Retail and FMCG
Forecasts at store and product level, fewer shelf gaps and lower losses on fresh products.
Context
In retail and FMCG, results are the sum of thousands of small decisions: hundreds of stores, tens of thousands of SKUs, weekly promotional leaflets, fresh products with a shelf life of a few days. Stock data describes the distribution centre, while the customer sees the shelf. The data needed for these decisions sits in POS, ERP and WMS systems, the loyalty programme and buyers' spreadsheets, but every source uses different identifiers and a different calendar.
We bring this data together and build tools on top of it: demand forecasts for every store and product, automated ordering with rules a buyer can follow, promotion analytics compliant with the Omnibus Directive, and data exchange between the retailer and its suppliers. Forecasts and promotion analysis run on POS data, without customers' personal data.
Challenges in the industry
Shelf gaps despite positive stock
The customer cannot find the product although the system shows positive stock. The causes usually lie in the store: phantom stock after receiving and stocktaking errors, goods in the back room instead of on the shelf, a display that departs from the planogram, a minimum stock level that is too low for rising sales.
Promotions that are hard to settle
The result of a campaign is inflated or understated by stock-outs during it and by overlapping mechanics. After the promotion sales fall, because customers have stock at home, and part of the uplift is taken from another variant in the same category. Costs shared with the supplier are settled months later, and the 30-day reference price changes the profitability of the same mechanic.
Fresh write-offs without causes
Dairy, meat, bakery and vegetables are ordered in advance and shelf life is counted in days. Write-offs reach accounting as a single figure, with no split between expiry, damage and ordering errors, so the person placing the order does not see the consequences of their own decisions.
Different versions of the same figures
Retailer sell-out, manufacturer sell-in, ERP, WMS, the loyalty programme and buyers' spreadsheets use different product identifiers, different hierarchies and different promotion calendars. Before a commercial negotiation, someone spends weeks establishing whose figure is right.
Where we deploy software and AI
Demand forecast at store and product level
The model accounts for seasonality, weather, prices, promotions and the calendar, including the eight trading Sundays a year and stores closed on Christmas Eve. Slow-moving products and new launches are forecast through the category hierarchy. The planner sees the forecast together with its uncertainty range.
Promotion planning and evaluation
We separate baseline sales from the promotion effect and calculate cannibalisation, the halo effect and the post-promotion dip, adjusted for stock-outs during the campaign. The buyer compares mechanics before approving the plan, and cost settlement with the supplier relies on the same statement.
Automated ordering of fresh products
The system calculates the order from the forecast, stock, shelf life at delivery and minimum display quantity, within ordering windows agreed with suppliers. It does not change a placed order unilaterally. Routine items are ordered automatically and exceptions go to an approver.
Detection of shelf gaps and phantom stock
The system compares expected hourly sales with actual sales and flags products that stopped selling despite positive stock. The store receives a short list to check on the shelf and in the back room, and confirmed phantom stock goes to correction and root-cause analysis.
Pricing, elasticity and reference price
We estimate price elasticity, taking competitor prices, substitution and the products by which customers judge the retailer's price level into account. Promotion recommendations calculate margin from the lowest price of the previous 30 days, and the system checks that the price is consistent in the leaflet, on the shelf and in the app.
Retailer-manufacturer data exchange
We provide the manufacturer with an agreed scope of data, namely sell-out by product and region, the promotion calendar and stock levels, on contractual terms and without competitors' prices or personal data. Both parties plan promotions and deliveries on the same figures.
Regulatory requirements are taken into account from the analysis stage
Contractual advantage in the food trade
The act sets payment terms (30 days for perishable products, 60 for others) and prohibits, among other things, unilateral changes to orders, cancelling orders for fresh products shortly before delivery, and charging the supplier for promotion costs without a prior agreement. The penalty can reach 3% of turnover. We build these rules directly into the ordering and promotion settlement tools.
Lowest price from the previous 30 days
Every price reduction notice must state the lowest price from the 30 days before the reduction, and for fast-perishing products the price before the first reduction. According to guidance from UOKiK (the Polish competition and consumer protection authority), the reference point is calculated separately for each store. The promotion system therefore keeps price history at store and day level.
Reducing food waste
Stores above 250 m² where food generates at least half of revenue must have an agreement with a non-governmental organisation, run an annual campaign and pay PLN 0.1 per kilogram of wasted food. The report is due by 31 March. Withdrawal data split by cause simplifies the report, and the EU directive sets reduction targets for 2030.
GDPR and loyalty programmes
Purchase history linked to a card or app is personal data. Profiling for coupons requires a legal basis, clear information for the customer and the right to object to direct marketing. We run forecasts, gap detection and promotion evaluation on POS data without a customer identifier.
How we start
Choosing a category and a metric
We start with one category or region and one metric with a clear definition: on-shelf availability, write-offs or forecast accuracy. We agree who places orders, who acts on the gap list and who makes decisions.
Review of data and systems
We check stock accuracy, price and promotion history per store, the quality of product master data and the delivery calendar. We identify gaps, duplicates and missing history.
Pilot with a control group
The model runs alongside the current process in selected stores or categories. We compare results with a control group on real orders and write-offs, over several full ordering cycles and at least one promotion.
Rollout and training
We extend the scope, integrate the solution with the ERP, EDI and the ordering process, and train buyers and store ordering staff.
Related services
Frequently asked questions
No. Forecasts, gap detection and promotion evaluation can be done with POS, stock and price data. Loyalty data allows analysis of baskets and customers, but requires additional GDPR arrangements.