Per-SKU demand models
Built from your sales history with seasonality, trend, and promotional lift separated out, so a Q4 peak does not get smeared across the whole year.
Forecasting
A number you can actually place an order against.
Per-SKU forecasts built from sales history, seasonality, promotions, and channel mix, then reviewed weekly against what actually happened.
The problem
What this actually feels like.
Most e-commerce forecasts are a trailing average with a growth percentage on top. That works until the thing you actually needed to predict happens: a season, a promotion, a channel launch, a competitor going out of stock.
The result is a buy plan that is confidently wrong in both directions at once. Too much of the slow movers, because the average flattered them, and too little of the seasonal peak, because a twelve-month average cannot see a six-week spike.
And because no one owns the forecast, no one ever checks it against what happened. So the same error repeats every cycle, and everyone learns to distrust the number rather than fix it.
What we own
Off your plate, not just advised on.
Promotion and launch modelling
Planned promotions, drops, and channel launches are modelled as events with their own lift, rather than left to distort the baseline afterwards.
Channel mix
Amazon, Shopify, Walmart, and TikTok Shop are forecast separately and then reconciled, because the same SKU does not behave the same way on each.
Forecast accuracy tracking
Every forecast is scored against what actually sold. The error is visible, and it is what drives the next revision. This is the step most brands skip.
Translation into a buy plan
A forecast that does not become a purchase order is a report. We turn it into order quantities and dates against supplier lead times and your cash position.
Month one
What you see in the first thirty days.
- A baseline forecast per SKU across every channel you sell on, with seasonality separated from trend.
- A written accuracy baseline, so improvement is measurable rather than asserted.
- A buy plan derived from the forecast, sized to lead times and minimums.
- A weekly review where the forecast is corrected against actuals rather than left to drift.
Proof
We publish our approach to the hard cases, including how seasonal brands forecast demand that an average cannot see, and how to plan Q4 without either stocking out in week two or carrying the excess into February.
Questions
The ones we always get asked.
- How much sales history do you need?
- Twelve months lets us separate seasonality properly. We can work with less, but the first months lean harder on category patterns and your own judgement, and we say so rather than pretending the model is confident.
- What accuracy should we expect?
- It depends heavily on your category and how promotional you are. What we commit to is measuring it honestly from week one and showing you the error, rather than quoting a number that flatters us.
- Is this software or a service?
- A service, run on our own operations ERP. You get the forecast, the buy plan, and an operator who owns both, not a login and a tutorial.
- Can you forecast new products with no history?
- With analogues from your own catalogue and category data, yes, within honest error bars. New product forecasts get reviewed more often because they correct faster.