AI forecasting readiness is whether the decision, the history, the test, the owner and the review behind a forecast are in place before anyone relies on it. This checklist covers forecasting only. Instead of a general AI readiness check, it scores one use, and wider questions of strategy or infrastructure sit outside it. In short, think of it as an AI data readiness check for forecasting. It has 15 checks in five groups, each with what to bring and when it counts as settled, plus a way to score the result. It sits under the Signisys trend forecasting use case.
Last reviewed 2026-11-02 by Manish Sharma.
AI Forecasting Readiness Checklist: What the 15 Checks Test
The checklist is for a finance, sales, operations or IT leader who plans to rely on a forecast, or already does. It is a self-check, then, not an audit. In Signisys’ working method, the 15 checks fall into five groups, and each group answers one question:
| Group | Question it answers | Checks |
|---|---|---|
| Decision | Is the decision written down before the forecast starts? | 1 to 3 |
| History | Is the history ready to use? | 4 to 6 |
| Baseline and test | Is there a baseline and a test? | 7 to 9 |
| Ownership | Who owns the forecast and its inputs? | 10 to 12 |
| Review | Will the forecast be reviewed? | 13 to 15 |
IBM’s explanation of AI forecasting sets out four difficulties: the quality of the data, shifting conditions, transparency with trust, and the processes that run the system. Our reading relates them to the groups as follows. Data quality sits with the history checks, and shifting conditions sit with the input and review checks. Transparency sits in check 12, and the running processes sit with ownership and review. The groups and their tests are ours, not IBM’s and not a standard.
Marks for AI Forecasting Readiness: Settled, Under Way, Not Started
Each check takes one of three marks. In Signisys’ working method, they mean the following:
- Settled (2): the test is met, and a dated record shows it.
- Under way (1): the test holds for some items or periods, or the record carries no date.
- Not started (0): the test is not met.
A belief that something is in place does not make it Settled. Until a dated record exists, the highest mark is Under way. Mark a check N/A only when it has no bearing on the forecast, take it out of the maximum and write down why. Score a sample that you choose and name, for example the one forecast behind the monthly purchasing plan, and do not assume the other forecasts match it. When no check is N/A, the maximum is 30. However, the checklist sets no target score or band.
For example, take check 9. Settled: the team has written the gap and its direction for the baseline and for the model, with a date. Under way: the baseline gaps are in, but the model rows are blank or undated. Not started: nobody has drawn up a test table yet.
Key Takeaway A check is settled when the record shows what was done and when it was done.
Is the Decision Written Down Before the Forecast Starts?
These three checks ask whether anyone has said what the forecast is for. Otherwise, the later groups measure work that may serve no decision.
| Check | Bring | Settled when |
|---|---|---|
| 1. The team has written down the decision the forecast serves and the person who acts on it | One line per forecast | Someone outside the forecasting team can say what changes in the business if the number moves |
| 2. The team has written down both the forecast period and the decision date | The two dates, and the time the decision takes to come into effect | The period reaches past the decision date by at least that time |
| 3. The team has picked what the forecast counts: all sales, one product line or one item | The level, and where its history sits | The history exists at that same level |
Write the lead time for check 2 down yourself; this checklist sets no figure for it. If two people name different decisions for check 1, mark it Under way until they agree on one line.
Is the History Ready to Use?
Among the predictive analytics requirements behind a forecast, the history comes first. Specifically, this group stays with coverage, events and the meaning of one figure.
| Check | Bring | Settled when |
|---|---|---|
| 4. The team has written down the first and last periods it holds for each item, with any gaps | A coverage line per item | The line exists and each gap has a written reason |
| 5. Someone has marked the one-off events in the history | The history sheet | Each large single order, closure, promotion or stock-out has a note saying whether the period stays, changes or drops out |
| 6. The demand figure means one thing to all who use it | A sheet with one sentence and one owner per term, such as whether demand counts at order date or ship date | Two owners, asked separately, give the same sentence |
Note also that this checklist sets no minimum length of history. For cleaning wrong or duplicate values, see the use case on improving data quality. Connecting the systems that the figures come from is a separate subject, and this group does not score it.
Is There a Baseline and a Test?
These checks score whether anyone has tried the forecast on past periods before relying on it.
| Check | Bring | Settled when |
|---|---|---|
| 7. A baseline forecast exists, from a simple rule anyone on the team could repeat | The baseline for each test period | Each test period has a baseline figure beside it |
| 8. The team chose the test periods before any model ran and kept them out of training | The periods and the rule for choosing them | Both are in writing, with a date earlier than the first model run |
| 9. The team has recorded the gap and its direction for the baseline and for any model | The test table: period, baseline, model, actual figure, gap | Both sets of gaps appear in the table |
For instance, IBM describes judging a forecast against outcomes already known, including whether a model leans high or low, and testing it on earlier periods first, which it calls backtesting (IBM). If no model exists yet, score check 9 on the baseline alone and note that on the sheet. The six steps in a guide to forecasting demand with a model include a test on past periods.
Checks 7 to 9 still apply. Score the baseline and the test rule now, then add the model’s row when a model exists.
Who Owns the Forecast and Its Inputs?
Ownership checks ask whether a named person can find, explain and update the forecast.
| Check | Bring | Settled when |
|---|---|---|
| 10. The firm has named an owner and a deputy for the forecast | Two names | Each of them can say where the last forecast is kept |
| 11. The team has listed the inputs beyond history, such as planned promotions or the calendar, with a source and a refresh date | The input list | Each input has an owner and a date |
| 12. A person other than the builder can explain one forecast figure from its inputs | That explanation in writing | The explanation matches the input list |
IBM lists transparency and trust as a challenge and describes it as seeing what shaped the figures and being able to question them. Check 12 is where this checklist tests that point. If only the builder can explain the figure, the firm depends on one person. Similarly, if the deputy in check 10 has not yet opened the forecast, mark it Under way.
Will the Forecast Be Reviewed?
The last group checks that someone holds each forecast up against the figures that followed it.
| Check | Bring | Settled when |
|---|---|---|
| 13. The team publishes the forecast as a low, a middle and a high figure, with the date of the next one | The one-page sheet | All three figures and the date appear on it |
| 14. Someone keeps a log of each forecast, its actual figure and the gap | The log | The last closed period appears in it |
| 15. A written rule decides what follows each review: keep, change the inputs or go back to the history | The last review decision and its date | That decision is on record |
For example, the log can sit beside the regular management reports. Where a firm builds those reports automatically, see the use case on automating management reporting. Keep this group to the forecast itself, while other reports belong elsewhere.
AI Forecasting Readiness: Which Group to Fix First
Add the marks within each group, and write the number of Not started checks beside each one. In Signisys’ working method, the decision and history groups come first. The baseline, ownership and review checks rest on them. A test run at the wrong level, or on unmarked history, tells the team little.
Within any group, close the Not started checks before improving an Under way one. The checklist gives no bands, no target and no comparison with other firms. Write the date and the sample beside the score, and then score again after each change. Read that way, an AI forecasting readiness score is a record of where the gaps sit, not a grade.
Key Takeaway Score the same named sample each time, so that a later score describes the same forecast.
Where Your Own Judgement Takes Over
This is a self-check. It scores whether records exist, not whether a forecast will be right, and it covers forecasting inputs and records only. Likewise, connecting systems, choosing between analytical approaches, model selection, infrastructure and security sit outside it. Nor is it an audit or legal advice. For why forecasts go wrong, see the four reasons a business forecast misses.
A second view helps when two owners score the same check differently and neither can show the record, or when the test table in checks 8 and 9 is still blank.
Signisys provides data and AI services through its business intelligence and analytics solution and its data and AI service page. This box describes a Signisys service, not neutral advice, and it makes no promise.
AI Forecasting Readiness FAQ
What does AI forecasting readiness cover?
Can an AI forecasting readiness score be full and a forecast still miss?
Which group should we fix first?
Do we need a model before we score?
Why does this checklist cover forecasting only?
Security Insights, Delivered.
Practical, vendor-neutral analysis of threats, tools, and architecture decisions.