Demand Forecasting (LT calendar)
Forecasts with honest prediction intervals and scenario planning - cut inventory cost without stockouts.
SKU 4212 · PIENAS 2.5% 1L · STORE 4 - 14-DAY FORECAST
Order proposal
96 units
covers P90 through Friday at your 95% service target
drivers: trend +4% · Fri peak · promo widens the interval - honestly
How it works
Connect ERP / WMS, read-only
Sales at SKU-store-day level, promotions, stock, lead times. Hierarchical: SKU → category → store → total, reconciled so the numbers add up.
Forecast as a range, not a guess
P10/P50/P90 for the next 14 days, with the Lithuanian calendar built in - holidays, paydays, and that first hot week of summer.
Safety stock from the math
Order proposals sized to your service-level target per category - perishables and slow-movers get different treatment, automatically.
Coverage is watched, honestly
If the 90% interval stops containing ~90% of outcomes, the model recalibrates and tells you - a forecast that can say "I'm less sure this week" is one you can plan on.
Methodology reference
The published method this build is based on, and what it did in production.
Amazon cut forecast error ~15% with DeepAR by predicting a full probability distribution instead of a single number.
Probabilistic forecasting · as reported by arXiv ↗
Referenced for methodology only. AimRank is not affiliated with, endorsed by, or a customer of these companies; each links to its own public source. Sources last checked September 2026.
The Four Guarantees™ - this build
The scaffold wires all four in. The PoC proves them on your data, as of that day. Only Implementation keeps them true over time: watched, drift-checked, with recourse.
Measured value
~30% less overstock writedown + ~25% fewer stockouts + working capital freed. Pinball-loss + interval-coverage eval gate vs seasonal-naive.
Defensible
Driver decomposition per forecast - trend, seasonality, promo, holiday. Auditable assumptions; every scenario logged.
Self-improving
Interval coverage (PICP) tracked live; recalibration triggers on coverage breach; demand-pattern drift alerts.
Yours & everywhere
ERP read + write-back of proposals; your cloud, full source. MCP endpoint for procurement agents.
The number, sized honestly
Reference buyer: Lithuanian / Baltic retailer or distributor, ~€10M revenue, ~500 SKUs, ~5% stockout rate, ~€100k/yr overstock writedowns.
Three ways to own it
| Tier | What you get | Price | |
|---|---|---|---|
| Scaffolding ★ | The full repo - models + conformal intervals + pinball/PICP eval harness, audit log, MCP server. Reference run on public retail data. | €1,990 | |
| PoC ★★RECOMMENDED | Trained on your sales history - forecast accuracy and interval coverage backtested against your actual demand, LT calendar included. Code + quality report yours. No guarantee before the backtest speaks. | from€6,000 | |
| Implementation ★★★ | Production: live forecasts in your planning flow, coverage monitoring, re-forecast cadence, retraining, monthly accuracy-vs-naive report - the agreed number (conservative) attaches here. | from €18,000+ from €900/month assurance |
★ = engagement depth. PoC is the recommended path: quality proven on your data before production money. The PoC carries no performance guarantee by design; the agreed number (conservative) attaches at Implementation, informed by the PoC report.
What we don't promise
Ready to see your own number?
Request the build: within 48h you get a personal reply with the value sized to your volume.
No commitment · reply within 48h · your data stays in the EU