ujans
Blog
Retail

Demand forecasting and stock planning: a practical guide for retail, supermarkets and restaurants

Too much stock ties up capital and spoils; too little loses sales. Forecasting demand with reasonable accuracy reduces stock cost and lost sales together.

ujans team··5 min read

Retailers, supermarkets, cafés and restaurants answer the same question constantly: how much should I order? Order too much and money and shelf space are tied up, goods spoil or go on discount; order too little and customers see an empty shelf and go to a competitor. Demand forecasting is the way to move this decision from instinct to data.

Why is demand forecasting hard?

Demand is not a single number; it is the result of many factors: season, day and hour, weather, holidays, promotions, price, competitor moves, trends, events. Also, sales data is not "real demand" but "demand that was met": if a product sold out, sales show zero, but demand existed. So marking out-of-stock periods in the data matters.

Core approaches

  1. Simple historical averages: The average of recent weeks, the same period last year. Quick, simple and useful for stable products.
  2. Seasonality and trend adjustment: Adding day-of-week, period-of-month and season effects and the overall growth or decline.
  3. Event and promotion effect: Measuring the effect of past promotions, holidays and events on sales and applying it to future plans.
  4. Statistical and machine learning models: More accurate forecasting with many products and variables; they can use external variables such as weather, price and promotion.
  5. AI-assisted tools: Process data automatically and produce suggestions; verify with your own data; see AI in hospitality and retail and AI process automation.

The most advanced model is not always needed. For most small businesses clean data, regular tracking and a simple method are enough; the real gain is in the data being clean and continuous.

First prepare the data

From forecast to order: core concepts

  • Forecast demand: Expected sales in a given period.
  • Lead time: Time from placing the order to delivery.
  • Safety stock: A buffer against demand or supply deviation. It rises as demand variability and lead time increase; set by product criticality and cost.
  • Reorder point: When stock falls to this level, order (expected demand during lead time plus safety stock).
  • Order quantity: By procurement cost, storage cost, discount tiers and shelf life.
  • Perishable products: For food and fresh produce, ordering is short-cycle and more frequent; waste risk is the cost of forecast error.

Product classification

Not every product needs the same care:

  • By sales value (ABC): A (few high-value products) tight tracking, C (many low-value products) simple rules.
  • By demand stability (XYZ): X (stable) is easy to forecast, Z (irregular) needs safety stock and flexibility.
  • By shelf life: Fast-perishing items get tight control.

Focusing on critical products (best sellers and quickly perishing ones) makes your effort most productive.

Application by sector

Restaurants and cafés

Prep and ordering by daily or weekly sales and reservation data; frequent small orders for perishables such as fresh fruit and vegetables, dairy and meat. For prep planning and waste reduction see restaurant cost control and inventory management; for menu decisions menu engineering; for staff planning shift and staff management software.

Supermarkets and shops

Seasonal and promotional effects, multiple branches and warehouse transfers; see central management for multi-branch businesses, warehouse management and omnichannel retail. Promotion and loyalty data shape demand; see how to set up a loyalty program.

Hotels

Occupancy forecasting; demand linked to season and events; see hotel revenue management. Plan stores, laundry and staff by occupancy.

E-commerce

Promotion, traffic and season effect; return rate; see e-commerce SEO and things to get right when building an online store.

Measure accuracy

Compare the forecast with actuals:

  • Mean absolute error and percentage error: How far did the forecast miss?
  • Bias: Are you consistently over- or under-forecasting?
  • Stock indicators: Stock turn, stock-out rate, waste rate, service level.

Analyse error with its source (promotion, weather, stock-out?) and update the model or rule. To track indicators on a panel see KPIs and management dashboards and Power BI.

Human judgement and exceptions

A model knows data history but may not know a new rival next door, a local event or a trend that exploded on social media. Adjusting the forecast with expert judgement and recording adjustments (to learn later) is good practice. New products have no history; start from similar products and small pilot orders.

Data security and continuity

Sales and stock data is the base of decisions: back it up, limit access and keep an alternative method for system outages; see backup strategy, POS security and the disaster recovery plan.

Common mistakes

  • Treating past sales as real demand and not correcting stock-out periods,
  • Trusting the forecast when the stock balance is wrong,
  • Applying the same method to all products,
  • Not calculating safety stock or keeping it too large,
  • Never measuring forecast accuracy,
  • Not putting promotions, weather and events in the calendar,
  • Large, infrequent orders for perishables,
  • Building an overly complex model without data cleaning.

A starter plan

  1. Pick the 20–30 most important products (by ABC analysis).
  2. Clean the last 12 months of sales and stock-out data.
  3. Forecast with a simple method (moving average plus season).
  4. Set safety stock and reorder point for each product.
  5. Measure accuracy monthly and improve.
  6. As data accumulates, move to more advanced models or AI tools.

ujans makes demand forecasting and stock planning work with your data by integrating POS, stock and reporting systems.

Looking for help with this?

Tell us what you need and our team will get back to you within one business day.

Get a quote