Dynamic pricing and demand forecasting for South African retailers

South African retail margins are under constant pressure from exchange rate movements, long supply chain lead times and intense promotional competition. Pricing that never changes leaves margin behind when demand is strong, and ties up cash in stock that does not move when demand is weak. Demand forecasting and dynamic pricing address both, but only when they are built on good data and clear rules.

What are demand forecasting and dynamic pricing?

Demand forecasting predicts how many units of each product will sell, where and when, using past sales and known future events. Dynamic pricing adjusts prices over time using that forecast together with stock levels, competitor prices and margin targets. Forecasting tells you how much to order; dynamic pricing helps you sell it at the best achievable margin.

Which demand signals matter in South Africa?

How does demand forecasting work?

A forecast starts with history: sales by product, location and day, ideally for at least two years so annual patterns are visible. Simple methods such as moving averages and seasonal baselines are a sound start. Machine learning models add value when there are many products and many influences at once, because they can learn how payday, promotions, price and weather interact. Whatever the method, a forecast is only useful when it is checked against actual sales each week and the error is tracked.

1. Clean the sales history: remove duplicates, mark stockouts (days with no sales because there was no stock, not no demand) and label promotions.
2. Build a baseline forecast by product and location, including seasonality and payday effects.
3. Add known future events: promotions, holidays and price changes.
4. Turn the forecast into reorder quantities using supplier lead times and safety stock.
5. Measure forecast accuracy every week and improve the products with the largest errors first.

How does dynamic pricing work?

Dynamic pricing uses price elasticity, how much demand changes when the price changes, to choose prices that meet a goal such as gross profit or sell-through before the end of a season. It works best on products where demand responds to price and where you have enough sales to measure that response. Typical uses are clearing slow-moving or seasonal stock earlier at a smaller discount, protecting margin on lines that sell regardless of small price changes, and responding to competitor moves on key products.

What guardrails does dynamic pricing need?

Automated pricing without limits can damage margin and customer trust quickly. Every system should run within rules the business sets:

• A minimum margin below which no price can fall.
• A maximum price change per day or week, so customers do not see erratic prices.
• Price consistency across shelf labels, website and till. Under the Consumer Protection Act, a customer generally cannot be charged more than the price displayed, so a price change that has not reached the shelf costs you the difference.
• Clear rules for essential goods. Charging excessive prices for essential items, particularly during a declared national disaster, can attract action from the Competition Commission.
• A person who reviews and approves larger changes.

What data do you need to start?

• Sales history by product, location and day, from your point-of-sale and online store.
• Stock levels over time, including when items were out of stock.
• Product cost, so margin can be calculated for every price.
• Promotion and price-change history.
• Supplier lead times and minimum order quantities.
• Competitor prices for key products, where you can collect them reliably.

Most retailers already hold this data, spread across a point-of-sale system, an e-commerce platform, spreadsheets and supplier emails. Bringing it into one place is usually the first and most valuable step; see our guide to what a growth intelligence layer does.

How should a retailer get started?

1. Connect and clean sales and stock data in one place.
2. Start with demand forecasting and automated reorder suggestions, which protect cash flow with little risk.
3. Use the forecast to plan promotions and markdowns, and measure the result against a control group of stores or products.
4. Introduce automated price recommendations on a limited range, with guardrails and human approval.
5. Expand to more products once accuracy and margin results are proven.

What does it cost?

It depends mainly on how many systems hold the data and how clean that data is. At Lulonke, data and forecasting work starts with a diagnostic audit; the initial architecture for a unified intelligence layer is typically R65,000 to R220,000, scoped from that audit. Smaller retailers can often start with forecasting and reorder reporting on their existing systems before investing further. See AI growth intelligence for how we approach it.