Fundamentals
Hierarchical forecasting: make the pieces add up
You rarely forecast one series. You forecast stores that must sum to a region, and SKUs that must sum to a brand. Hierarchical forecasting predicts at more than one grain and reconciles so finance and ops are not holding two official numbers.
Updated Aug 26, 2026·8 min read
Why one grain is never enough
Finance wants a chain week. Ops wants a store day. Brand wants a category. If you forecast only the top, you invent a shape when you split. If you forecast only the leaves, the total will not match the number the CFO already put in a slide. Hierarchical forecasting is the boring work of predicting more than one grain and making the pieces add up.
One tree, one meeting
Energy does this with hour → day → region. Hospitals do it with department → site → system. Schools do it with course → campus. The noun changes. The treaty does not: one tree, one meeting, no private forecasts that cannot reconcile.
Bottom-up, top-down, reconcile
Bottom-up sums the leaves. Honest when leaves are stable; noisy when they are not. Top-down forecasts the total and spreads with historical shares. Calm totals, fictional stores. Reconciliation (MinT and friends) forecasts several levels, then projects onto numbers that add up. You pay for a bit of math. You stop arguing about which export is official.
- Forecast the grains people actually buy, staff, or book against — not a catalog tree nobody uses.
- Keep a naive at each grain. A pretty reconciliation that loses to last week is still a loss.
- If a leaf is mostly zeros, do not make it the foundation of the chain number.
The tree has to match the meeting
A product hierarchy from PIM is not an S&OP hierarchy. A legal entity tree is not a dispatch tree. If you reconcile on the wrong parent, you will produce a coherent fiction. Draw the tree on a whiteboard with the people who spend money. Then put signals at those grains.
How Predict.ai holds the grains
Open a goal per grain you actually decide on. Predict.ai will score each target on its own folds — a region-day WAPE is not a chain-week WAPE — and keep each path live. Roll-ups stay in the workspace as their own signals, not a second spreadsheet that diverges on Friday. Driver discovery can differ by grain: weather at the site, a national campaign at the brand.
If you need the pieces to add up in a planning tool, export the pinned paths. Do not pretend one model stretched across every leaf is “the hierarchy product.” It is a shortcut that will show up in bias at the store that never looked like the chain.
FAQ
- What is hierarchical forecasting?
- Forecasting a family of series that must add up: SKU to category to chain, hour to day, site to region. After the models run, reconciliation projects the set onto numbers that are coherent.
- Should I always forecast the lowest grain?
- No. Leaves are noisy. Forecast the grains people buy against, then reconcile. A pretty product tree ops does not use will reconcile into a fiction.
Keep going
Guide
What is good forecast accuracy?
Good is beating a naive baseline on the same grain, horizon, and folds — by enough to change a decision. Industry round numbers are gossip until they match your SKU mix and your clock.
Guide
Forecasting new products and locations
A cold start is a forecast with almost no history of its own. You borrow shape from similar items or places, use leading signals, keep the horizon short, and replace the borrowed prior as soon as real sales show up.
Guide
FP&A forecasting
Finance needs a slower, cleaner number than a warehouse. The mistake is either ignoring the operational forecast or forcing it into a 12-month line that cannot be true. FP&A should consume a reconciled path, plus scenarios, plus an interval.
Use case
Retail demand forecasting
Anticipate demand by product, store, channel, and region before buying or allocation decisions are locked.
Use case
Revenue forecasting
Maintain a current revenue outlook grounded in bookings, pipeline, usage, renewals, pricing, and seasonality.