AI Economics in 5 Minutes

The problem in one sentence

There is a gap between what organisations spend on AI and the share of that spend they can connect to a measured outcome. That difference is the AI Value Gap. Cost data is precise, real-time and automatic. Value data is manual, delayed and contested. So leaders can say exactly what they spent last month, but not what they gained.

A March 2026 benchmark study of 2,048 enterprise decision-makers found that 79% of organisations report no measurable EBIT impact from AI, despite 70% adoption. Bain’s April 2026 survey of 951 large companies found that among those measuring AI cost savings, roughly 40% came in at 10% or less — below their expected 10 to 20% — and that 44% are funding their next wave of AI from savings that have not yet materialised.

The one number to lead with

Our position, stated plainly

The gap is a governance failure, not a measurement impossibility. Organisations already measure harder things, including marketing attribution and cyber risk. AI simply has not yet been given the same discipline.

Waiting for prices to fall is not a strategy. Per-token prices are falling, but agentic systems multiply the tokens used per task, and providers do not pass all savings through. Expect total AI invoices to keep rising for at least the next few years even as unit prices drop. Cheaper tokens, bigger bills.

Time saved is not the same as value realised. Of every pound of reported time saved, only around twenty pence tends to reach the profit and loss account — once you account for savings too scattered to use, capacity no one redeploys, adoption gaps, and the cost of making AI work in production. The rest is narrative.

The frameworks, in brief

What to do first

You can start with finance extracts, vendor portals and a spreadsheet. In the first 30 days:

1. Name one owner accountable for AI economics, with a mandate from the CFO or CIO.
2. Build one spend inventory in four categories: direct consumption, licensed seats, embedded tiers inside SaaS contracts, and infrastructure.
3. Capture baselines for the top three use cases by spend, before that history ages out of your systems.
4. Define one unit cost per major use case, fully loaded — for example, cost per resolved ticket or per accepted code change.
5. Hold a first value review with finance in the room, publish the pack, and make three decisions: one tool to consolidate, one use case to baseline properly, one contract to renegotiate.

The four questions a leader should be able to answer

Organisations that can answer these are not just measuring AI better. They are governing it better.