Token burn is the cost nobody is watching.
Analyse. Find where your tokens go, and what to change.
AI spend now follows tokens, not seats. Most businesses cannot say which tools, teams and models burn them, or whether the model fits the job. Our advisers, human and agentic, show you, then help you cut the burn and forecast what comes next.
Tokens are the new meter, and almost nobody is reading it.
rise in tokens used between January 2025 and April 2026, while AI spend rose 497%
Ramp AI Index, 2026 (vendor data)Usage grows with every prompt, agent run and long context, so cost climbs faster than headcount or licences and no one owns the total.
Without a view of which model each task needs, expensive models handle jobs a smaller one could do.
Long histories, large instruction files and pasted documents travel with every request, and each person writes their own prompts. Nobody has decided what context is worth keeping.
Tokens are rising faster than the money spent on them, and the invoice does not say why.
Between January 2025 and April 2026 tokens used rose far faster than spend. A bill shows the total. It does not show which team, tool or task produced it.
Animated chart comparing growth in tokens used with growth in AI spend
Tokens used (the units AI is billed in)+1,001%
AI spend+497%
Usage has grown about twice as fast as spend.
Where your tokens go, what to change and what next quarter will cost.
Worked out from your usage data, not a rule of thumb. We do not promise a fixed saving.
Routine work stops running on the most expensive model.
A forecast of AI spend you can plan against, with the assumptions shown.
Instruction files are plain-text files, written in a format called markdown, that tell AI tools how to behave. We review them and your prompts.
What to keep between requests and what to drop, so you stop paying to resend it.
Advice on structuring teams, roles and shared context so AI is used efficiently.
Our advisers, human and agentic, take you from questionnaire to findings.
Most routine work does not need the most expensive model.
Matching each task to a suitable model is one of the plainest ways to cut cost without losing quality.
Animated chart matching tasks to the model that suits them
- Reading, drafting, summarising or sorting varied language.
- A person can check the result quickly.
- A mistake is cheap to catch and fix.
- The answer must be identical every time.
- A database lookup or form check does the job.
- The decision is about credit, hiring or medical matters, where people decide.
Part of every analysis is checking whether a task needs AI at all. Sometimes a rule, a formula or a workflow tool is cheaper, quicker and easier to defend.
Send each task only the context it needs, and stop paying to resend the rest.
Company-wide context, role context and task context each have a job. Keep the first two lean and stable, and add task detail only when it is needed.
Animated chart of company-wide, role and task context layers
- Company contextShared by everyone
Who you are, how you write, the rules that apply to all work. Kept once and kept current.
- Role contextShared by a team
What finance, legal or sales needs to know to do its job well. Added for that role only.
- Task contextUsed once, then dropped
The documents and facts for this one job. Loaded when needed, so tokens are not spent carrying the rest.
Beyond the invoice: how your teams actually use AI.
We break token use down by team, tool and task, so you can see where it concentrates.
Which tools and models are in use, who pays for them, and what each is used for.
We review the instruction and context files your teams and agents rely on, and bring them under control.
We assess how prompts are written and reused, and set standards that cut waste and improve results.
Which context is worth keeping between requests and which to drop.
What belongs in company-wide context and what belongs to a role, so each task carries only what it needs.
Get a first estimate of what you could save on AI cost.
Enter your monthly AI spend and a few assumptions to see an indicative range for what better model choice, shorter context and tidier instruction files could save. It is an estimate from the figures you enter, not a quote, a forecast or a promise of saving. Share your usage data for a figure based on your own spend.
AI cost calculator
Indicative, from the numbers you enter
- Only the work a smaller model could do is moved. Premium models stay for the rest.
- This covers model choice only. Your health check looks at more, using your own usage.
Items marked assumption can be edited. No figures are stored or sent.
How this is worked out
Monthly saving = monthly spend x share on premium models x share a smaller model could do x price difference. The low and high figures move the middle share down and up by the range you set. Annual figure = middle monthly figure x 12.
Built for the people who answer for AI spend and risk.
Know where AI money goes and whether it is under control.
One number for AI spend, a forecast, and a plan to bring it down.
See which tools, models and instruction files burn tokens, and fix them.
Analyse shows each kind of business where its tokens go.
Choose the tab that sounds most like your business.
Visibility from the first hires
Habits set early tend to stay. We map the tools and models your small team already uses, set a simple rule for context and instruction files, and give you a forecast so a growing bill does not surprise you.
No IT department and plenty of AI
Tools arrive on expenses and personal accounts. We find what is in use, show where tokens go, and leave you with one policy, company accounts and one number a month for the owner.
See what suits small businessesEvery department bought its own AI
Teams use different tools and models, often for the same job. We compare them side by side, show which tasks justify a premium model, and give finance one view and a plan to bring the bill down.
See what suits mid-market firmsOne picture across business units
Large estates collect tools, agents and instruction files. We build a single view of tokens by unit, separate shared context from role context, and give the board a forecast it can plan against.
See what suits enterprisesCost and control in one review
Where AI touches customers or regulated work, cost is only half the question. We review spend and model fit, and show where Guard and Evidence would help you control and prove what AI does.
See GuardAnalysis is delivered through our advisory service, by human and agentic advisers.
Advisory is the only way we offer Analyse. Want to see the process and what you will receive? Schedule a demo and we will walk you through it.
Keep reading
Straight answers on Analyse.
What is GenWA Analyse?
GenWA Analyse is our advisory analysis of your AI use and token cost. Our advisers, human and agentic, review your usage data, talk to your business about what people do with AI, and show you where tokens are burned, which models fit which tasks, and what your spend is likely to do next.
How does the analysis work?
It starts with a short questionnaire, followed by conversations with the people who use AI. You then share an export of tool and token usage, including enterprise data where you have it. We review cost, model fit, context and instruction files, and report savings, model recommendations and a spend forecast.
What usage data do you need?
Exports of tool and token usage from your AI vendors and platforms, plus the instruction files and prompts your teams rely on, where you can share them. A local usage check can read usage files in your browser without any AI calls. Data handling terms are confirmed in the security review before anything else is shared.
How do you reduce token burn?
We find where tokens go by team, tool and task, then check whether each task needs the model it uses. We also review prompts, instruction files and how much context is stored and resent on every request. The result is a ranked set of changes and a forecast, so you can act on the biggest sources first. We do not promise a fixed saving.
What are instruction files and why do you review them?
Instruction files are plain-text files, written in a format called markdown, that tell AI tools and agents how to behave, such as company-wide guidance or role-specific instructions. Large, duplicated or contradictory files inflate every request. We review them, separate shared company context from role-specific context, and set rules for what to store and what to drop.
Can I get an estimate before sharing any data?
Yes. The cost calculator on this page gives an indicative range from the figures you enter. It is an estimate, not a quote or a promise of saving. Sharing your usage data gives a figure based on your own spend.
How is Analyse priced?
Analysis is part of our advisory service and is scoped to the size and complexity of your business. We confirm the price in writing before you commit. The quickest way to start is the AI health check, or by sharing your AI usage data so we can scope it with you.
Find out where your tokens go.
Prices are confirmed in writing before you commit.