The CTO briefing

Decision Ownership After the Rubber StampGoverning a consequential decision once it is approved

A short brief for CTOs and programme directors. It covers the owner gap after approval, what the public record shows on Novopay and the Boeing 737 MAX, how an approved decision can fail, what makes a decision governable, a worked example and a Watchcard.

By Gareth A Holebrook, November 2026. The text of the briefing, as in the PDF.

In brief

  • After approval, the programme and its budget each keep an owner. Whether the decision behind both still holds rarely has one, and that is where the cases in this brief go wrong.
  • Novopay's go-live decision was made at a stage gate whose criteria had not been met, and the gate had no stated consequence for missing the criteria.
  • An approved decision can fail in 16 recognisable ways, and each one appears in a public record I have written up, from inquiry reports to court judgments.
  • Kettei adds one discipline to the governance you already run: 7 properties declared before a decision goes live, and one named owner, often the sponsor, holding it until it ends.
  • Kettei is Japanese for the act of decision making. It is the title of the work that takes my self-governance discipline of Ishi to organisations, where intent exists outside a single skull.

The problem, as I have seen it

Over the past 30 years I have worked with leadership teams in technology, and for most of the last 15 I have been in the room as a facilitator. The biggest impediment I have seen is decision latency, the weeks and months a consequential decision waits to be made while the work it would authorise waits in a queue.

Latency is half the problem. The second half begins with the thud of the rubber stamp, when the decision moves into execution. The programme keeps its sponsor, its stage gates and its business case, while checking whether the decision still holds stops being anyone's explicit job.

How an organisation governs a decision once it is made decides whether it gets what it approved, and I have rarely seen one do it on purpose. Large companies are slow to decide and very hard to move once a decision is made, so a strategy agreed 3 years ago, on assumptions left unchecked since, is defended long after its premise has collapsed.

Startups decide on a dime and decide again before anyone has executed the first call, and I have seen departments burn out chasing direction changes that went unrecorded as decisions. In neither case was the decision treated as a thing with an owner, a boundary and a set of conditions under which it ends.

What the public record shows

Novopay, the payroll for New Zealand's teachers and school staff, went live on 20 August 2012. The ministerial inquiry found that "The actual Go Live decision was made on 31 May 2012", although the criteria for the second of 2 stage gates had not been met. The gates had explicit criteria, but "did not articulate the implications of not meeting the criteria".

Boeing launched the 737 MAX in August 2011 as a derivative of the 737, and that classification was the founding decision of the programme. In March 2016 its managers approved a redesign that raised the authority of the MCAS flight-control software from 0.6 to 2.5 degrees.

The House Committee called it "a significant technical change with dramatic operational implications" and found that Boeing did not reevaluate the system afterwards. The 2 crashes that followed killed 346 people.

In Kettei terms, Novopay's gate had criteria and no abort condition, and Boeing's redesign changed an assumption the founding decision stood on, with no reassessment of that decision. Neither report uses those terms, and the interpretation is mine.

Sources: Ministerial Inquiry into the Novopay Project (Jack and Wevers, June 2013), printed pages 10 and 43. US House Committee on Transportation and Infrastructure, The Design, Development and Certification of the Boeing 737 MAX (September 2020), pages 5, 20, 21 and 103.

How an approved decision can fail

An approved decision can fail in recognisable ways. I count 16, and the right-hand column names a case I have written up where each one appears in the public record.

Failure modeWhat it looks likeIn the record
Decision VacancyNo single person holds the decision's continuing validity. "Who is driving this?"UK Ajax armoured vehicle
Authority CaptureAuthority taken in a crisis stays after the crisis, and commitments made inside execution go undeclaredPrevention of Terrorism Act
Strategic-Tactical ChasmA board decision reaches delivery as a task list, with no scope or conditions the people doing the work can govern againstTarget Canada
Coupling Blindness2 coupled decisions are treated as independent. One changes and the other does not hearNovopay
Unforged GovernanceA practice is adopted and adapted before anyone drilled it to competenceGM and NUMMI
Documentation-Reality DivergenceThe written process and the process people actually run have separatedPiper Alpha
Capability-Governance MismatchDecision authority grows faster than the capability to govern itUber
Assumption DriftThe world moved and the decision stayed put. "We decided this last year."Peloton
Trigger Register DecayTriggers are written down, and the register goes unwatchedEquifax
Normalisation of DevianceDeviations become background noise. "That error rate is normal for us."Challenger
Signals IgnoredThe trigger fires and the owner does not actMid Staffordshire NHS Trust
Escalation of CommitmentPersisting because of what has been spent. "We have come too far to stop."Taurus, London Stock Exchange
Ungoverned PersistenceNo triggers and no abort criteria, so the organisation has no definition of when to stopKnight Capital
Governance TheatreDashboards green and forums held, with no decision changed by eitherCarillion
Reflection TheatreThe post-mortem is filed, and 12 months later the same failure recursColumbia
Institutional MoralisationQuestioning the decision reads as questioning the institution, so it goes unrevalidatedPost Office Horizon

The cases rest on inquiry reports, regulators' findings, court judgments and company filings, read at source. Uber is the exception, because the investigators' findings were not published, so the failure mode there is my inference from the remedies the board adopted.

What makes a decision governable

Most programmes already have a sponsor, stage gates and a business case that is meant to stay justified. Kettei works at a finer grain, on the few consequential decisions inside a programme, and gives each one a single named owner, who can be the sponsor, so it adds no committee.

Each decision gets 7 properties, written on its Watchcard before it goes live. I treat the 7 as one object, because a decision with an owner and no abort criteria is still ungoverned, and so is a decision whose triggers go unwatched.

PropertyWhat it means
OwnerOne named person accountable for the decision's continuing validity. Not shared, vacant or a committee
ReversibilityHow hard the decision is to unmake, judged by the cost of changing it against the cost of delay. Set at creation and monitored, up to a point of no return
Lifecycle stateProposed, Active, Suspended, Superseded, Revoked or Closed. Each change of state is recorded with a date, a name and the evidence
CouplingWhat the decision depends on and what depends on it. When it is Suspended, each dependent decision goes to mandatory revalidation
ScopeWhat the decision covers, what it leaves out and what authority it delegates
Abort criteriaObservable conditions for stopping, tuned to this decision and set before execution begins. Only the owner declares the abort
Validity boundaryThe assumptions the decision stands on, a horizon by which it is reviewed regardless, and the triggers that force an earlier review

In flight, a trigger is the andon cord. The pull opens one question, and the owner answers it inside the window on the card with AHA (pronounced A-Hey). Adjust changes the tactic and keeps the strategy, Hold holds the line, and Abort ends the decision. When the owner judges that a boundary may have been breached, the decision is Suspended while it is revalidated, and a confirmed breach ends it.

Kettei is for the small cluster of decisions a programme can least afford to get wrong. A decision earns a Watchcard when it meets 2 or more of 5 tests: hard to reverse, coupled to other decisions, lasting beyond a quarter, consequential, or resting on an assumption. Below that bar the team runs it as an operational decision.

How it runs: a worked example

WANZCORP Bank is a fictional teaching case of mine. Its board approved closing 189 inbound enquiry roles once a voice bot was live on the line, and this is that decision's Watchcard in brief.

FieldEntry
DecisionClose 189 inbound enquiry roles once the voice bot is live on the inbound line
OwnerThe Executive General Manager, Customer Contact. One named person
Principal assumptionThe bot takes calls off the line as forecast
TriggerWeekly inbound volume against the forecast, watched by the workforce planning lead, answered within 5 business days
Abort criterionIf volume has not fallen as forecast by 8 weeks after go-live, no termination notice goes out
Point of no returnThe date the termination notices go out
Depended on byThe results-day statement to the market, and the staff consultation

At 6 weeks the trigger fires. The Owner holds for the 8-week data, and at 8 weeks the abort criterion is met, so the Owner declares the abort and the decision is Revoked before any notice goes out. Had the Owner only suspected a breach, Suspending the decision would have sent the market statement and the consultation to revalidation with it.

The card does not make the call for the Owner. An Owner can still read the data badly or decline to act, and Signals Ignored is a failure mode because it happens. What the card changes is that the call is visible, dated and owned, and watchkeeping shows each week which cards were reviewed and which were not.

Writing a Watchcard is done once, as the decision is formed. The recurring review runs inside a steering meeting you already hold, where each card is read and owners speak only on exception, and a fired trigger does not wait for that meeting.

I hold the review to 20 minutes at most, and about 15 is typical in my clients' steering meetings. Start with one programme and one card per steering decision, and the blank card on the next page is yours to copy.

Where to go next

Each Kettei essay takes one case from inquiry reports, regulators' findings, court judgments or company filings, and reads it through the failure modes. My third book, Kettei: The Discipline of Governed Intent, is in draft on Leanpub as I write it. If you read 3 essays, start with these.

EssayWhy read it
11 Decisions Taken One at a TimeNovopay, the New Zealand case, where the real go-live decision was made almost 3 months before go-live
45 minutesKnight Capital, and what a missing stopping condition costs at the speed of software
The Honest YieldWhat a stopping rule written in advance does produce, and the limit of what Kettei claims

The cases so far are below, each linked to its essay. The longer analyses, the research behind the essays and the working templates go to the Authors Table, my Substack publication for readers who want that depth.

The PDF adds a blank Watchcard to copy. The cards are also in the free library.

Kettei Workshop: The Board Question

POA. Email for details.