veto on narrow grounds only. Grounds for each person in the agree role: ____
- A veto goes back to the recommender first, then to the D after ____ days.

## Live decision
Decision: ____
Recommender: ____
Inputs consulted (name and date): ____
Options considered: ____
Recommendation: ____
Agree sign-offs: ____
Decided by: ____ on ____
Performed by: ____, first milestone on ____

## Quarterly health check
- Decisions escalated twice or more: ____
- Decisions reopened after the D decided: ____
- Roles to change: ____
Case study · USS Santa Fe (U.S. Navy)

A crew trained to wait for orders was running the worst-rated submarine in the fleet.

The problem
In 1999 David Marquet was diverted to command the USS Santa Fe, then ranked last in retention and operational standing. His publisher describes poor morale, poor performance and the worst retention rate in the fleet. Marquet concluded that the usual approach, where the captain takes control and gives orders, would not fix it.
What they did
Marquet gave control instead of taking it. Officers and crew stopped asking for permission and told him what they intended to do. In his words: "Tell me what you intend to do, and if I don't stop you, you do it." He paired that with building the crew's technical competence and making the ship's goals clear, so the people closest to the work could make the call.
What happened
The Santa Fe went from worst to first, with the highest retention and operational standings in the Navy. After Marquet left, it kept winning awards and promoted a large share of its officers and enlisted crew, including ten later submarine captains.

Lesson for your teamGive control step by step, and match each step with skill and a clear goal.

Sources: Penguin Random House: Turn the Ship Around, by L. David Marquet InfoQ: Marquet's Agile 2017 keynote on intent-based leadership Wikipedia: David Marquet

How it works

Seven rungs of ownershipRead left to right: each rung moves more of the decision to the person doing the work.
  1. 1 Tell me what to do Leader decides everything An analyst asks which vendor to call about a late shipment.
  2. 2 I see... Team reports facts, leader decides "Two vendors missed the delivery window last week."
  3. 3 I think... Team offers a view, leader decides "I think we should move volume off the slower vendor."
  4. 4 I would like to... Team proposes, leader approves "I would like to move the account to vendor B."
  5. 5 I intend to... Team decides unless the leader stops it "I intend to move it Monday unless you object."
  6. 6 I have done... Team acts, then reports "I moved it and updated the contract tracker."
  7. 7 I've been doing... Team owns it; leader hears in review "Vendor changes have been mine since last quarter."

How to run it

  1. Find the rung for each decision type

    For each recurring call, note how people talk about it today. "Tell me what to do" is the bottom rung and "I've been doing" is the top.

  2. Stop answering on the low rungs

    When someone asks what to do, ask what they see and what they think. Each answer moves the speaker one rung up.

  3. Teach "I intend to"

    Ask people to state the action, the reason and the check they ran. The leader then says yes, asks a question or, rarely, stops it.

  4. Build skill before you hand over control

    Control without competence is a gamble. Pair each step up with training, a rehearsal or a checklist the person owns.

  5. Share the goal behind the order

    People can only intend the right thing when they know what the team is trying to achieve. State the goal and the limits, then stop there.

  6. Move up one rung at a time

    When "I intend to" calls hold up for a few weeks, ask for "I have done" reports instead. Drop back a rung after a miss and say why.

Done when

  • each recurring decision type has a written rung and a target rung.
  • most requests you get start with "I intend to" rather than a question.
  • you can name the training or checklist behind every rung you raised.
Template Markdown, ready to paste into a doc
# Ladder of leadership: ____ (team)

Leader: ____
Date: ____

## Rungs
1. Tell me what to do
2. I see ____
3. I think ____
4. I would like to ____
5. I intend to ____
6. I have done ____
7. I've been doing ____

## Decision types
| Decision type | Rung today | Target rung | Skill or checklist needed | Review date |
|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ |

## The "I intend to" message
I intend to ____
Because ____
I checked ____
Unless you stop me by ____, I will go ahead.

## What the team needs to know to decide
- Goal: ____
- Limits on money, customers and safety: ____

## Moves this month
| Decision type | From rung | To rung | Reason | Agreed by |
|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ |
Case study · Amazon (Fulfillment by Amazon)

A seller fulfillment idea kept stalling because nobody's full-time job was to build it.

The problem
Letting outside sellers use Amazon's fulfillment network was an idea that bounced around the company and sat on a long list of initiatives. As Amazon veterans Colin Bryar and Bill Carr tell it, the demands of the core business soaked up the bandwidth of the leaders who could have built it.
What they did
Jeff Wilke, who ran operations, asked one of his strongest VPs to drop a very large role in the operations organization and focus on this alone. With a full-time leader, the team identified the resources it needed and hired and developed the people to build it. Amazon's David Limp put the principle plainly: the best way to fail at inventing something is to make it somebody's part-time job.
What happened
Bryar and Carr say the program was built and launched in only a matter of months. Amazon announced Fulfillment by Amazon on September 19, 2006, opening its order fulfillment and customer service to sellers for a handling fee as low as $0.50 per item plus $0.40 per pound.

Lesson for your teamIf an initiative matters, make it one leader's only job.

Sources: First Round Review: lessons from Working Backwards by Colin Bryar and Bill Carr Amazon press release: Fulfillment by Amazon launch (2006)

How it works

One leader, one initiative, a team that can move aloneRead from the top: the leader owns the initiative, and each branch is something the leader controls without borrowing it.
  • Single-threaded leaderWorks on this initiative and nothing else
    • Dedicated teamHired and run by the leader
      • EngineeringFull time on this initiative
      • ProductOwns the requirements and the launch
      • OperationsRuns it once it ships
    • BudgetSpent without another group's approval
    • InterfacesEvery dependency written as an agreed contract
      • Shared servicesUsed through a service or data feed
      • Partner teamsFixed handoffs with named owners
    • MetricsInputs and outputs reviewed every week

How to run it

  1. Pick the initiatives that need a full-time owner

    List the bets that matter but keep losing to the core business for attention. Each one is a candidate.

  2. Name one leader per initiative

    The leader works on that initiative and nothing else. A strong person freed from a large role beats a junior hire whose title says owner.

  3. Give the leader the team and the budget

    Engineers, product and operations people report into the initiative, so progress does not depend on borrowing time from other groups.

  4. Make the team separable

    Write down where the team depends on others, and turn each dependency into an agreed interface such as a service, a data feed or a fixed handoff.

  5. Judge the leader on one set of metrics

    Agree the inputs and outputs the initiative is judged on, and review them on the same rhythm as the rest of the business.

Done when

  • each listed initiative has a named leader with no second role.
  • the leader can hire, move people and spend inside an agreed budget without asking another group.
  • every dependency on another team is written down as an interface with an owner.
Template Markdown, ready to paste into a doc
# Single-threaded initiative charter: ____

Initiative: ____
Leader (no other role): ____
Sponsor the leader reports to: ____
Start date: ____

## Why it needs a full-time leader
How long it has been on the list: ____
What it has lost to so far: ____

## What the leader controls
- People (names or roles): ____
- Budget for the first two quarters: ____
- Decisions the leader makes alone: ____
- Decisions that go to the sponsor: ____

## Dependencies
| Depends on | What we need | Interface (service, feed, handoff) | Owner on their side | Date agreed |
|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ |

## Metrics
Output we are judged on: ____
Inputs we control: ____, ____, ____
Reviewed at: ____ (weekly meeting)

## Exit
We stop or fold it back in if: ____
Decided by: ____
Case study · Amazon

As Amazon grew, reversible calls drifted into the slow process built for irreversible ones.

The problem
In his 2015 letter to shareholders, Jeff Bezos wrote that larger organizations tend to run most decisions through the heavyweight process meant for consequential, irreversible ones. The result, he wrote, is slowness, unthoughtful risk aversion, too little experimenting and less invention.
What they did
Amazon split decisions into two kinds. Type 1 decisions are one-way doors, made methodically and slowly with deliberation and consultation. Type 2 decisions are two-way doors, made quickly by high-judgment individuals or small groups. In his 2016 letter Bezos added that most decisions should be made with around 70% of the information you wish you had, and that a leader can disagree and commit. He did that himself when he approved an Amazon Studios original despite his doubts.
What happened
Bezos tied the approach to Amazon's habit of experimenting. Amazon took two big swings that missed, Auctions and zShops, before it launched Marketplace, and by his 2015 letter close to 50% of units sold on Amazon came from third-party sellers.

Lesson for your teamMost decisions can be undone. Make those fast, and spend your care on the few that cannot.

Sources: Amazon 2015 letter to shareholders (SEC filing) Amazon 2016 letter to shareholders (SEC filing)

How it works

Two kinds of decision, two speedsSort each decision to one side first, then use that side's process and nothing heavier.
Two-way door (Type 2)
  • Reversible, and a wrong call costs little for long.
  • Made by one person with good judgment or a small group.
  • Made with about 70% of the information you wish you had.
  • Watched closely and corrected fast if it proves wrong.
One-way door (Type 1)
  • Irreversible or nearly so.
  • Made slowly, with deliberation and consultation.
  • Owned by a senior person with a written case.
  • Tested in a reversible slice first where possible.

How to run it

  1. Ask what it costs to walk back

    For each pending decision, write down what undoing it would take in money, time and customer trust. If it can be reversed quickly at a modest cost, it is a two-way door.

  2. Push two-way doors down

    Let one person with good judgment, or a small group, make the call. Bezos argues these can and should be decided quickly.

  3. Decide with about 70% of the information

    Bezos suggests most decisions need around 70% of the information you wish you had. Make the call, then watch closely for signs it was wrong and correct fast.

  4. Slow down for one-way doors

    Irreversible calls get the careful process: a written case, consultation and a senior owner. Look first for a way to turn part of the bet into a reversible test.

  5. Disagree and commit

    When the group cannot agree, the owner asks the others to commit anyway. Record the dissent so the review can check who was right.

  6. Review the doors you misjudged

    Each quarter, look at decisions that took longer than they should have and calls that proved hard to undo, and adjust how you sort the next ones.

Done when

  • every open decision is tagged one-way or two-way, with the reason written down.
  • two-way decisions are made by a named person without a committee.
  • each one-way decision has a written case and a named senior owner.
  • the decision log shows each reversal and how long it took.
Template Markdown, ready to paste into a doc
# Decision sort: ____ (week of)

Sorted by: ____

## Open decisions
| Decision | Cost to undo (money, time, trust) | Door | Decider | Decide by |
|---|---|---|---|---|
| ____ | ____ | two-way / one-way | ____ | ____ |
| ____ | ____ | two-way / one-way | ____ | ____ |
| ____ | ____ | two-way / one-way | ____ | ____ |

## Two-way door
Decision: ____
Information we have (about 70% is enough): ____
Signal that would tell us it was wrong: ____
Check date: ____

## One-way door
Decision: ____
Written case linked here: ____
People consulted: ____
Part we can test first as a two-way door: ____
Senior owner: ____

## Disagree and commit
Who disagreed and why: ____
Committed on: ____
Review date to see who was right: ____

## Quarterly look back
- Two-way calls that took more than ____ days: ____
- Calls we tagged two-way that proved hard to undo: ____
Case study · Amazon

Amazon's teams raised the selection metric they were given, and sales did not move.

The problem
Amazon picked selection as a controllable input and first measured it as the number of product detail pages. Teams added pages, but sales, the output, did not rise. Some teams had bought products that were not in high demand, and that stock took up warehouse space.
What they did
The metric was reworked through the Weekly Business Review. It became detail page views, since a page nobody sees earns no credit. Then it became the share of detail page views where the item was in stock, and finally the share where it was in stock and ready for two-day shipping, called Fast Track In Stock. The review runs every Wednesday for 60 minutes over 400 to 500 metrics, and owners speak only to unusual variation.
What happened
Fast Track In Stock measured selection as items customers could buy and receive in two days. Bezos's 2009 letter shows how far the input focus went: 452 detailed goals for 2010 with owners, deliverables and dates, where revenue appeared eight times and net income, gross profit and operating profit not once.

Lesson for your teamPick inputs a team can move, and keep redefining them until moving the input moves the output.

Sources: Commoncog: The Amazon Weekly Business Review Commoncog case: Amazon's selection metric, from Working Backwards Amazon 2009 letter to shareholders (SEC filing)

How it works

The weekly input loopRead clockwise from planning: the yearly goals set the inputs, and the weekly turns test whether those inputs still move the outputs.
  1. YearlySet input goalsInput goals with owners, deliverables and dates
  2. WeeklyRefresh the deckSix-week and twelve-month charts against last year and target
  3. WeeklyReview exceptionsOwners explain unusual moves or say they are investigating
  4. WeeklyAct on causesFixes assigned to the owners of the inputs
  5. When an input stops predictingAudit the metricA redefined input, such as Fast Track In Stock

How to run it

  1. Split outputs from inputs

    Outputs are results such as revenue or active users. Inputs are things a team can move this week, such as price, selection, in-stock rate or delivery time.

  2. Give every input an owner and a date

    Set the yearly goals mostly on inputs, each with an owner, a deliverable and a target completion date.

  3. Show the same charts every week

    Plot each metric over the last six weeks and the last twelve months on one axis, with last year and the target on the same chart.

  4. Talk only about exceptions

    Owners say "nothing to see here" for routine variation and the meeting moves on. For an unusual move they explain the cause or say they are still investigating. Nobody guesses.

  5. Check that the input still drives the output

    If an input rises and the output does not follow, redefine the input. Expect several versions before one settles.

Done when

  • every metric in the deck has one named owner.
  • most yearly goals are inputs, each with an owner and a date.
  • the review runs on the same day and for the same length every week, including weeks the CEO is away.
  • each input has been tested against its output at least once, then kept or redefined.
Template Markdown, ready to paste into a doc
# Weekly business review: ____ (company or unit)

Day and time (fixed): ____
Length: ____ minutes
Deck owner: ____

## Outputs (we watch these)
| Output | Definition | Target this year |
|---|---|---|
| ____ | ____ | ____ |
| ____ | ____ | ____ |

## Inputs (we move these)
| Input | Exact definition | Owner | Target | Output it should move |
|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ |

## Each chart shows
- [ ] Last six weeks and last twelve months on one axis
- [ ] Same period last year
- [ ] Target

## This week's exceptions
| Metric | What moved | Cause, or "still investigating" | Action and owner |
|---|---|---|---|
| ____ | ____ | ____ | ____ |

## Metric audit
Input that rose while its output did not: ____
New definition: ____
Version number: ____
Case study · Intel

Motorola's faster chip was beating Intel's 8086 in the field, and sales knew first.

The problem
In late November 1979, district sales manager Don Buckout sent an eight-page telex warning that the 8086 was getting beaten by Motorola's faster 68000. Intel needed every function pointed at the same fight.
What they did
Within two weeks an executive task force set the corporate objective: establish the 8086 as the highest performance 16-bit microprocessor family. Its key results included publishing five benchmarks, repackaging the 8086 family, getting the 8MHz part into production and sampling the arithmetic coprocessor by June 15. Engineering turned one of them into its own objective, to deliver 500 8MHz parts to CGW by May 30, with dated key results for masks and test tapes. Sales and marketing moved to selling long-term system value to CEOs.
What happened
By January 1980 Andy Grove had teams deployed around the world. The goal was 2,000 design wins, and by the end of the year Intel had about 2,500. The 8088, a variant of the chip, went on to power the first IBM PC. John Doerr later brought the method to Google in 1999.

Lesson for your teamOne clear objective, cascaded into dated key results, lets every team see its part of the fight.

Sources: What Matters: John Doerr on Operation Crush What Matters: Intel's company-wide OKR for the 8086 What Matters: Intel engineering OKR, 500 8MHz parts by May 30 What Matters: Google's OKR playbook

How it works

Operation Crush, cascadedRead top down: the company objective sits at the root, its key results below, and one key result becomes engineering's own objective.
  • Establish the 8086 as the highest performance 16-bit familyIntel corporate objective, 1980
    • Publish five benchmarks showing superior 8086 performanceCorporate key result
    • Repackage the entire 8086 familyCorporate key result
    • Get the 8MHz part into productionCorporate key result
      • Deliver 500 8MHz parts to CGW by May 30Engineering objective; masks April 9, test tapes May 15
    • Sample the arithmetic coprocessor by June 15Corporate key result

How to run it

  1. Write the objective as a direction

    One sentence on what you want to achieve. It should be significant and action-oriented, with no number in it.

  2. Add three to five key results

    Each key result is specific, time-bound and verifiable. At the end of the quarter you either hit it or you did not.

  3. Cascade one level down

    A team takes a company key result and writes its own objective from it, with its own dated key results.

  4. Mark committed and aspirational

    Committed OKRs are expected to reach 1.0, and the company will move schedules and resources to deliver them. Aspirational OKRs are expected to average about 0.7.

  5. Grade at quarter end, apart from pay

    Score each key result from 0.0 to 1.0 and explain any committed miss. Keep grades out of compensation, as Grove did at Intel.

Done when

  • each objective has three to five key results a stranger could verify.
  • every team objective traces to a company key result.
  • each OKR is marked committed or aspirational before the quarter starts.
  • last quarter's grades are written down and every committed miss has an explanation.
Template Markdown, ready to paste into a doc
# OKRs: ____ (team), Q__ ____

Owner: ____
Parent key result this cascades from: ____

## Objective 1 (committed / aspirational)
____

| Key result | Baseline | Target | Due | Owner | Grade (0.0 to 1.0) |
|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |

## Objective 2 (committed / aspirational)
____

| Key result | Baseline | Target | Due | Owner | Grade (0.0 to 1.0) |
|---|---|---|---|---|---|
| ____ | ____ | ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ | ____ | ____ |

## Rules
- Committed OKRs aim for 1.0. Any lower score needs an explanation.
- Aspirational OKRs aim to average about 0.7.
- Grades are not used in pay decisions.

## Quarter-end review
Committed misses and why: ____
What we carry into next quarter: ____
Case study · Netflix

Netflix split off its DVD business, and executives who saw the risk stayed quiet.

The problem
In 2011 Netflix raised prices and announced it would move its DVD-by-mail service into a separate business called Qwikster. Members would have needed two websites and two subscriptions. Reed Hastings later called it the biggest mistake in Netflix history.
What they did
Hastings found that people who expected it to fail had said nothing. One vice president told him: "I should have laid down on the tracks screaming that I thought it would fail. But I didn't." Netflix made dissent part of the process. A proposal now goes out in a shared document where colleagues rate it from -10 to +10 with comments, and the culture memo says every significant decision has an informed captain who must seek out different opinions.
What happened
Netflix reversed the split on October 10, 2011, after losing about one million subscribers and watching its stock fall more than 50%. Hastings has since called Qwikster his favorite failure, because it taught him to seek more input before major decisions.

Lesson for your teamGive people the call, and make asking for disagreement part of making it.

Sources: Netflix culture memo Ozan Varol: what you can learn from Netflix's biggest mistake CNBC (via NBC New York): Reed Hastings on farming for dissent Popular Science: Netflix abandons Qwikster (2011)

How it works

What the leader does under each modelCompare the rows across: the left is where most teams start, and the right is what the leader does instead.
Control
  • The leader approves each significant decision.
  • A new rule follows every mistake.
  • Disagreement reaches the top late, if at all.
  • Speed depends on one person's calendar.
Context
  • The leader shares strategy, stakes and assumptions.
  • An informed captain owns each significant call.
  • The captain asks for ratings from -10 to +10 before deciding.
  • The leader steps in only for ethics, material harm, a crisis or missing context.

How to run it

  1. Write down the context

    Share the strategy, the goal behind the work, what is at stake and the assumptions you are making, so people can judge for themselves.

  2. Name an informed captain

    For every significant decision, name the one person who will make the judgment call. A committee does not vote on it, and the boss does not take it back.

  3. Farm for dissent

    The captain writes a short memo and asks colleagues to rate the idea from -10 to +10 with comments. It is a read on how strongly people disagree, and it is not a vote.

  4. Let the captain decide

    The captain weighs the comments and makes the call. Leaders step in only for something unethical, a decision that could materially harm the company, a crisis, or a new person who lacks the context.

  5. Look back in the open

    After a big bet, the captain writes up what happened. If it failed, ask what context was missing before asking who got it wrong.

Done when

  • each significant decision has a named informed captain before work starts.
  • every big proposal has a dissent memo with ratings and comments attached.
  • leaders can list each time they overrode a captain and the reason.
Template Markdown, ready to paste into a doc
# Informed captain memo: ____

Captain: ____
Decision needed by: ____

## Context from leadership
Strategy this serves: ____
What is at stake: ____
Assumptions we are making: ____
Limits (money, legal, brand): ____

## The proposal
What I intend to do: ____
Why now: ____
What it costs to undo: ____

## Dissent (rate -10 to +10, with a comment)
| Name | Role | Rating | Comment |
|---|---|---|---|
| ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ |
| ____ | ____ | ____ | ____ |

Strongest objection, in its author's words: ____
How I answered it: ____

## Decision
Decided by the captain on: ____
Leadership override (only for ethics, material harm, crisis or missing context): yes / no, reason ____

## Look back on ____
What happened: ____
Context that was missing: ____
Measurement

How I measure KPIs

I measure work against a target, cap the credit at that target and put a source on every number. The formulas below come from systems I built, and each one links to its write-up.

  1. Cap the credit at the target

    Attainment is min(100, value ÷ target), so beating one KPI by a wide margin cannot hide a miss on another. Team performance is the KPI-weighted total of those capped scores.

  2. Write the weights down

    Every composite score shows its weights. Unit health is growth, retention, runway and execution at 30/25/25/20, so anyone can see why a unit ranks where it does.

  3. No number without a source

    Every figure on a dashboard carries the record it came from. When there is no source, the screen says so instead of showing an estimate.

  4. Derive state from events

    Priority and SLA come from server timestamps, and the AI can propose a change but never sets a stage, status, priority or score.

  5. Count proof, not presence

    Hours and metric values count once a manager approves the work log, which carries its proof. A flagged log needs a reason.

  6. Automate on the third repeat

    Any task that recurs three times becomes a workflow, with a person checking what it produces.

My metric tree: unit healthRead top down: the root is the weighted score each unit gets, and below it are the inputs I watch between weekly submissions. This is how I organize them, not a formula.
  • Unit health scoreGrowth 30, retention 25, runway 25, execution 20
    • GrowthWeekly KPI submission; missing data ranks high
    • RetentionWeekly KPI submission per unit
    • RunwayWeekly KPI submission per unit
    • ExecutionHuman-owned; AI only proposes changes
      • Key-result driftPoints behind a straight line; 10 points grace
      • Blocker SLASLA 1, 3, 7 or 14 days by priority
      • Missing KPI dataRanked high on Today
      • Missed standupsSurfaced daily
My review rhythmRead clockwise from the work log: each stage feeds the next, and the loop closes when repeat work becomes a workflow.
  1. Daily to monthly, per KPILog the workHours, metric values and proof
  2. As logs arriveApprove or flagApproved logs count; each flag carries a reason
  3. DailyTriage TodayBreached blockers first, then P0 and P1, drift and missing data
  4. WeeklySubmit and review KPIsUnit health, key-result drift and the Monday pre-read
  5. On the third repeatAutomate the repeatsA workflow replaces the hand-done task

Worked examples from my systems

SystemMetricHow it is computedWhy it is built this way
VA KPI OS Team performanceAttainment per KPI = min(100, value ÷ target); performance = KPI-weighted attainmentThe cap stops one big number from covering a missed KPI, and only approved logs feed it.
Ops health Unit health0.30 × growth + 0.25 × retention + 0.25 × runway + 0.20 × executionOne score per unit gives founders a 60-second read across the company, with the slipping component one click away.
Ops health Blocker priority and SLAImpact × urgency sets P0 to P3, with SLAs of 1, 3, 7 and 14 days from server timestampsPriority comes from the math and the clock, so a breached blocker reaches the top of Today without anyone arguing for it.

Where the practice comes from

AmazonGoals are set on controllable inputs and reviewed every week. Bezos's 2009 letter lists 452 goals for 2010 with owners, deliverables and dates, and revenue appears in them eight times. In the Weekly Business Review, owners speak only to unusual variation.

Like my Today view, the review spends its time on exceptions and on inputs a team can move this week.

Sources: Amazon 2009 letter to shareholders (SEC filing) Commoncog: The Amazon Weekly Business Review

GoogleKey results are graded from 0.0 to 1.0 at the end of each quarter. Committed OKRs are expected to reach 1.0, and aspirational ones to average about 0.7, with high variance.

A fixed scale against a target set before the quarter is the same idea as capped attainment: credit is measured against the bar that was agreed.

Sources: What Matters: Google's OKR playbook What Matters: OKR meaning, definition and grading

IntelDuring Operation Crush, key results were dated and checkable, such as masks to the fab on April 9 and test tapes by May 15, and the company counted design wins against a goal of 2,000.

A dated key result has a source and a deadline, so nobody argues about whether it happened.

Sources: What Matters: Intel engineering OKR, 500 8MHz parts by May 30 What Matters: John Doerr on Operation Crush

Send me what you are working on.

A product idea, a team that needs another builder, a competition that needs a judge. Tell me what it is and I will say straight away whether I am the right person.

Book a 30-minute call LinkedIn
↑ ↓ to moveEnter to openEsc to close