Know what a forecast can do before you rely on it
The aim is a model you can evaluate, understand and use in an agreed planning process. Each stage should produce evidence for deciding whether the next stage is worthwhile.
1 Define the decision
We begin with what you need to plan, who will use the forecast and when they need it. Together, we define the forecast horizon, the level of detail and what a useful result would look like.
The cost of being too high may differ from the cost of being too low. That belongs in the brief from the beginning.
Stage output: A focused problem statement and proposed evaluation criteria.
2 Review the data
We examine the history available, gaps and changes in how information was recorded. We also discuss events that numbers alone may miss, such as a promotion, supplier disruption or a period when products were unavailable.
Some questions can be investigated with existing exports. Others need better records or a narrower scope first.
Stage output: A data-readiness assessment, known limitations and a recommendation on the next step.
3 Develop and compare approaches
We establish straightforward reference forecasts, such as recent averages or the equivalent seasonal period, and compare suitable alternatives against them. More sophisticated methods are considered where the data and decision justify them.
Stage output: A prototype and a clear comparison of the approaches tested.
4 Test against past outcomes
Where the history allows, we recreate earlier forecasting points using only information that would have been available then. We compare the predictions with what happened, over a horizon that matches your planning needs.
This backtesting helps reveal where an approach performs well and where it struggles. Strong historical performance cannot guarantee future accuracy.
Stage output: A validation report covering performance, important weaknesses and uncertainty.
5 Agree how the forecast will be used
A useful prototype needs a workable home. That may be an export for your existing spreadsheet, a reporting view or a separately scoped integration.
We agree who refreshes the data, who reviews unusual results, and which decisions remain subject to human judgment. The delivery format, ownership, documentation and support arrangements are defined in the project scope.
Stage output: An agreed handover or deployment plan.
6 Review performance as conditions change
If ongoing support is included, the scope should set out when results are reviewed, how forecast errors are tracked and when a model or assumption needs attention.
Changing products, customers or operating conditions can affect performance. A forecast needs a review process appropriate to the decision it supports.
Stage output: A documented monitoring and maintenance arrangement, where commissioned.
A range can be more useful than a single number
Where the method and data support it, a forecast can include a prediction interval: a range intended to express uncertainty about future outcomes. Scenario analysis answers a different question by showing what happens under chosen assumptions. We explain which kind of range you are looking at and what it means.
Questions you may have
How much historical data do we need?
It depends on the forecast frequency, horizon, seasonal patterns and how much the business has changed. The initial review establishes whether the available history is sufficient for the question you want to answer.
Will a more advanced model always be better?
No. A simpler approach can perform well and be easier to maintain. Model selection should be supported by testing and the practical needs of the business.
Can we keep using our existing spreadsheets?
An export that fits your current planning process may be a sensible starting point. Any automation or integration is agreed separately after assessing your systems.
What if the prototype does not improve on our current approach?
The comparison still helps you decide where to invest. The recommendation may be to keep the baseline, improve the data, change the forecasting question or stop further development.