Book · Forthcoming 2027
The Predictability Loop
An Operating Model for Enterprise Software Delivery in the Age of AI
AI can accelerate the production of work. Leadership still has to decide what deserves commitment, whether the system improved and what benefit can be defended.
Forthcoming from Apress · March 2027
For technology and delivery leaders
After reading it, you will be able to
- 01See where work and judgement are actually waiting, not where the dashboard says.
- 02Tell whether a change improved the system or only moved the delay.
- 03Connect an operational gain to a benefit you can defend to finance.
- 04Change what leadership rewards, so that honest reporting stays possible.
It is written for technology and delivery leaders, and for the practitioners who operate the underlying measurements.
Before you commit the time
- Running the loop costs about two to three hours a week of meeting time on one team. Setting it up costs more than running it.
- It is the wrong instrument for a team of three or four, for work that doesn’t recur, and where the constraint sits outside engineering.
- A companion toolkit with fillable versions of every artefact will be on GitHub, via the book’s Apress page.
The method
The loop
A weekly control layer for software delivery: read signals you can trust, find the constraint, change the least you can, check it worked, and say only what the evidence supports.
Point at or tab to a stage, a rail arc or a gate to see what it does and where the book covers it.
Stages
- ObserveRead signals that expose system behaviour, including downstream pressure.Ch. 1–2
- InterpretTurn signals into a hypothesis about the constraint, the risk or the change in process.Ch. 3–5
- InterveneChoose the smallest action likely to improve the system without rewarding inflated output.Ch. 6–7
- ValidateCheck whether the action changed the process, not just moved a chart for a week.Ch. 8–10
- CommunicateTell the truth about uncertainty, so leaders can separate local gains from system outcomes.Ch. 11, 14
The rail
- GovernWhat the AI is allowed to decide, and what the moved constraint costs.Ch. 12–13
- SustainKeep the measurement system honest over time.Ch. 15
The four gates
- Data qualityIs the data clean enough to measure?Ch. 2
- StabilityIs the process stable enough to promise from?Ch. 8
- Forecast readinessAre the inputs fit for a commitment-grade forecast?Ch. 10
- DistrustIs the number still safe to present?Ch. 11
From the manuscript
Turning a delivery crisis into a leadership conversation
Chapter 14 — Running the loop across the portfolio
A portfolio commitment has been missed. Leadership wants a name.
When a portfolio is in trouble, leadership reaches for a blame surface. One team. One miss. One manager. One escalation. Flow metrics are useful here only if they stop that reflex from hardening into the operating story.
Maya’s best sentence of the quarter explained no metric at all. It was the reframe.
“We did not miss because Team Delta failed. We overloaded the shared constraint and kept starting work through it.”
That sentence widens the room’s responsibility. Now the conversation can take in funding for the shared service, the sequencing of starts, ownership of the dependencies, the policy on how many initiatives run at once, and whether the organisation rewards starting work more than finishing it. At organisation scale, flow metrics are not there to rank teams. They are there to show leadership where its own system is creating delay.
From The Predictability Loop, a manuscript in progress. Maya and Team Delta belong to the book’s illustrative scenario.
Contents
Part 1 · Why AI needs a control layerObserve
- 1How AI moved the constraint
- 2Signals you can trust
Part 2 · Reading an AI-accelerated systemInterpret
- 3Where did the constraint go?
- 4What’s at risk right now?
- 5Are we keeping our promises?
Show all chapters
Part 3 · Acting on the systemIntervene
- 6Choosing the least-harmful intervention
- 7Which generated work earns commitment?
Part 4 · Rigour you can defend to the boardValidate
- 8Are we stable enough to commit?
- 9Did the system actually improve?
- 10Commitments without false certainty
Part 5 · Running the control layer at enterprise scaleCommunicate, Govern, Sustain
- 11When to distrust your analytics — and your AI
- 12Governing AI inside the engineering delivery system
- 13The economics of AI-accelerated delivery
- 14Running the loop across the portfolio
- 15Keeping the system honest
Appendices: artefact index · statistical methods reference · guarded-AI prompts · the weekly operating cadence
The loop in practice
Related work: Selected work · Advisory · The Art of Creating Self-Organizing Teams
Join the launch list
You’ll get an invitation to the launch session, a live walkthrough of the loop using Flow Observatory, and a note when pre-orders open. Occasional emails, no more than monthly. Unsubscribe at any time.