Customer footfall forecasting means estimating how many people will enter a mall or store on each day and in each hour ahead. That figure is the starting point for many tasks, from staff rosters and cleaning and security plans to campaign measurement. Many stores already estimate from their managers' experience. An AI forecasting system does the same job but can look at far more data: every branch, every hour and every factor at once. This article explains how such systems work, what data they use, how to read the results, where their limits are, and what mall and retail-chain executives can use them for. If you are interested only in staff scheduling, continue with Scheduling store staff from real customer numbers.

How Footfall Forecasting Differs from Sales Forecasting
| Footfall forecasting | Sales forecasting | |
|---|---|---|
| Answers | How many people will come in, and when | How much will be sold |
| Main data | Entries from the people counting system | Sales from the point-of-sale system |
| Used to plan | Customer-facing staff, services, security, energy | Stock, budgets, sales targets |
| Strength | Shows total service demand, including people who enter but do not buy | Links directly to revenue |
The two work well together, because sales equal footfall multiplied by conversion rate multiplied by average transaction value. Once you know how many people will come in, the team can see whether sales are moving because fewer people came, or because fewer visitors were turned into buyers. These metrics are covered in 8 retail store KPIs.
How an AI Forecasting System Works
The principle is simple. The system learns patterns from the past, then adjusts for what is known in advance about the future. It has four parts.
1. Learning Regular Patterns
Historical footfall contains repeating patterns: lunchtime and evening peaks, weekends that differ from weekdays, year-end that differs from mid-year. The system separates these into hour-of-day, day-of-week and seasonal patterns.
2. Adding Factors Known in Advance
Things that make one day differ from the usual pattern include public holidays, long weekends, promotions, mall events and the weather. The system learns from the past how much each factor has raised or lowered footfall, then applies that to upcoming days.
3. Adjusting for Each Branch
A branch inside a mall and a street-front branch react differently to rain. A branch near offices and one near housing peak at different times. Machine learning lets the system learn each branch's pattern separately rather than applying one formula everywhere.
4. Comparing with Actuals and Adjusting the Next Round
Once the day has passed, the system has actual figures from the people counting system to compare against the forecast. Errors that repeat are used to adjust the next forecast.
Key principle: AI does not know the future. It condenses what has happened, across large amounts of data, into figures you can plan with. People who know the local context still need to review the figures and adjust them when they know something the system does not.
Data Used for Customer Footfall Forecasting
| Data | Source | Why it is needed |
|---|---|---|
| Hourly historical entries | People counting system | The basis of every forecast |
| Sales and transaction counts | Point-of-sale system | Supports the forecast and converts footfall into sales |
| Holiday and seasonal calendar | The organisation's calendar | Explains patterns that rise and fall over time |
| Weather | Weather forecasts | Rain or heat changes whether people go out |
| Promotions and events | Marketing and events teams | Explains unusually busy days |
| One-off incidents | Branch logs | For example partial closures or a closed entrance, used to remove abnormal days from the data |
The quality of the footfall data matters most. If the counting system misses an entrance or includes staff, the forecast learns from wrong figures and is off as a result. How to test a counting system's accuracy is covered in the people counting buying guide.
How to Read a Forecast
Read It as a Range, Not a Single Number
Every forecast has error, so read a forecast figure as "likely around this level". If the system shows a high-low range, use the upper bound to plan work that cannot fail, such as security cover, and the middle value for routine planning.
Measure the Error Yourself
An easy method is to compare forecast with actual hour by hour, then average the error as a percentage (mean absolute percentage error).
Hourly error (%) = |actual − forecast| ÷ actual × 100
Example (hypothetical figures): a store forecasts 200 customers between 18:00 and 19:00, and 180 actually come in. The error for that hour = 20 ÷ 180 × 100 ≈ 11%. Averaging every hour of the week gives a figure you can track to see whether forecasts are getting better or worse.
Points to keep in mind
- Look at hourly error, not just daily. The daily total may match because an over-forecast in the morning cancels out an under-forecast in the evening. If you schedule by the hour, hourly error matters more
- Watch quiet hours. When only a few people come in, a small miss produces a very high percentage. Look at the absolute miss as well
- Be wary of accuracy figures without conditions. Forecast accuracy depends on the type of store, the length of the historical data and how far ahead you forecast. A credible figure should say what it was measured against, how, and over what period
Limits to Be Aware Of
- New branches have no history. The system needs a period of historical data first. Early on, it may have to rely on patterns from similar branches plus manual adjustment
- Events that have never happened before. A competitor opening next door, roadworks outside the mall, or a disruption to travel across the city are not known to the system in advance. People must adjust for them
- Layout or entrance changes. When a new entrance opens or a store moves, data from before and after cannot be compared directly. Tell the system or record the change
- External data must actually exist. If marketing does not feed the promotion calendar into the system, the forecast will not know a promotion is coming
What Malls and Stores Use Forecasts For
| Task | How the forecast is used | Who uses it |
|---|---|---|
| Store staff scheduling | Place working hours where customers are busiest | Branch managers, operations |
| Mall services | Plan cleaning, security and customer-service points around expected busy periods | Facilities and services |
| Energy and air conditioning | Let the engineering team prepare cooling for days expected to be busy | Building engineers |
| Campaign and event measurement | Use the forecast as a baseline for "how many people would have come without the event" | Marketing |
| Stock and replenishment | Prepare stock and shelf-filling staff ahead of busy periods | Branch managers |
| Targets and tracking | Compare actuals with forecast to tell whether sales fell because of fewer visitors or fewer sales | Executives, head office |
Examples of use (hypothetical scenarios)
A mall over a long weekend: the system forecasts that the Saturday of a long weekend will be clearly busier than a normal Saturday. Facilities add extra cleaning rounds for toilets and the food court, and security adds staff at the car-park entrance in the afternoon.
A retail chain on a rainy day: the weather forecast shows rain all afternoon. The system lowers figures for street-front branches and raises those for branches inside malls. The area manager reviews the figures and decides to move some staff between nearby branches.
Event measurement: marketing uses the forecast for the event day, calculated without the effect of the event, as a baseline against actual footfall. The difference is the number of extra people the event drew in. The detailed method is in Measuring mall events with visitor data.
How footfall supports a mall's energy management is covered in Mall electricity costs: where to cut when you know real occupancy.
How to Get Started with Forecasting
- Make footfall data complete and reliable first. Count every entrance, exclude staff, and validate accuracy
- Collect calendars of holidays, promotions and events for each branch, in a format the system can take in
- Start with pilot branches. Choose branches of different types to see which kinds of store forecasting works for
- Agree how to measure results before starting: how to measure forecast error, and how to compare with the current planning method
- Let managers adjust the figures, and record the reasons, so you learn what data the system is still missing
How IMAS Helps with Customer Footfall Forecasting
IMAS uses the platform of Retail Insight, from Norway, for visitor forecasting, together with data from the people counting system.
- Hourly customer forecasts up to 8 weeks ahead, according to Retail Insight
- Several data sources combined: historical footfall, sales, seasonality and weather
- Learns each store's pattern with machine learning, and recommends where staff hours should go
- The store manager makes the final decision and can edit every recommendation
- Connects with existing systems such as point-of-sale, staff scheduling, people counting, and accounting or enterprise resource planning (ERP)
The footfall data behind the forecast comes from 3D people counting, which counts anonymously, and results can be compared with actuals on the XPERIO dashboard. See use cases by sector on the shopping malls and retail chains pages.
Frequently asked questions
−What is customer footfall forecasting?
It is estimating how many people will enter a mall or store on each day and in each hour ahead, used to plan staff, services and marketing measurement.
+What data does AI use to forecast customer numbers?
Mainly historical entries from the people counting system, combined with sales, seasonality, holidays, weather, promotions and events.
+Do we need a people counting system first?
It is recommended, because accurate historical footfall is the basis of the forecast. Using sales alone hides the people who came in but did not buy.
+How accurate is the forecast?
It depends on the type of store, the quality and length of historical data, and how far ahead you forecast. The best approach is to measure it yourself in pilot branches by comparing forecast with actual hour by hour.
+Can AI make roster decisions instead of managers?
No, and it should not. The system provides figures and recommendations. Managers decide and can change every one, because they know context the system does not.
Want to Know in Advance How Many Customers Each Branch Will Get?
The IMAS team helps set up people counting, connect sales data, and configure forecasting so executives and branch managers can actually plan with it. You can start with pilot branches. Details and costs are provided in a quotation.
References
- Retail Insight AS (Norway). Customer forecasting and staff-hour planning platform information. Website retailinsight.no
- Retail Insight AS. Predictive staffing transforms store operations. Website retailinsight.no
- IMAS Analytics AI. Our Solutions: Predictive Forecasting. Website imas-analytics.ai
- Hyndman, R. J. & Athanasopoulos, G. (2021). Forecasting: Principles and Practice, 3rd edition. OTexts
