This original design draws on Chronos-2 time-series materials and has not undergone client backtesting. General model capabilities do not establish accurate demand forecasts for particular products.
Check historical data first
Record products, dates, sales, stockouts and promotions, identifying gaps or backfilled entries. Otherwise, low sales may be mistaken for weak demand. Client-appointed staff confirm context; the model must not fabricate sales records.
Make comparisons reproducible at historical cutoffs
A trial could forecast selected products using only information available at the time, then compare actuals. Retain a simple baseline to assess additional information. Use fictional acquisition samples and do not describe demo charts as client gains.
Keep decisions with the operator
Replenishment also depends on payment terms, purchase quantities and storage. Reports provide supporting observations. Price cleaning and review, excluding automatic orders. Discuss ongoing work only after the client uses results in regular meetings.
Further reading: business projects.