A half-day session for leadership teams
Your AI spend is rising. Can anyone tell you what it bought?
Seven per cent of senior leaders say they have established a return on their AI investment. The ones who can see where the money goes are five times more likely to be in that group.
A taught half-day for the people signing off AI work. Where the cost comes from, which levers move it, and the questions that separate a real number from a hopeful one. Nothing to prepare, nothing to send in advance.
The problem
The bill is a lagging indicator of a decision made months ago
By the time an AI invoice looks wrong, the architecture that produced it has been in place for two quarters. The choices that set the number were made early, in a design review, usually by people who were not thinking about cost and were not asked to.
Which is why reading the invoice more carefully does not help much. What helps is understanding the handful of decisions that generate it, so you can recognise them while they are still being made.
31 → 98
Per cent of finance operations practitioners managing AI spend: 31 in 2024, 63 in 2025, 98 in 2026. Two years ago this was a niche concern. It is now universal.
FinOps Foundation, State of FinOps 2026, 1,192 practitioners
7%
of senior leaders report an established return on AI. Among those with strong cost visibility it is fifteen per cent; without it, three.
KPMG Global AI Pulse, Q2 2026, 2,145 leaders across 20 markets
24%
of leaders say their CEO is accountable for AI-driven business outcomes. Where accountability is defined, established ROI runs at 14 per cent against 4.
KPMG Global AI Pulse, Q2 2026
The half-day
Four blocks, taught
Worked through on the board with live pricing from the current models, using workloads that look like the ones your teams are building. Questions throughout; the session is small enough for that to work.
01
Where the number actually comes from
What you are billed for and why it moves. Tokens in and tokens out, why the reply costs several times what the question did, what context length does to a bill that looked fine in the pilot, and why the same feature costs a different amount every time it runs.
02
The shape of a workload
Almost no AI feature is one call to one model. It is a chain of steps, and the steps are not equally hard. We take common workloads apart on the board and look at where the money concentrates, which is rarely where people assume.
03
The levers, and what each one costs you
Routing steps to different models, caching, retrieval, batching, smaller models with checks around them. Every lever trades cost against quality, latency or engineering time. We go through what each one buys and what it takes away.
04
The questions that expose a weak number
A set of questions to put to any AI proposal, internal or from a vendor: what happens to this figure at ten times the volume, what is being left out of it, what is measured, what breaks if the model is replaced next quarter. You leave with the list.
Afterwards
What your leadership team can do that they could not before
The session sells judgement, not information. Everything in it exists somewhere on the internet. What is hard to assemble alone is the working picture that lets you challenge a number in the room, while the decision is still open.
- Read a cost estimate for an AI feature and say where it will be wrong
- Ask a vendor the four questions their pricing page is built to avoid
- Tell the difference between a pilot that is cheap and a system that will be
- Judge whether a proposed AI feature is worth building before it is built
- Explain to a board why the spend moved, in language they will accept
Right for
- Technology leadership at organisations with AI features already in production or close to it
- Teams whose AI spend has started to attract questions from finance
- Leaders who approve AI work but do not build it, and are tired of taking the numbers on trust
Wrong for
- Organisations that have not started building with AI yet, where the useful conversation is a different one
- Engineering teams wanting to implement the optimisations themselves, who are better served by a hands-on programme
- Anyone looking for a vendor recommendation. The session is provider-agnostic and does not route you to a product
Who teaches it
Someone who builds with these models and teaches engineers to
I am Tamas Piros. Twenty-five years across software engineering, developer relations and technical training, a Google Developer Expert in Web Technologies and an ambassador for the Agentic AI Foundation. I have delivered 138 talks and workshops in 33 countries, and I teach engineering teams to build AI systems.
That matters here for one reason. The cost of an AI system is decided by engineering choices, and I spend my working life teaching the engineers who make them. The session explains those choices from the inside rather than summarising what analysts say about them. More about me.
Free, monthly, forty-five minutes
See how it is taught before you buy it
Once a month I run a forty-five minute online briefing: thirty minutes on where an AI bill actually comes from and the two decisions that set most of it, then fifteen minutes of questions. Live, and the questions are the half most people get the value from.
Attendance is capped so the questions stay useful. Leave your details and I will send you the next date as soon as it is set. It is free, and it is the front door to the paid half-day, which I would rather say than pretend otherwise.
Or skip ahead
Tell me who is in the room and what they are being asked to approve. I will tell you whether this session is the right one and what it costs.