// Playbook — AI value · conceptual · ~5 min

All in on AI. Still waiting for the value.

For organisations that have bought the licences, run the pilots and put a copilot on every desk — and still can’t name the line on the P&L that moved. The problem isn’t too little AI. It’s too much, pointed at nothing in particular. Worked example: Tallowood Bank, a listed regional bank — pick your function.

11AI use cases
· in procurement
3Pilots
· live at once
1Contract tool
· bought first
0Cost-to-income points
· attributed to AI
// See it in your function
01 —

Tallowood’s procurement team.

Procurement · composite · illustrative

An AI contract tool. Three supplier lists underneath it.

Three mergers left procurement with 1,800 suppliers across three vendor masters. When the board went all in, procurement moved first — a contract-review tool, a supplier chatbot, a copilot for RFPs.

The tool reads contracts fast but can’t see that two supplier names are one company. Only 41% of spend is under contract, and the register of material service providers lives in a spreadsheet.

1,800Suppliers
3Vendor masters
41%Spend under contract
12People in procurement
02 —

What all in looks like.

Five symptoms · after Bain

// Bain calls it the AI value paradox: frenetic activity, flat earnings. About four in five CEOs say their AI programme is at best partly delivering.

// 01

The swirl.

More ideas than anyone can prioritise. Effort spread too thin to matter.

// 02

Minutes, not margin.

Faster emails. Enterprise value doesn’t move.

// 03

Tools first.

A copilot on the old workflow. Decorated, not redesigned.

// 04

Pilots forever.

Demos that never reach production.

// 05

Owned by IT.

No visible owner at the top, so it stalls.

03 —

Four shifts.

Value first · tools last

// Not more AI. Different AI. In order — each shift depends on the one before.

01
From

The use-case register

To

The cost-to-income ratio.

Judge every idea against spend under contract, supplier count and material service providers known.

02
From

Choosing a tool

To

Redesigning the work.

Map how a purchase really happens. One vendor master first, AI second.

03
From

Many pilots

To

A few bets, in production.

11 use cases → 2 bets: supplier de-duplication and a material service provider register.

04
From

Launching

To

Landing the change.

Risk co-owns the register. Spend under contract goes to the board.

The tool was good. The data under it wasn’t ready.
04 —

Three ways it stalls.

Pitfalls
// Pitfall 01

Cutting without a story.

Pilots stopped by spreadsheet. The people behind them stop bringing ideas.

The fix

Explain each cut against the value plan, in person. Keep what was learned.

// Pitfall 02

A new swirl, better organised.

A new intake form and steering group. Same volume, same result.

The fix

Measure what gets declined, not what gets captured.

// Pitfall 03

Winning at the pilot.

The bets get a launch. Production quietly loses its funding.

The fix

Fund the path to production up front. Done means the value moved.

05 —

Five questions for Monday.

Start here

// If the leadership team can’t answer these in one meeting, that’s the diagnosis.

Which three to five outcomes does our strategy actually depend on?

Which of those does each live AI initiative move? Name the line.

If we could fund only four, what would we stop tomorrow?

Who, by name, owns AI value — not AI technology?

Have the people who do the work redesigned it — or just been given a tool?

06 —

Using this in practice.

Closer

Fewer bets. Bigger lines.

Conceptual by design: the shape of the problem and the order of the way out. The four shifts hold in every function. What changes is where the function’s line sits — and who asks which line? For procurement, it’s the Chair, against cost-to-income.

// Where this connects — the full method lives in the transformation methodology: Getting past the swirl →

// Reference — Kmet, Zhu, Cooke & Howgego, Getting Past the AI Value Paradox in Private Equity, Bain & Company, 16 September 2026. Symptoms and shifts adapt Bain’s obstacles and steps; the worked example is my own composite.