KEY TAKEAWAY
Check stockouts and availability before equating sales with demand. Better forecast error and better inventory decisions are related but distinct evaluation questions.
1. Understand zero sales
Zero sales can mean no demand, a stockout, a pre-launch period or a closed store. Treating these alike can understate demand. Review stock and availability alongside daily sales.
If inventory history is incomplete, mark the gaps. Keep estimated values distinct from observed data and consider narrowing the evaluation period or products.
2. Align products and returns
Variants, bundles and retired codes affect historical continuity. Match products to actual buying units rather than similar display names. Preserve both sale and return dates to explain adjustments.
When combining stores and online channels, distinguish orders, lines and units.
3. Use information known at the time
Realized ad spend or discounts decided later can leak future information into a historical test. Reconstruct the information available at the forecast date.
One campaign does not establish reliable behaviour for future promotions. Review regular periods, promotions, new items and continuing products separately.
4. Compare with simple baselines
Retain seasonal, moving-average or current buyer forecasts as baselines. Compare by category and sales volume so high-volume items do not conceal poor results elsewhere.
Low-volume items can produce unstable percentage errors. Consider category-level forecasts and replenishment rules instead of valuing model complexity for its own sake.
5. Include the ordering constraints
A better forecast may still be unusable with large pack sizes, long lead times or limited storage. Compare current and proposed orders under the same constraints.
Stockout and excess-stock cost assumptions influence decisions. Label assumptions and record buyer adjustments to separate forecasting issues from operating constraints.
6. Scope the first engagement
Start with one category, a defined history and a buying cadence. Request data definitions, baselines, evaluation, a review interface and exclusions as deliverables.
For f can help with commerce and inventory reconciliation through forecasting validation. Availability records, product mappings and current buying methods define what can be tested.
A concrete acceptance check
Reconcile against existing reports at the same cutoff and explain differences at line-item level. Agree how later returns affect figures before extending into purchasing or forecasting.
Before commissioning
- Distinguish stockout, closure and retirement.
- Prevent future campaign information from entering evaluation.
- Compare with simple baselines.
- Evaluate lead time, pack size and buyer adjustments.
Explore support for this industry
Retail and e-commerce — data and AI overview ↗