Job warehouse-ai-function-cost-estimate-and-cap · revised 12 October 2026
Before we run AI over a million rows, tell me what it will cost and cap it
A sampled cost estimate with a range, a per-job cap and a usage query that reproduces the billed credits, checked against a 1,000-row trial that you run in your own account.
You might be seeing
- Nobody can say what the full run will cost
- A previous test ran up more credits than expected
No passwords, keys, card details or admin invites needed to start.
What usually happened
Snowflake's AI functions bill by tokens. Label text and examples are counted for every record, and a document page can count as many hundreds of tokens. A trial on a thousand rows tells you little about a million unless the token counts are modelled.
Who it’s for: Data lead about to run a warehouse AI function over a large table.
Usually starts when: A pilot ran on a thousand rows for pennies, and nobody can say what the full table will cost.
The result: A sampled token and cost estimate with a range, a per-job cap, and a usage query that reproduces billed credits. The estimate is checked against a 1,000-row trial that you run in your own account and is within an agreed band of the billed cost. The full-table range scales the billed trial's cost per row to your row count, with the low and high from the sample's text-length spread; the full run is not tested.
Check whether this job fits
Answer these without sending files or logins. Nothing is submitted unless you choose to contact us.
What you get
- The estimate with its range and assumptions
- The per-job cap setting
- The usage query
- The trial comparison
- Run steps
Included
- One warehouse AI function applied to one text column
- A sample of 200 non-personal rows to measure token counts
- A cost estimate with a low, expected and high range for the full table
- A per-job cap and a usage query that reproduces billed credits
- A 1,000-row trial that you run in your account, compared with the estimate
Not included
- Running anything in your account
- Cost advice on the rest of your warehouse bill
- Choosing which function or model to use
- Credentials or logins
- A guarantee that the full run will cost what is estimated
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
The estimate for the 1,000-row trial is within an agreed band of the billed cost (we propose within 25%). This is one trial, so it checks the method on one function and column, not every future run.
Evidence: The estimate, the billed cost from your usage view and the difference.
The usage query reproduces the billed credits for the trial to the unit.
Evidence: The query and its output beside the bill.
A deliberately small cap that you set in your own account stops a trial job at the cap.
Evidence: The capped trial run and the stop record from your usage view.
The estimate states its range and its assumptions: row count, text-length spread, and what is counted per record.
Evidence: The assumptions list.
Sign-off. Your developer runs the trial and the usage query, inspects the comparison and signs off in writing. Payment follows sign-off.
If it fails. If a check above is not met, you do not pay for this fixed scope. You keep the results and the failures, and we agree whether to stop or re-quote. No surprise work.
When it fits, and when we stop
It fits when
- A warehouse with AI functions and a usage view you can query
- A sample of 200 non-personal rows, or aggregate token counts
- A developer who runs the 1,000-row trial and the usage query
- You can set a per-job limit in the warehouse
We stop and tell you if
- No usage view exists
- The trial cannot be run in your account
- The sample holds personal data that cannot be removed
- You want us to run the job
What could go wrong
The job reads copies you send and writes files. It changes nothing in your systems, so there is nothing to revert. Our copies of your files are deleted when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| The estimate is far from the real bill. | It is checked against a billed 1,000-row trial. If outside the agreed band, the report says so and shows why. |
| A cap does not stop a job. | The cap setting is tested with a deliberately small cap in the trial. |
| Your files go to a model provider outside your control. | Before any file is shared we name the provider, its country and its retention terms in writing. Entry jobs run on redacted, synthetic or public copies, or on your own provider account with a key you hold. |
A separate automated reviewer checks the estimate, the cap and the trial comparison before hand-over. A developer on your side runs the trial and the usage query and signs off. No person on our side is included unless your agreement names one, and at launch the work is largely automated.
How we deliver
We arrange the work and independent review, then show you the result against the agreed checks. You keep authority over your systems.
- Agree the function, the column, the cap and the allowed estimate band
- Measure token counts on the 200-row sample
- Model the full-table cost as a low, expected and high range
- You run the 1,000-row trial and send the usage query output
- Independent review of the estimate against the billed trial
- Hand over the estimate, cap and queries; delete our copies
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, the price, a start window and a delivery date before you accept, and work starts only after inputs, a secure handoff route and permissions are in place. Model usage up to a cap agreed in writing is included, and an enquiry creates no charge or booking.
Need to keep it working?
Need the real run set up with the cap in place? Ask for a quote.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
Explore an ongoing managed lane, or mention the responsibility you need in your enquiry.
What you can check
This is a new service. We have not delivered this job for a client yet.
Other ways to get this done
- Your warehouse's own cost documentation and usage views explain the billing. Read them first and run a trial. docs.snowflake.com
- Our cloud bill review covers non-AI cost on servers, not warehouse AI functions.
Questions
Will you run the job?
No. You run a small trial in your account and the full job when you choose.
Is the estimate a guarantee?
No. It is a range, checked against a billed trial.
Does it work on other warehouses?
The method does. Billing details differ, and we read your provider's own documentation.
Send an enquiry
Send us
- Which warehouse and which AI function
- Roughly how many rows and how long the text is
- No real records, personal data or credentials in the first enquiry
Later, once you agree
- A 200-row non-personal sample or token counts by an agreed secure route
- Your trial results and the usage query output
- A named developer who runs the trial
You keep your warehouse and your account. We work on a sample or token counts you send by an agreed secure route, return an estimate and queries you run yourself, and delete our copies when the job ends. We take no credentials.
Email fallback: open your mail app
If website submission is unavailable, review and send the fallback email yourself. An email fallback is not a website receipt. Or write to hello@syntheticindustry.ai with “warehouse-ai-function-cost-estimate-and-cap” as the subject.