Demand forecasting software that works
at SKU level
SKU Science forecasts every SKU individually — and at any level of aggregation you plan on. The AI engine tests hundreds of combinations per item, every cycle, and keeps the one that holds up over time. Forecast accuracy KPIs, SKU dashboards and the S&OP process around them come with it.
| Part number | Client | ABC | ABC % | XYZ | XYZ % | Var (%) |
|---|---|---|---|---|---|---|
| 00-3 | Customer 3 | A | 24 % | Y | 22 % | 337 |
| 0014-1 | Customer 1 | C | 7 % | Y | 25 % | 21 |
| 00-2 | Customer 1 | C | 2 % | Z | 32 % | 18 |
| 005-1 | Customer 1 | C | 1 % | Z | 31 % | 257 |
| 022-1 | Customer 2 | C | 2 % | Y | 24 % | 5 |
| Total | ||||||
| 6/2026 | 7/2026 | 8/2026 | 9/2026 |
|---|---|---|---|
| 0 | 0 | 0 | 0 |
| 296 | 308 | 234 | 241 |
| 51 | 800 | 52 | 35 |
| 181 | 0 | 0 | 0 |
| 64 | 90 | 100 | 78 |
| 2,532 | 2,272 | 1,530 | 1,553 |
SKU forecasting: hundreds of AI combinations tested on every item, every cycle
Statistical models and machine learning are combined into hundreds of candidates and backtested on your own history. The engine keeps the model that stays accurate on future periods, not just the past. No model to pick, no parameter to tune.
-
A combination per item, re-selected every cycle
Statistical models and machine-learning models are fitted item by item. When an item's behaviour changes, the selected combination changes with it: nobody has to notice and intervene. Each forecast is compared with the previous cycle, so the items that moved most rise to the top.
-
Aggregated SKU forecasting, reconciled
Forecast by item, family, customer, channel, warehouse or country. Change the aggregation level and the engine recomputes, the SKU detail stays consistent with the aggregate.
-
Hard cases handled explicitly
Promotional uplifts are modelled rather than absorbed into the baseline. Launches borrow from comparable products, successors inherit the history of the item they replace, and end-of-life items ramp down without a manual override every cycle.
-
Editable at any level, fully traceable
Adjust the forecast at item or any aggregation level, in whatever unit the decision needs — quantity, price, cost, margin, or a multiplier field such as weight or pallets. Edits reconcile across levels, and an audit trail logs every change.
| Period | 4/2026 | 5/2026 | 6/2026 | 7/2026 | 8/2026 | 9/2026 | 10/2026 | 11/2026 |
|---|---|---|---|---|---|---|---|---|
| Demand | 164 | |||||||
| Demand adjustments | ||||||||
| Forecast baseline | 183 | 289 | 296 | 237 | 234 | 241 | 231 | 233 |
| Forecast adjustments% | 308 | |||||||
| Revenue (USD) | 169 248 | 298 248 | 305 472 | 317 856 | 241 488 | 248 712 | 238 392 | 240 456 |
Forecast accuracy KPIs, measured every cycle, not once a year
Accuracy, bias, error and forecast value added are computed automatically every cycle — at item level or any aggregate. Each can be weighted by revenue, so a miss on a high-value item counts more than one on a long-tail SKU.
-
Accuracy by forecast lag
Error is measured at the lag that matters for planning (the forecast made one, two or three cycles before the period it covers) and averaged over the window you choose. A forecast that is reliable at a short lag but drifts further out shows as two different numbers.
-
Bias, tracked with a target band
Systematic over- or under-forecasting is the expensive error. Bias is reported next to accuracy at every level so you can see which items, families or customers drift.
-
ABC / XYZ, so effort goes where it pays
Items are classified by importance (A/B/C) and by predictability (X/Y/Z), ranked on revenue rather than units. Planners spend their review time on the AX and AZ items that move the business, not on a flat list.
-
Forecast value added on every edit
Value added is measured against the best forecast the engine can produce, not a naive baseline. When someone overrides a value, SKU Science shows whether the change improved the forecast or made it worse.
| Part number | ABC | XYZ | Var % | Avg acc | Accuracy (%) · by period | ||
|---|---|---|---|---|---|---|---|
| 2/26 | 3/26 | 4/26 | |||||
| 005-1 | C | Z | 7 | 69 | 88 | 25 | 93 |
| 010-1 | A | Z | 6 | 83 | 69 | 97 | 84 |
| 033-2 | B | Y | 62 | 89 | 93 | 93 | 81 |
| 044-1 | C | Z | 8 | 83 | 92 | 85 | 71 |
| 022-1 | C | Y | 5 | 89 | 98 | 77 | 93 |
| 0011-3 | A | Y | 9 | 80 | 64 | 81 | 94 |
| Total | 73 | 82 | 82 | 86 | |||
Business review dashboards, ready the moment your forecast runs
Because the platform already holds your historical sales and computes your forecasts, business review dashboards are there without any data prep. Read the trend, spot the gap against plan and take corrective action.
-
Performance against budget and last year
For each period, compare actuals against the fiscal-year budget and last year (FY-1) on one chart, so you see where you stand versus plan without building a report.
-
Any aggregation, updated dynamically
Aggregate your data in any format (item, product family, territory, customer) and the dashboard recomputes on the fly for an easy operational performance review at the level you plan.
-
Backlog against forecast and budget
Visualize the orders already booked against the remaining forecast and the budget line to quickly understand what is left to close for a successful year end.
-
Switch the view, save it, export it
Move between budget comparison, order backlog and cumulative trend, in units or value, then save the view or export it — no pivot tables, no BI project.
| PART NUMBER | LOCATION | |
|---|---|---|
| 00-3 | Paris | |
| 0014-1 | Milano | |
| 0011-3 | Milano | |
| 022-1 | Paris | |
| 005-1 | Paris |
Not just forecasting, the whole demand planning cycle
A forecast on its own is a number in a file. SKU Science ships the process around it: granularity, temporality, metrics, review and hand-off are already defined, so your S&OP cycle runs without being re-invented each month.
Sales history in
Excel or CSV upload with column mapping remembered between imports, or an API and MCP connector for scheduled loads from your ERP or data warehouse.
AI baseline forecast
The engine produces the unbiased baseline forecast per SKU, at the aggregation level you plan on, with the forecast error from previous cycles so you know how much to trust it.
Demand collaboration
Sales, key accounts and finance each have a number. They are held side by side with the baseline, and the consensus is scored against it, so you know whether it beats the engine.
Monthly cycle, locked and comparable
Cycles are dated, locked and archived. Each one can be compared with the last: what changed in the forecast, on which items, and because of whom, the question the S&OP meeting always asks.
Promotions and lifecycle
A planned promotional uplift stops looking like new baseline demand, and the post-promo dip stops looking like a lost customer. Launches, phase-outs and substitutions are linked, not re-keyed.
Plan out to execution
The agreed plan leaves as an export or through the API, at SKU and location level, ready for the ERP, the replenishment tool or the production schedule.
First forecast in 2 days, then a guided path to go-live
Start on your own data the same week. Our team runs the set-up alongside you, and stays with you from the first cycle to full go-live.
Send your sales history
Upload Excel or CSV, or push data via MCP. Bring at least 24 months of history at SKU level so seasonality can be detected. Columns are mapped once — date, item, quantity, and whichever dimensions you plan on.
Set-up done with you
Aggregation levels, ABC/XYZ, computation granularity and user access are configured together in one working session. This is guided onboarding, not a statement of work.
First forecast in hand
The engine has run on your own data. You review the SKU-level forecast, adjust what needs adjusting, and lock a first cycle you can take into a meeting.
Support and configuration help, included
Demand planning consultants answer the questions, not a generic help desk. Support is included, not a separate line item.
- Guided onboarding on your own data
- Help with configuration and re-configuration as your portfolio changes
- Token-gated access: we only enter your account when you let us in
Demand Forecasting Best Practices training
This platform agnostic training will help you shape the ideal demand planning & forecasting process: the framework first, the tool second. Useful whether your team is new to forecasting or rebuilding an S&OP process that has drifted.
- Process design: granularity, horizon, cycle, roles
- Understanding and reading forecast accuracy KPIs
- Running the demand review and the S&OP consensus
Your sales data, handled to enterprise standard
SOC 2 Type II certified, audited annually by an independent third party, and GDPR compliant. Hosted on AWS, SSL encryption end to end, backups every 6 hours, token-gated support access — only you let us in.
2
SKU-level forecasting, answered
Top-down or bottom-up — SKU Science forecasts at SKU level and keeps every aggregation reconciled. Plus the questions planners ask most.
You choose the direction per portfolio or per client. SKU Science runs both and keeps the aggregated view and the SKU detail reconciled automatically. Either direction feeds the same S&OP: the forecast still drives production, purchasing and replenishment.
What is SKU-level forecasting?
SKU-level forecasting is the practice of predicting future demand for each individual stock-keeping unit
(each product, pack size and variant) rather than for a product family, category or total volume. Each SKU is modelled on its own history, with its own seasonality and its own accuracy measurement, so replenishment, production and purchasing plans can be built directly from the forecast.
Which analytics software tracks SKU-level performance?
Demand planning platforms track SKU-level performance natively, because they hold both the forecast archives and the actuals for every item. SKU Science ships an SKU performance tracker as standard: forecast accuracy and bias per SKU and per period, ABC/XYZ classification, movers between cycles, and portfolio-level dashboards that rebuild at any aggregation level. General-purpose BI tools can display the same figures, but the accuracy and bias calculations have to be built and maintained by hand.
How is SKU demand forecasting different from aggregate forecasting?
Aggregate forecasting predicts total demand for a family, brand or region. It is more stable, because individual errors cancel each other out, and it is the right level for budget and capacity planning. SKU demand forecasting predicts each item, which is harder but executable (orders are placed on SKUs, not on categories). SKU Science computes both and keeps them reconciled, so the aggregated view discussed in the S&OP meeting and the SKU detail sent to the ERP never diverge. You can forecast top-down or bottom-up, whichever direction reads the demand best, and both levels stay reconciled either way.
Why is SKU-level demand forecasting harder?
At SKU level the demand signal is thinner: volumes are smaller, series are shorter, promotions distort history, new items have no past and slow movers sell intermittently. A single model applied to a whole portfolio fits some items well and the rest badly, which is why the work is done by software that fits and re-selects a model per item, cycle after cycle.
How much sales history do you need to forecast a SKU?
Bring at least 24 months of history so seasonality can be detected, the engine works with less, but with less confidence in the seasonal pattern. Items with little or no history are handled separately: launches borrow from comparable products and a successor inherits the history of the item it replaces, rather than being forced through a model that has nothing to learn from.
How long does implementation take?
Two days from first upload to a reviewed forecast on your own data, with our team running the set-up session with you. There is no integration project to schedule: the first cycle runs on a file extract, and API or MCP automation can be added later without changing anything you have configured.
See SKU-level forecasting run on your own data
Thirty minutes with a demand planning specialist: your portfolio, your aggregation levels, your accuracy KPIs.
