Job forecast-beats-naive-baseline-or-says-so · revised 12 October 2026
Forecast next quarter's orders and show it beats repeating last year, or say it does not
For one series or up to 50 SKUs, a forecast with an interval and a rolling backtest against naive and seasonal-naive methods. If ours is no better, you get that finding and the baseline forecast.
You might be seeing
- Stock orders rest on a spreadsheet trend line nobody has tested
- A forecast was bought or built before, and nobody knows if it beat a simple rule
No passwords, keys, card details or admin invites needed to start.
What usually happened
A forecast is only worth having if it beats a simple rule such as 'same as last year'. In a peer-reviewed study, forecasters built on language models did not beat their simpler versions. Without a test on periods the method never saw, nobody knows whether the forecast adds anything.
Who it’s for: Owner or operations manager who orders stock from a spreadsheet extrapolation.
Usually starts when: Stock has been run out of or overbought, and the next order needs a number better than 'same as last year plus ten percent'.
The result: For one series or up to 50 SKUs, you receive a forecast with an interval and a rolling backtest against naive and seasonal-naive forecasts. If our method does not beat the seasonal-naive baseline by the agreed margin, you receive that finding and the baseline forecast.
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 forecast file
- The backtest table: error per method and horizon
- The method note and limits
- The list of promotions, holidays and stock-outs used
- The script and run steps
Included
- One series or up to 50 SKUs with at least two full seasonal cycles before the first test period, plus 12 test periods
- A forecast with an 80% interval for the agreed horizon
- A rolling-origin backtest against mean, naive, seasonal-naive and drift methods
- A list of the promotions and holidays used, and stock-out periods flagged
- A plain-English limits note, the script and run steps
Not included
- Placing orders or changing stock
- A promise about future accuracy
- Pricing decisions
- Forecasts about individual people
- Series with fewer than two seasonal cycles plus 12 test periods
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
Errors are computed only on periods never used for fitting, with at least 12 test periods. The method and its settings are fixed before those periods are scored, and the score is reported once, whatever it shows. Interval coverage is claimed only with at least 30 test periods; otherwise the report says it cannot be judged.
Evidence: The backtest table and the fit-and-test split.
Our method's scaled error beats the seasonal-naive method by the agreed margin (we propose below 1.0), or the report says 'no improvement' and delivers the baseline forecast.
Evidence: The error table and the report's first paragraph.
Running the script twice gives identical forecasts.
Evidence: Both outputs compared.
The promotions, holidays and stock-out periods used are listed, and the report names the series where they weaken the result.
Evidence: The assumptions list.
The report promises no future accuracy. Where a statistical method is what runs, the report says so and does not call it AI forecasting.
Evidence: The report text.
Sign-off. You inspect the backtest and the limits, and sign off in writing. Payment follows sign-off, including when the finding is 'no improvement'.
If it fails. If a check above is not met (for example the figures do not reproduce), you do not pay for this fixed scope. A finding of "no improvement" is a valid result and is paid for. 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
- At least two full seasonal cycles before the first test period, plus at least 12 test periods (36 months for monthly data), with dates of promotions and stock-outs if known
- Each series has at least 30 non-zero periods; intermittent series get the baseline forecast only
- No personal data in the series
- A named person receives the forecast and decides what to order
We stop and tell you if
- History is shorter than two seasonal cycles plus 12 test periods
- You want a promise about future accuracy
- You want the system to place orders
- Stock-outs hid most of the demand and the dates are unknown
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 forecast is read as a promise. | The report states limits in plain words and promises no accuracy. If it does not beat the baseline it says so. |
| The method looks good because it saw the test periods. | Errors are computed only on periods not used for fitting. The script shows the split. |
| Stock-outs make low sales look like low demand. | Stock-out periods are flagged and listed, and the report says where they weaken the result. |
A separate automated reviewer checks the backtest, the baselines and the limits note before hand-over. A named person on your side receives the forecast, checks the backtest 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 series, the horizon, the baseline methods and the margin to beat
- Receive the history, promotion dates and stock-out dates
- Fit the methods using only periods before each test point
- Score every method on periods it never saw, and compare with the baselines
- Independent review of the backtest and the limits note
- Hand over the forecast, backtest and script; 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?
Want the forecast re-scored each quarter? Ask about a standing re-check.
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
- Inventory tools such as Prediko forecast demand for a monthly fee. Compare their forecast with the seasonal-naive baseline on your own last year before paying. www.prediko.io
- Forecasting: Principles and Practice describes the baselines and the rolling-origin test used here, free to read. otexts.com
- If you already have a forecast and want it scored against a simple baseline, use our sales forecast back-test job. This job produces a new forecast. For cash, see our cash-flow back-test job; for reorder points on a store, see our reorder-sheet job.
Questions
Is this AI forecasting?
It is a statistical forecast checked against simple baselines. A model, if used at all, only helps write the summary.
What if your forecast is no better than last year?
Then the report says so, you pay the same, and you get the baseline forecast.
Will it be right?
We cannot promise that. We show how it would have done on periods it never saw.
Send an enquiry
Send us
- What is forecast (orders, units, revenue) and the horizon
- How many years of history there are, and at what step (daily, weekly, monthly)
- No real records, personal data or credentials in the first enquiry
Later, once you agree
- The history as a file by an agreed secure route
- Dates of promotions, holidays and stock-outs
- A named person who receives the forecast
- A written cap on model usage
You keep your data and every ordering decision. We work on a history file you send by an agreed secure route, return a forecast, a backtest and a script, and delete our copies when the job ends. We connect to nothing.
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 “forecast-beats-naive-baseline-or-says-so” as the subject.