Demand at SKU and channel level, learned not guessed.
An ML forecast that learns from every channel signal — so the buy starts from real demand, not last year plus 5%.
Demand forecast
Demand projections — WMAPE-ranked accuracy. The plan moves from here and ties to the buy.
Why the forecast decides the season.
Full-price sell-through gap between retail leaders and the industry — most of it down to how they act on the forecast.
Source: Incisiv × WRC × Anaplan 2026
Share of unplanned markdown cost attributed to upstream buying and forecasting decisions, not in-season demand.
Source: Coresight industry research
Face the Future's full-price sell-through gain after switching to a connected forecast.
Source: Audited customer outcome
What breaks in the forecast before the buy is written.
Most planners know the forecast is off by week 2. The cost lands at week 12 in markdowns no one planned for.
Three steps from data to a forecast you can plan a buy against.
Stream every channel signal
POS, e-com, basket adds, returns, supplier confirmations and cost moves — all into one model.
Forecast at SKU × channel × week, with confidence
A per-line WMAPE confidence label tells the team what's safe to commit to and what needs a second look.
Cold-start new styles from real cohorts
2–3 real comparable products from your own history — image-match scored — blended into the new forecast.
A forecast you can
plan a buy against.
ML forecasts at SKU and channel, refreshed daily — so the buy starts from real demand, not last year plus a percentage.
- Per SKU × channel × weekNot category averages — channel mix and size depth are first-class data.
- Cold-start from real cohortsNew styles baselined on 2–3 actual comparable products — image-match, channel, price band.
- Refreshed dailyPOS, e-com, basket adds, returns and supplier confirmations stream in — never stale by Monday.
Demand forecast·8-week horizon · per SKU × channel| Top SKUs · 8-wk forecast | Variant · channel | Units | WMAPE | Trend |
|---|---|---|---|---|
| Salmon Wide-Leg Trouser | Salmon · DTC | 12,480 | 7.4% | |
| Pop Tapered Trouser | Cobalt · DTC + WS | 9,840 | 13.8% | |
| Olive Cropped Wide-Leg | Olive · DTC | 7,260 | 9.1% | |
| Velvet Trouser | Black · Retail | 6,140 | 8.6% |
Meet your Demand agent
Re-forecasts the season on live sell-through, day and night — flags where the plan and the signal disagree.
Meet the agentsRe-forecast ready — 3 categories have drifted from plan this week. Want me to stage the moves for your review?
“We used to plan the season off last year plus a percentage and find out at week 8 it was wrong. Now the forecast updates as the season runs. Our team isn't reconciling spreadsheets — they're making decisions.”
Full-price sell-through (audited)
The forecast you have today, and the one Tightly delivers.
The forecast is one part of the connected plan.
Financial planning
The top-down plan reconciled to the same forecast — one number, both sides agree.
In-season management
Re-forecasts on live sell-through and proposes the rebalance before the markdown is needed.
Open-to-buy
OTB moves with the forecast, so the buy is always written against what's actually expected to sell.
How does Tightly forecast new styles with no sales history?
Each new style is baselined on 2–3 real comparable products from your own catalog — same price band, channel, materials, with an image-match score on the visual similarity. Their real trajectories are blended into the new forecast. It's the same logic a buyer uses to call out a 'this'll sell like that' analogue — done at scale and explicitly, not in someone's head.
How is this different from an ERP demand-planning module?
ERP demand-planning modules are statistical extrapolations from historical demand at category or item level. Tightly's forecast is an ML model trained on your channel signal at SKU and week, plus a cohort-based cold-start. It also feeds the financial plan and OTB directly, so there's no second model to reconcile to.
How often does the forecast update?
Daily by default — POS, e-com, basket adds, returns and supplier confirmations stream in continuously. Major movers (a viral SKU, a stockout, a supplier slip) surface in the trading view inside the trading day.
Can you forecast natively at channel level?
Yes. DTC, wholesale, marketplace and retail are first-class dimensions, not aggregations. A forecast for 'Knitwear' is the sum of channel-level forecasts, not an average smudged across them.
What's WMAPE and why is it on every line?
Weighted Mean Absolute Percentage Error — a confidence label on each forecast that tells the planner how trustworthy this specific number is. Most demand-planning tools surface a single global accuracy figure; Tightly puts the confidence on every line so the team commits where it's safe and reviews where it isn't.
How long until the model is trained on our data?
First useful forecasts come from cohort-based modelling on history within the first week of integration. Native ML refinements compound over the next 4–8 weeks as the model learns your channel-level seasonality, returns rate and trading rhythm.
Can our team see what changed and why?
Every forecast change carries a 'why' — the signal that moved it (a viral SKU, a tariff move, an outlier returned) and the magnitude of the contribution. Planning leads can review and accept or override, with the audit logged.
Sell more at full price. Plan from one set of numbers you can trust.
There's nothing to rip out. Tightly runs on your existing ERP, EDI, e-commerce and POS. Give us 30 minutes and we'll show it on your own categories.