Fundamentals
Choosing a forecast horizon (and the grain that goes with it)
The horizon is how far ahead the forecast looks; the grain is how finely it ticks. Pick both from the decision they serve — lead time, planning cycle, and how fast the world can change — not from what the model can print.
Updated Aug 10, 2026·8 min read
Horizon vs grain
Horizon is how far you look. Grain is how finely you tick. A 14-day horizon at daily grain is fourteen numbers. The same horizon at weekly grain is two. They are different products. Libraries will happily print both. Your lead time will not.
History · origin · horizon
Start from lead time
The useful horizon is the slowest delay in the decision. Overnight replenishment does not need a 16-week path. A container that left Shenzhen last Thursday does. Staffing tomorrow’s clinic is a 24–48 hour problem. Hiring for next semester is not.
Teams try to “just do 12 months” so one model feeds everyone. What they get is a chart that is precise about a week nobody can act on, and vague about the week they can. Split the products. Share the data.
When the horizon is too long
Error compounds. Each step sits on the last. By month ten you are often forecasting a forecast. Direction can still be useful — “we will need more capacity in H2” — but do not send that number to a daily replenishment engine and then blame the algorithm for the shrink.
Long-range work wants wide intervals and a human in the loop. If a vendor shows a skinny band at week 40, they have not told the model it is allowed to be unsure.
When the grain is too fine
Hourly SKU-store demand is a legitimate series. It is also mostly zeros. You will spend the project explaining intermittent demand to people who wanted a dashboard. Sometimes the move is to forecast daily or weekly and disaggregate with a simple profile only where ops truly needs the hour.
Nested horizons
Grown-up stacks keep more than one. A nowcast for this afternoon, a short forecast for the week, a slower one for the buy. They should not contradict each other by accident — that is a hierarchy problem — but they also should not be the same model stretched like taffy.
FAQ
- What is a forecast horizon?
- It is the stretch of future the model is asked to fill in. A 14-day horizon produces 14 daily points (or 2 weekly ones). It is a product choice, not a model setting you twist for a prettier chart.
- Should I forecast daily or weekly?
- Daily if the operation moves daily — staffing, perishable replenishment, energy dispatch. Weekly if buying and reporting already live on weeks. Forecasting daily and summing up is fine; forecasting weekly and splitting into days is a fiction.
- Why do long-range forecasts look confident and still miss?
- Error compounds. Each step is built on the last. By month ten you are often forecasting a forecast. Use a long horizon for direction, a short one for the POs you will actually send.
Keep going
Guide
How much history do you need to forecast?
Enough to see the pattern you are betting on. A weekly seasonal business wants more than one year; an hourly series can learn a week of shape faster. Quality of the target beats another dusty decade of the wrong grain.
Guide
Prediction intervals
A prediction interval is a range the future value is expected to fall in, not a promise. p50 is the middle path; p10–p90 is a band for planning stock, staff, and cash without pretending the future is a single line.
Guide
What is demand forecasting?
Demand forecasting estimates how much of a product or service people will want in a coming period, so you can buy, staff, and price before the rush — not after it has already walked in.
Use case
Shipment volume forecasting
Forecast inbound and outbound volume by lane, facility, customer, and mode before capacity is committed.
Use case
Production output forecasting
Forecast finished output from schedule, staffing, material availability, cycle time, downtime, and yield.