Know what to reorder before the stockout.
We build a demand forecast on your own sales history — across every channel you sell on, not just the ones a tool supports — and turn it into reorder alerts your team can act on and your agents can draft.
What our clients say
Four artifacts, not a slide deck
Everything is deployed on your side and stays yours — including the model and its assumptions.
A forecast per SKU and per channel
Built on your own sales history, with seasonality, lead times and the promotional peaks your team already knows about but no generic tool does.
Reorder points you can defend
Coverage days, safety stock and reorder quantity per SKU — with the assumption behind each number written down, so purchasing can argue with it.
Alerts where the work happens
Not another dashboard nobody opens. Alerts land where your team already works, and the agent drafts the purchase proposal for a human to approve.
The accuracy report
Forecast versus reality, measured monthly. If the model degrades you will see it, because a forecast nobody audits is just a number.
Thirty days to the first usable forecast
Data: sales history, current stock, lead times per supplier. This is where most projects stall, so it goes first.
Baseline model and backtest against your last 12 months, so you see how it would have performed before you trust it forward.
Reorder logic, safety stock and thresholds tuned with the person who actually places the orders.
Alerts wired up, agent drafting purchase proposals behind an approval gate, and the accuracy report scheduled.
Four things. If you have them, we start this week.
This is the whole list. We ask for it up front so nobody spends a meeting finding out the project cannot start.
- At least 12 months of sales history per SKU, in any format you can export
- Current stock and supplier lead times — approximate is fine, missing is not
- The person who places the purchase orders today, for two sessions
- A decision on what an agent may propose and who signs off on a purchase
It pays for itself within 10 months, or we keep working for free.
We agree on the baseline before we start — your stockouts and your overstock over the last year — and we measure against it. If the accelerator has not paid for itself in ten months, we keep working on it at no extra cost until it does.
Questions we get before the first call
How is this different from the Platform stock forecast?
The product feature covers the marketplaces Epinium connects to, and if that is where you sell, use it — it is cheaper and it is already built. This accelerator is for the channels, data sources or logic the product does not cover yet.
Our history is messy. Is that a blocker?
Usually not, but it is the first thing we look at. Week one exists to tell you whether your data can carry a forecast — and if it cannot, what would have to change first.
Does the agent place orders by itself?
No. It drafts the proposal; a person approves it. Purchasing is exactly the kind of decision that stays behind an approval gate.
What if demand changes completely?
Then the accuracy report shows it and the model is retrained. That is why measuring forecast against reality is one of the deliverables and not an afterthought.
Want to know if your data can carry a forecast?
A free diagnosis: we look at your sales history and your channels, and tell you what a forecast could realistically do with them before you commit to anything.
