Job sales-forecast-back-test-against-a-simple-baseline · revised 12 October 2026
Find out whether your sales forecast beats a simple baseline
Your last forecasts are scored against actuals and against a plain seasonal baseline, period by period, with a verdict. If the baseline wins, you get the baseline. No future forecast is promised.
You might be seeing
- Forecasts are revised every month and nobody keeps the old ones
- The board asks whether a spreadsheet guess would have done as well
No passwords, keys, card details or admin invites needed to start.
What usually happened
A sales forecast is rarely scored against what happened, so nobody knows whether the method adds anything. A forecast method that cannot beat a simple repeat of last year is not worth the effort, and a model that sounds clever may not beat it either. Better inputs do not by themselves make a better forecast.
Who it’s for: Founder or sales leader who submits a monthly or quarterly forecast to a board or investor.
Usually starts when: Someone asks how reliable the forecast has been, and nobody has ever scored it.
The result: You receive your recorded forecasts scored against actuals and against a seasonal baseline, period by period, with the error measure written down first and a plain verdict. If the baseline is better, you are handed the baseline and told so. No forecast for the future is promised.
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 scored back-test by period
- The verdict, in the first paragraph
- The baseline forecast and the script
- A note on what the data could not show
Included
- One sales series by month or quarter, aggregate only
- At least 24 months of actuals and at least 12 recorded forecasts
- Scoring of your 12 recorded forecasts against actuals and against a seasonal-naive baseline; they were made at the time, so they are already out of sample
- A simple statistical forecast fitted without the last 6 periods and scored on them, with the baseline handed over if it does not win
- A script that reproduces every figure
Not included
- A promise about future sales
- Scoring individual reps or teams
- Advice on hiring, spend or targets
- Interval or confidence claims on fewer than 30 periods
- Forecast software
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
Every error figure in the report is recomputed from your actuals by a second run, with identical results.
Evidence: Both outputs compared.
The baseline forecast for each held-out period uses only data from before that period, shown by a leak check.
Evidence: The leak check output.
The verdict states in its first paragraph whether your forecast, the simple model or the baseline had the lowest error on the last 6 periods.
Evidence: The first paragraph of the report.
If the baseline is no worse than the simple model, the baseline forecast is handed over and the report says so.
Evidence: The handed-over baseline and the report text.
Sign-off. You re-run the test yourself, check the list of severe errors and sign off in writing. Payment follows sign-off.
If it fails. If the agreed pass mark is not met, we say so, hand over the results and the failures, and do not call the output working. You do not pay for this scope; we agree whether to stop or re-quote.
When it fits, and when we stop
It fits when
- Two full seasonal cycles of actuals and 12 forecasts recorded at the time, not rebuilt afterwards
- Aggregate numbers only
- A named person owns the series and agrees the error measure before the run
We stop and tell you if
- Forecasts were rebuilt after the fact, so the back-test would leak
- Promotions or stock-outs are undated and distort the series
- You want individual reps scored
What could go wrong
The job reads an export and writes a report, so there is nothing to undo.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| The test leaks the future into the baseline | A leak check confirms every baseline value uses only data from before its date. |
| A short series makes a score look more certain than it is | We make no interval claim under 30 periods and report the number of periods. |
| The verdict is read as a forecast for next quarter | The report says it scores the past and promises nothing about the future. |
A separate automated reviewer checks every figure, the baseline and the leak check before hand-over, and no person on our side reviews unless your agreement names one. Your series owner reads the verdict and decides what to change. 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 error measure and the held-out periods in writing
- Receive the actuals and recorded forecasts by the agreed route
- Build the seasonal baseline using only data known before each date
- Score your 12 forecasts, then score the simple model on the last 6 periods
- Independent automated recomputation of every figure and a leak check
- Hand over the report with the verdict first
This is a one-off job, not a subscription or emergency cover. We confirm eligibility, price, a start window and a delivery date before you accept, and work starts only after agreed inputs and access are in place. Nothing is sent, merged, posted or changed in your live systems by us, and an enquiry creates no charge or booking.
Need to keep it working?
Discuss re-scoring each new quarter's forecast.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
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
- A spreadsheet can compare each past forecast with the same month a year earlier in about an hour. This job adds the held-out scoring and a script.
Questions
Will you forecast next quarter?
No. We score the past and hand over a baseline you can reuse.
What if my forecast beats the baseline?
The report says so, with the periods and the error figures.
Do you use AI?
No language model is needed. The simple method is statistical; if a language model is tried it is scored the same way and must beat the baseline.
How is this different from the cash-flow forecast check?
That one replays methods on a monthly cash series. This one scores the sales forecasts you recorded at the time and also scores a simple method on the last 6 periods.
Send an enquiry
Send us
- The series, the period and how long you have recorded it
- How the forecast is made today, in a few sentences
- Whether past forecasts were saved when they were made
Later, once you agree
- The actuals and the recorded forecasts as an aggregate export
- The agreed error measure
- A secure handoff route agreed before any export is shared
You keep the data. We read an aggregate export and return a report and a script.
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 “sales-forecast-back-test-against-a-simple-baseline” as the subject.