Practical AI · 3-minute read
The forecast says next month’s demand will rise 12.73 percent. Nobody knows why the customer’s expansion was delayed, but the spreadsheet has apparently made peace with uncertainty.
A forecast can help a business prepare. Extra decimal places do not provide extra evidence. Before asking AI what will happen, ask what your plan depends on.
Start with the decision you need to make
Orchard Bend Supply is a fictional agricultural supplier planning its packing-material deliveries. The manager needs to decide whether to arrange additional delivery capacity. The team has past orders, current bookings and several customers who have said ‘probably.’ Those are three different levels of evidence.
Create a short assumptions register. Record each input, its source, when it was checked, who owns it and what remains uncertain. Separate confirmed orders from tentative requests. Explain unusual periods in the history, such as a closure or a one-off large order.
A chatbot can help organize that material and question the assumptions. It should not fabricate missing sales history, current weather information or customer commitments. Use a suitable forecasting method and validated calculations when numerical predictions are required.
Build scenarios that change something useful
For a simple planning exercise, Orchard Bend tests three invented demand cases: 800, 1,000 and 1,300 units. Staff check delivery capacity and lead times under each. These are chosen scenarios, not a statistical prediction interval and not assigned probabilities.
Forecasting: Principles and Practice describes scenario forecasting as exploring possible futures based on different assumptions. It emphasizes plausible scenarios and preparation for what could happen, rather than treating one imagined future as certain. [1]
Help build a scenario worksheet for [operational decision] using only these approved inputs. Separate confirmed facts, assumptions and missing information. For each scenario, show changed assumptions, operational consequences, capacity limits and a signal that would trigger reconsideration. Do not invent probabilities or label the scenarios a confidence interval. Identify calculations and source records a person must verify.
Check the assumptions with the people who know
The operations lead checks capacity and arithmetic. Sales confirms the status of customer requests. Purchasing verifies supplier lead times. Keep the source, assumptions and method visible so another person can challenge the result.
Try changing one important assumption at a time. If a modest shift changes the decision, that input deserves attention. Consider what happens to smaller customers or remote delivery routes when capacity is tight, rather than quietly treating them as disposable.
Agree on a review trigger, such as a major booking change or a confirmed supplier delay. Then compare the assumptions with what actually happens and revise the process. A useful forecast helps you respond; it does not excuse you from noticing reality.
Your turn: Which assumption would change your next staffing or inventory decision if it turned out to be wrong?
Source [1]: Hyndman and Athanasopoulos: Forecasting Principles and Practice, Scenario forecasting. Supports scenario forecasting and explicit assumptions. The supplier, quantities, worksheet and operational example are original illustrations, not a forecast.
