The rota is still written on instinct
A city-centre bar with a 400-person capacity schedules staff from last year's takings and the manager's memory. On some Thursdays the bar is three deep at 9pm with two people serving. On others, five staff stand around polishing glasses. Tills show what was sold, not how many people gave up and left.
Video analytics can fill that gap. The cameras already covering the entrance and bar can count people in, measure how long the queue at the bar is, and show how full each area is at any moment. Over a few months, that becomes a much better staffing model than instinct.
Hospitality is also a setting where guests are sensitive about being watched, so how you use it matters as much as what it measures.
Useful measurements for venues
- Entries and occupancy. Real-time headcount against capacity, useful for licensing conditions and door staff.
- Queue length and wait time at bars, counters, ticket gates and hotel reception.
- Table occupancy and turn time. How long tables sit empty after guests leave, and how long a cover really takes by day of week.
- Zone heat maps. Which areas of a beer garden or exhibition hall actually get used.
- Walk-aways. People who join a queue and leave before being served, a direct measure of lost sales.
- Crowd density at events, for safety at pinch points and exits.
All of these can be computed as anonymous counts and timings. None requires knowing who anyone is.
Turning numbers into decisions
| Measurement | Decision it informs |
|---|---|
| Queue length by 15-minute slot | When to put a second bartender on, and when to send one on a break |
| Table reset time | Whether bussing staff are in the right place, and how many covers a section can take |
| Occupancy by area | Which rooms to open early, where to heat or light, where to place a pop-up bar |
| Walk-away count | The cost of understaffing, in a number the finance director will believe |
| Entry rate at events | How many gates to open and when to switch to exit mode |
The pattern that works is to join camera data with till, booking and rota data. Occupancy alone tells you it was busy. Occupancy next to sales per head and staff on shift tells you whether busy was profitable. That joining is often more work than the vision, and it is where data science earns its fee.
Do not use facial recognition
It is tempting to go further: recognise regulars, flag banned customers, track individual guest journeys. We would advise strongly against it for almost every venue.
Facial recognition is biometric processing, which under UK and EU data protection law is special category data with a high bar for lawful use. The EU AI Act restricts several biometric uses outright and places heavy obligations on others. It is also the kind of thing that ends up in a local newspaper. The anonymous version, counting people and timing queues, gives nearly all of the operational value without any of that exposure.
Guests accept that a venue counts how busy it is. Very few accept that it recognises their face.
Practical set-up in a venue
- Audit existing cameras: which cover entrances, bar fronts, counters and main floor areas
- Add overhead cameras at entrances if current ones are angled for identification rather than counting
- Process video on a small on-site device and keep only counts, discarding frames
- Update signage and your privacy notice to describe the analytics honestly
- Feed counts into the tool managers already use, not a new dashboard
Dim lighting, coloured stage lights and crowded dance floors reduce accuracy. Test on footage from your busiest night, not a quiet Tuesday afternoon. For the angles and tracking issues in more depth, see our post on counting objects with existing cameras.
When video analytics is overkill
A 40-cover restaurant with a booking system already knows its covers by slot and its table turn times, roughly. A hotel with good property management data knows arrival patterns. In those cases the cheaper win is usually better use of data you have, through AI integration for hospitality tools or a simple forecasting model on bookings and sales.
Video analytics pays back most in venues where lots of demand is unbooked and invisible to the till: walk-in bars, food halls, attractions, stadiums, festivals and large hotel lobbies.
What we would look at first
At SpiderHunts we would ask for a month of till data and one weekend of footage from the bar or counter cameras. If the footage shows queues and walk-aways that the till data cannot see, there is a case. If the till data already explains most of the variation in demand, we would tell you to start with a staffing forecast and leave the cameras doing security.
If there is a case, a single venue pilot over eight to ten weeks is usually enough. Run it through at least one bank holiday or big event, compare the camera-informed rota against the old one on labour cost and walk-aways, and only then decide whether to roll it across the group. Venues differ more than head offices expect, so the second site is a test too, not a formality.