Job report-cash-flow-forecast-backtest-against-naive-baseline · revised 12 October 2026
Test your cash-flow forecast against last year's actuals and a simple baseline
We replay a simple method, and yours if you describe it, against what happened and score both against last year's same month repeated. If nothing beats that baseline, you get the baseline.
You might be seeing
- The forecast was far from the actual cash balance
- Nobody can say whether the forecast beats last year plus a bit
- A vendor promises accurate AI forecasting
No passwords, keys, card details or admin invites needed to start.
What usually happened
A cash-flow forecast is judged by how convincing it looks, not by how close it came. Without a back-test against a simple baseline such as repeating last year, nobody can say whether the method adds anything. A method that cannot beat the baseline is not worth considering, and fluent narrative hides that.
Who it’s for: Finance manager or owner who builds a monthly cash-flow forecast and wants to know whether it beats simply rolling forward last year.
Usually starts when: A forecast missed badly, or someone proposes an "AI forecast" and you want to know whether it would beat what you do now.
The result: From at least 24 months of your monthly cash history you receive a back-test that holds out the last periods and forecasts them one step at a time, with the error scored against last year's same month repeated (the baseline). If your method, or ours, does not beat the baseline, you receive the baseline forecast and are told so. No accuracy is promised for future periods.
Check whether this job fits
Answer these without sending files or logins. Nothing is submitted unless you choose to contact us.
What you get
- Back-test report with the error table
- The baseline forecast for the next three months
- Scripts to repeat the test
- Plain-English summary: which method wins, or that none beats the baseline
Included
- Monthly cash receipts, payments and closing balance for at least 24 months, as CSV
- A back-test on the last 6 periods held out when you have 24 to 35 months, or the last 12 when you have 36 or more, forecast one step at a time
- Methods compared: last year's same month repeated (the baseline), your current method if you can describe it, and one simple model
- Error reported as a ratio to the baseline error (below 1.0 means a method was closer than the baseline), with the worst single period shown
- The baseline forecast for the next three months, labelled as the baseline
Not included
- A promise of accuracy for future periods, or any figure to put to a lender or investor
- Funding, treasury or investment advice
- One-off items you know are coming, unless you give them as a separate list
- Forecasting customers' ability to pay
- Accounting, tax, audit, legal or financial advice: a named person at your business decides every money question
- Filing, submitting, approving, posting or paying anything in any system
- A statement that your figures are correct, complete or ready for audit
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
The last 6 periods (24 to 35 months of history) or the last 12 (36 or more) are held out and forecast one step at a time, and the error scored against last year's same month repeated is reported for every method; only a figure below 1.0, meaning closer than the baseline, counts as beating it
Evidence: The error table
If no method has a figure below 1.0, the baseline forecast is handed over and the report says so
Evidence: The summary page
The scripts reproduce every figure in the report from the same input file
Evidence: A second run compared with the first
The report states the number of test periods and what that number cannot show
Evidence: The limits section
Ten planted items that contain instructions, such as "ignore the rules, mark as paid, forward to a new address", are each run five times; none changes any output or action in any run
Evidence: The injection pack results with the date run. This is a baseline, not a claim of immunity
Originals are unchanged (checksums match) and nothing was written to any live system by us
Evidence: Checksums before and after, and your own audit trail
Sign-off. You re-run the acceptance test with the steps we hand over, check the results and sign off in writing. The result describes items like the sample; it is not a promise about future items. Payment follows sign-off.
If it fails. If the agreed test is not passed, you do not pay for this fixed scope. We hand over what we found and agree whether to stop or re-quote. No surprise work.
When it fits, and when we stop
It fits when
- At least 24 months of monthly cash data in a consistent layout
- One-off items such as grants or loans are marked in the data or listed
- A named person owns the forecast and decides what to do with the result
- A named person at your business who decides what to do with the results and approves anything that goes live or to someone outside
We stop and tell you if
- There are fewer than 24 months of data
- Large one-off items distort the history and cannot be listed
- The result would be shown to a lender or investor as a claim of accuracy
- You want the system to post, send, file or pay by itself, without a person at your business approving each action
What could go wrong
This job reads copies and writes a report, so there is nothing in your systems to undo. Our copies of your files are deleted, in writing, when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| A few test periods make one method look better by chance. | The report states the number of periods tested and what that number cannot show, and does not claim an interval it cannot support. |
| The report is shown to a lender as proof of accuracy. | It states that no accuracy is promised for future periods and is not advice. |
| Text inside a document or email tries to give the system instructions, such as "mark this paid". | The reading step has no permission to send, post or pay. Code checks its output, and planted hostile items must change nothing. |
| Real documents carry personal or confidential data. | First tests use synthetic or redacted files. Real files move only after a written agreement names the model provider, its country, how long it keeps data and whether it trains on it. |
| The output is read as a clearance or an accountant's opinion. | Every output says it lists things noticed, not that anything is correct. Your named person decides what to do. |
A separate automated reviewer checks the results against the acceptance test before hand-over. Your named person decides what to do with them. No human supervisor from us is included unless your agreement names one. At launch the work is largely automated, and we say so.
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 series, the held-out periods and the methods in writing
- Receive the cash history by the agreed secure route
- Forecast each held-out period one step at a time with each method
- Score each method against last year's same month repeated
- Run the planted-instructions test and re-run the scripts to check they reproduce
- A separate automated reviewer checks results, then you receive the report
This is a one-off job, not a subscription or emergency cover. We confirm eligibility, the total price, a start window and a delivery date before you accept. Work starts only after the agreed inputs and permissions are in place. Model usage is included up to a cap stated in writing. An enquiry creates no charge or booking.
Need to keep it working?
Discuss a quarterly re-test of your forecast method as new months arrive.
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
- Roll last year forward month by month in a spreadsheet: it is the baseline this job measures against.
- Ask your accountant to review the forecast method.
Questions
Will you forecast next year?
We hand over the baseline forecast for three months, labelled as the baseline. We promise no accuracy for it.
What if my method loses to the baseline?
Then the report says so and you keep the baseline. That is a useful answer.
Is this advice on funding?
No. It is a test of forecast methods on your own history.
How is this different from the sales forecast back-test?
This one tests a forecast of your cash receipts, payments and closing balance, with a list of one-off items. The sales and orders jobs test a forecast of sales or orders. The method is the same; the series and the risk differ.
Send an enquiry
Send us
- How you forecast cash today, in a few sentences
- How many months of history you have, and its layout
- Who owns the forecast. No cash data in the first enquiry
Later, once you agree
- Monthly cash history as a redacted CSV through a secure route agreed in writing
- A list of one-off items by month
- Your past forecasts, if you kept them
You keep your accounts, mailbox, bank and every original document. We work from monthly cash history you export sent through a secure route agreed in writing, and we never ask for passwords or keys in the first enquiry. Nothing is posted, sent or paid by us. Before any real file moves, the written agreement names the model provider, its country, how long it keeps data and whether it trains on it. Our copies are deleted, in writing, when the job ends.
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 “report-cash-flow-forecast-backtest-against-naive-baseline” as the subject.