Twenty five lessons on setting up and scaling AI FinOps in a large enterprise, published every other Tuesday from 8 April 2025 to 24 March 2026.

It is a serial rather than twenty five standalone essays. One jungle, one cast, two budget cycles, told from inside the rooms where the arguments happen. The Crow runs finance and counts everything. The Fox leads AI and has already promised the board something. The Crocodile has survived mainframe, client-server, virtualization and three separate cloud migrations, each announced as the last one. The animals carry the argument rather than decorating it, which means the fable and the analysis are the same thing rather than two layers stacked on each other.

No vendor, product or provider is named anywhere in the series. Every lesson closes with a Field Kit split by role, so finance, IT and the business each leave with something specific to do, and a numbered Jungle Lesson that stands on its own. The twenty five assemble into a usable framework by the end rather than being a retrospective list.

Act I: The Jungle Floor

Why AI money behaves badly, and why the machinery that worked for cloud does not transfer.

• Lesson 1: Why AI Spend Does Not Behave Like Cloud Spend (8 April 2025). Scene: a bill four times the plan. A cloud bill tells you what you bought. An AI bill tells you what you did.
• Lesson 2: Three Tribes, One Bill (22 April 2025). Scene: the same number explained three ways in one day, all of them true. A disagreement you cannot resolve is usually a translation failure wearing a disagreement costume.
• Lesson 3: Anatomy of an AI Invoice (6 May 2025). Scene: a tour of the platform, opening panels nobody knew were there. Inference is the layer you budgeted and rarely the layer that hurts.
• Lesson 4: Termites in the Walls (20 May 2025). Scene: a four-page list nobody had asked for. Shadow AI is not defiance, it is an unfunded requirement with a credit card.
• Lesson 5: The Pilot Trap (3 June 2025). Scene: somebody finally counts the pilots. A hundred experiments is the same experiment funded a hundred times by people who have not met.

Act II: Learning to Count

The measurement spine. Denominators, allocation, productivity claims that survive scrutiny, and evidence.

• Lesson 6: Choosing a Denominator (17 June 2025). Scene: an offer to fund against any metric, provided you own it for two years. Count what was accepted, not what was produced.
• Lesson 7: Who Pays for the Watering Hole (1 July 2025). Scene: four meetings about a shared platform everyone wants and nobody wants on their books. Decide who funds the shared foundation before you decide how to split the water.
• Lesson 8: The Hours Saved Fallacy (15 July 2025). Scene: a very large number, and one question about where the time actually went. Time saved is not money saved until somebody does something specific with the time.
• Lesson 9: Baselines, Counterfactuals and Holdouts (29 July 2025). Scene: a control group nobody meant to run. You cannot measure a change you did not measure before.
• Lesson 10: The Value Ledger (12 August 2025). Scene: compliance and the AI lead discover they have been asking for the same document. Write down what you promised, and never edit the entry.

Act III: Architecture Is a Financial Decision

Where the unit economics are actually set, which is months before anybody looks at a bill.

• Lesson 11: Frontier or Local (26 August 2025). Scene: a magnificent presentation and one question about utilization. Two cost curves, not two products. The only questions are where they cross and how much you trust the volume.
• Lesson 12: The Routing Layer (9 September 2025). Scene: a weekend of plumbing that nobody notices for a month. Most of what you send to the cleverest animal does not need the cleverest animal.
• Lesson 13: Context Is the New Compute (23 September 2025). Scene: a two-line question wrapped in six pages of everything else. You are billed for the question as well as the answer, and the question is usually bigger.
• Lesson 14: Data Gravity and the Control Plane (7 October 2025). Scene: a constraint that turns out to have been a cost driver for a year. Price the constraint as a design input rather than arguing about it as a tax.
• Lesson 15: Agents and the Multiplication Problem (21 October 2025). Scene: twelve thousand operations over a weekend, no errors, no alerts. Autonomy is where you stop buying answers and start buying attempts. Budget the tail, not the average.

Act IV: The Operating Model

Capability-led funding, the operating rhythm, guardrails, and the commercial posture underneath all of it.

• Lesson 16: Capability, Not Project (4 November 2025). Scene: somebody proposes a structure that shrinks their own function. A project is funded to end and a capability is funded to continue. Only one gets cheaper the twentieth time.
• Lesson 17: From Annual Budget to Persistent Funding (18 November 2025). Scene: the CFO proposes giving up the forecast. Fund persistently, reallocate quarterly, and cut something visible in the first year.
• Lesson 18: Inform, Optimize, Operate (2 December 2025). Scene: six weeks in which nothing goes wrong, which nobody knows what to do with. Optimization has moved from the contract to the code, and cost and quality cannot be reviewed in separate rooms.
• Lesson 19: Guardrails That Do Not Strangle (16 December 2025). Scene: a cap, and the quiet reversion that follows it. A ceiling lands hardest on whoever is using the thing most, which is usually whoever is getting the most out of it.
• Lesson 20: Commitments and Capacity (13 January 2026). Scene: a genuinely good discount, and one question about the engineering roadmap. A commitment locks your volume, your price, and quietly your capability tier. Only two are on the term sheet.

Act V: The Long Game

What happens to all of it once time passes. Decay, decommissioning, assurance and org design.

• Lesson 21: The Second Year Problem (27 January 2026). Scene: a two-year-old slide titled Conservative Estimate, found in an archive and read aloud. Nothing you optimized stays optimized, and year two has to be funded by evidence rather than novelty.
• Lesson 22: Nobody Gets Promoted for Switching Things Off (10 February 2026). Scene: something switched off after two years, purpose unknown, owner departed. Write the sunset criteria at launch, while nobody is attached to it yet.
• Lesson 23: The Cost of Trust (24 February 2026). Scene: compliance and the AI lead presenting together, which nobody saw coming. Assurance is not what you pay to satisfy the tortoise, it is what you pay for the right to change anything quickly.
• Lesson 24: Growing the Troop (10 March 2026). Scene: a hiring decision reversed in public. Spend on literacy rather than on tooling that answers questions nobody knows how to ask.
• Lesson 25: Field Notes from the Canopy (24 March 2026). Scene: a bigger bill that everybody in the room can explain. You cannot manage what you cannot divide.

The pattern across all twenty five

Every one of these lessons is a measurement problem wearing a cost problem’s clothes. The bill that arrived four times too large, the productivity case that fell apart under one question, the cap that killed adoption while the numbers improved, the thing that ran correctly for two years serving nobody. In each case the money was the symptom and the absence of an honest denominator was the cause.

Which is why the series ends where it does. AI FinOps is not a cost discipline that happens to involve measurement. It is a measurement discipline that produces cost outcomes, and organizations that get that the wrong way round will cut the wrong things, quickly, with great confidence, and will not find out for two years.

The full archive lives under Scaling AI FinOps in the Enterprise Jungle. The series concludes four days before In Practice: AI in the Enterprise begins, and the two are designed to be read in that order: this one covers the money, the other covers the governance.

This is one of three series. The others are In Practice: AI in the Enterprise on governance, and In Practice: Building an AI Company on the view from the other side of the table. All three are listed on the series page.