These templates are for a store or a small chain. Three questions that retail has not changed in years: what actually brings the money and what only takes up shelf space; whether demand has changed for good or it is just noise; and what a forecast should be to serve the whole chain and a single store equally well. All templates run on one demo database: 104 weeks, 8 stores, and 60 products. This section has 7 templates.
Every template is a live walkthrough on demo data: open it in the Templates section of Analytics Agent and press “Use this template” to run it on your own data.
ABC-XYZ & Pareto — What Runs the Business
Shows: products grouped by money in two ways at once — by revenue and by profit — and also by demand stability; next to it a curve showing what share of products makes the bulk of the profit, and a list of what should leave the shelf, with the money that list keeps sitting on it.
You get: you see that the products which “bring in shoppers” do not necessarily bring money, and how much cash is frozen in slow sellers.
Demand Level Shifts — For Good or Just Noise?
Shows: for each product and for the chain as a whole, it checks whether demand has changed for good; if it has, you learn whether it was a one-time jump or a gradual shift, what demand was before and after, and how confidently this is established.
You get: the new reorder point is calculated from the new demand level, not the old one, and random noise does not turn the whole warehouse workflow upside down.
Hierarchical Forecast — Which Level Deserves the Model
Shows: simple forecasting methods compared with each other at every level — product, subcategory, category, chain — showing where a complex model truly wins and where taking the same weeks of last year is enough; separately, it measures whether pooling data across cells with a short history helps — and the error band is measured the same way for any shelf cell you choose.
You get: the forecast is built at the level where the decision is made, and extra complexity is not passed off as accuracy — including the case where the honest answer is “pooling does not help”.
Demand Forecast with Weather and Promos — Do the Features Earn Their Place?
Shows: two identical models — with weather and without it — are tested on history: where weather really lowers the error (drinks and ice cream in the heat), and where it adds nothing.
You get: you know whether collecting weather data is worth the time: for some products it is a real gain, for others it is extra work for nothing.
Price Elasticity and Markdowns — What a Price Move Is Worth
Shows: how strongly each product’s sales depend on price — on your own price history, with an honest caveat wherever the effect of price cannot be separated from a promo; separately, it computes what a 10% price increase would give and whether marking down unsold stock pays.
You get: the decision to discount or to raise the price is computed in money, not taken from the general rule that “a discount always lifts sales”.
Promo Impact – Cannibalization & Halo
Shows: each promo is judged on the whole category, not only on the hero product: how much the hero itself brought, how many sales it took from the neighbors, and how the neighboring shelf reacted; the total comes down to one profit number.
You get: a promo stops looking like a win just because the “hero” sold more — you see whether the business stayed in the black.
Safety Stock & Reorder Points
Shows: for each product, the point at which it is time to reorder, with lead time and demand variability taken into account; the chosen policy is tested on 26 weeks of history — how many sales are lost and how much money gets frozen; an alert for items below the point can be set up.
You get: the shelf stops running empty on fast sellers and growing overstocked on the rest — stock is held at a measured cost, not by habit.