AI Spend Forecast
The AI Spend Forecast is a three-week, fixed-price service from ETT that turns your AI invoices and workloads into model choices, a verdict on seats against tokens, and a budget your finance team can sign off. It runs on Analect.
AI spend arrives as one line on an invoice
Nobody can say which workload drove it, whether the model was the right one for the job, whether the seats you pay for beat paying per token, or what next quarter looks like.
Prices move every month, many organisations now buy from several AI vendors at once, and the people approving the spend are asked to sign a number nobody can explain. That is an operating problem, and a cheaper model does not fix it.
- Seat licences and token pricing bought side by side, with no fair way to compare them.
- Work running on the most expensive model, because nobody priced the alternatives.
- Budgets set from last month's invoice rather than from the work itself.
Four answers, each with its working shown
Your current spend, explained
We read your provider usage exports and break the spend down by workload and model. You see what you paid against list price, and your effective discount.
The right model for each job
Every workload priced across around 200 models from 27 providers, filtered by your rules on data residency and approved vendors and by what the job needs. Each pick comes with the reason it won and what it beat.
Seats against tokens
For every seat licence you hold, a verdict against the same vendor's pay-per-use pricing, with the break-even point. We hold the vendor constant, so the comparison is a fair one.
A budget you can defend
Your recommended monthly and annual run-rate by workload and by vendor, what it becomes at half and double the volume, and what a spending ceiling buys if you have one.
What each workload row holds
One row per workload. Finance signs off the report, so every row carries the reasoning as well as the number.
Written for whoever approves the spend. The finding in one paragraph, one row per workload with the model and the sentence that justifies it, the reasoning in full, caching priced as a separate lever, and every assumption named. Delivered as PDF, with your team's own logins to the working forecast for the length of the engagement.
- Workload
- The job being priced, described in business terms rather than as an API call.
- Volume and shape
- Requests per month, the size of each request and response, and tool calls per task.
- Recommended model
- The pick and its price, with the date of the catalogue it was priced against.
- Why it won
- One sentence, and the models it beat for this job.
- Seats or tokens
- Where a seat licence covers the work, the verdict against pay-per-use pricing and the break-even point.
- Caching
- Priced as its own lever, because it moves the bill whatever model you choose.
- Cost
- Measured from your bills or estimated from a stated assumption, and labelled as which.
Three weeks, about four hours of your people’s time
The three weeks start when we have your usage exports.
- Week 1
Gather
Kick-off. We take in your usage exports, seat licences and the workloads you want priced, and agree your rules.
Yours: Send the exports and a short list of workloads. Name a sponsor.
- Week 2
Model
A working session where we walk the forecast with your team and push on the numbers live. Your people get their own logins.
Yours: Two people for ninety minutes. Challenge every assumption.
- Week 3
Report
We finish the report and the budget and present them to your sponsor and finance lead.
Yours: A sixty-minute readout.
From you
- Usage or billing exports from each AI provider for the last one to three months. Any CSV with honest column names works.
- Your seat licences: vendor, number of seats, price, and which teams use them.
- Five to fifteen workloads you want priced: what the work is, how often it runs and roughly how big each request is. We help you describe them.
- Your rules: data residency, approved vendors, anything that must stay in-house.
- A sponsor, and about four hours of your people's time across the three weeks.
Where it goes
- Your billing files are read in your browser and never uploaded. We keep the forecast built from them: workloads, model names, token counts and costs. You can save it without the cost figures if you prefer.
- We need no documents, customer data or prompts. If you want one prompt priced, paste it in: it is counted, not stored.
- The forecast sits in a private store that only the named people at your organisation and at ETT can open, and is deleted at the end of the engagement unless you ask us to keep it.
What the forecast is not
Not a vendor quote
List prices come from a dated catalogue and the date is on the report. Your negotiated rates come from your own bills.
Not a verdict on everything a seat buys
A seat is also an interface, administration and support. We say so beside every comparison.
Not an instruction to switch
Where a cheaper model fits the job, we tell you what to test before you move anything.
Not a black box
Every figure is labelled as measured from your bills or estimated from an assumption we state.
One fixed fee covers
- Up to 15 workloads, 4 AI vendors and 5 seat plans
- Two working sessions, the Forecast Report and a readout for your sponsor and finance lead
- 30 days' access to the working forecast after delivery
- Optional: a quarterly re-forecast against your latest bills and the current price catalogue
What we only know once we see your bills
- Which of your provider bill exports parse cleanly, and which need mapping
- Your actual workload mix, taken from the bill rather than estimated
- Whether your contract pricing differs from list pricing, and by how much
About the AI Spend Forecast
How long does the AI Spend Forecast take?
Three weeks from the day we have your usage exports. Your people spend about four hours on it across the three weeks.
Do you need our documents or prompts?
No. We need billing or usage exports, your seat licences and a short description of each workload. If you want one prompt priced, you can paste it in; it is counted, not stored.
Which providers and models do you cover?
Around 200 models from 27 providers, from a dated price catalogue. The date is on every forecast, because a price without a date cannot be checked. Your negotiated rates come from your own bills.
Will you tell us to switch models?
No. Where a cheaper model fits the job, the report says so and says what to test before you move anything.
What happens after the readout?
The report, the budget and the working forecast are yours. You can add a quarterly re-forecast, which refreshes the forecast with your latest bills and the current price catalogue and reissues the report.
Start with a thirty-minute scoping call
We agree the workloads, vendors and seat licences in scope. Then an order form, an NDA you can accept by email, and the three weeks start when we have your usage exports.