Models Operations model

Decision rights

Battle tested

Who decides what, how decisions get recorded and reused, and what a founder keeps at each stage.

46% of facts from primary sources, 6 of 1313 verified facts, last checked 29 Oct 2026Primary source: primary sourceSecondary: secondary sourceEstimate: estimate

The Alphabet decision framework

How I run decisions, at any stage. Five steps, one per letter.

  1. Assign

    Every recurring decision gets one named owner, written down: who owns what, when and why. Founders keep only what only a founder can do.

  2. Bound

    Owners decide alone inside written money and risk limits. An engineer does not open a bank account. Above a limit, or when a call cannot be undone, it goes up.

  3. Challenge

    Anyone can challenge, with a reason, before the call. After it, everyone commits. It reopens only with new evidence.

  4. Document

    Every decision lands in a light log, captured with AI: what, who, why and the result expected. Search it before solving anything twice.

  5. Evolve

    Review what was decided, check whether it worked, and change the rule. That is how processes get built that hold up and last.

Who decides, and how?

Up to seven questions. The rust line is your path. Each faint branch is where the other answer leads. Turn the dial to your stage; slide or type your headcount.

  1. Is there a works council (Betriebsrat), and does the decision touch hiring or transfers, monitoring tools, working hours or pay systems?

    • Yes: Run the works council step first
  2. Is the group already deadlocked, after a recommendation and any agreement step?

    • Yes: Disagree and commit
  3. Can the decision be reversed at an acceptable cost within a few weeks (a two-way door)?

    • Yes, two-way door: More questions
  4. Is it on the founder-reserved list (fundraising terms, equity, exec hires and firings, legal entity or jurisdiction, pricing model, layoffs, mission)?

    • Yes: The founder decides, after a memo
  5. Does it recur, and do three or more functions need a say?

    • No: Use DACI with one Approver

Stage

Who decides what

Set by the dial on the instrument. It follows your headcount until you turn it.

Runs it
Founder as decider, with each early hire as single owner for their area. Ops generalist as backup.
You keep
Product direction, hiring, pricing and money.
Hand off
Tool choices, vendors and process in each area.

My stance

Write the owner map and the limits down, and start the log now. With AI it costs almost nothing.

How to run it

Numbered steps, each with an owner, a time cost and a document.

  1. List the decisions

    Write down the 15 to 25 decisions that recurred or stalled in the last quarter.

    Chief of Staff or ops lead, else the founder2 hoursDecision inventory

  2. Classify each one

    Mark each as a one-way or two-way door, and note how many teams it touches.

    Founder and ops lead1 hourDecision class sheet

  3. Assign a decider

    One decider per decision type, and a named backup.

    Founder1 hourDecision rights map

  4. Mark the reserved list

    Write down what the founder keeps. Everything else is delegated unless it is a one-way door.

    Founder30 minutesReserved decisions list

  5. Publish it

    Put it in the handbook, review it with the leadership team and add “who decides?” to the meeting template. Repeat every six months.

    Ops lead1 hourRights map and meeting template

Practice: practice

How decision rights go wrong

Failure modes

  • No single decider, or two

    One name per decision, and a mandatory decider column in the log. Bain: no decider stalls, several cause a tug-of-war.

  • The founder re-opens delegated decisions

    Count reopened decisions each quarter. Say “I disagree and commit”, or override in writing.

  • Late vetoes

    Gather agreement while building the recommendation, before the decision. Bain.

  • Framework theatre

    Use RAPID or DACI on at most 3 to 5 recurring decisions at first.

The biases behind them

  • Hindsight bias

    After the outcome, people think they knew it. Cassar and Craig found it among founders who never formed an operating business. Record the probability and reasons at decision time.

  • Authority bias

    People defer to the founder or an investor. Collect written input before discussion and let juniors speak first.

  • Anchoring

    The first number or option dominates. Collect options in writing before anyone shares a recommendation.

  • Status quo bias

    Choosing no change by default. Add “do nothing” as an explicit option, with its cost.

Where the evidence disagrees

Is RACI useless for decisions?

Yes

McKinsey (2022): too many people get a vote or a veto, and RACI is static and tracks tasks, not decisions.

McKinsey, The limits of RACI: primary source (opens in a new tab)

No

Use RACI for repeating process steps such as month-end close, never for decisions.

Practice: practice

The split is by use. Process steps and decisions need different tools.

Does the Agree role have a veto?

Bain’s article says Agree is a sign-off that cannot be ignored, and also that people in that role do not have veto power to block, undo or reject a decision. The two pull in different directions. Decide the rule yourself and write it on the page.

Bain, Decision Insights: primary source (opens in a new tab)

Efficiency savers

What each step saves. Where nobody publishes the number, it says estimate.

  • Run DACI only when a decision is stuck

    Atlassian budgets 15 minutes of prep and 60 to run, for 3 to 6 people. The saving is fewer unstructured meetings. No saving figure is published.

    Atlassian Team Playbook, DACI: primary source (opens in a new tab)
  • Treat two-way doors as owner decisions

    Removes 1 to 3 hours a week of meetings per manager once adopted.

    Estimate, no published data: estimate
  • Name one person to recommend and one to decide

    Bain: the recommender does about 80% of the work. Unassigned recommenders caused major delay in one mining company case. No time figure.

    Bain, Decision Insights: primary source (opens in a new tab)
  • Decision log with review dates

    Saves 30 to 60 minutes a week of “what did we decide” searches in a 20-person company.

    Estimate, no published data: estimate
  • Spend and headcount thresholds

    Removes 2 to 4 founder approvals a week at Series A.

    Estimate, no published data: estimate
  • Shorter meetings

    Porter and Nohria suggest one-hour meetings could often be 30 or 15 minutes. No percentage saved is given.

    Porter and Nohria, via CNN: secondary source (opens in a new tab)

Rules by country

General information, last checked 29 Oct 2026. Not legal or tax advice.

Disagree with a stance, or have data that should be here?

Email me