CASE STUDY · COMPUTER VISION APPLICATIONS
Building a retail foot-traffic analytics system using existing cameras.
- 15%Checkout queue time down 15%
01 · PROJECT OVERVIEW
The client and the brief.
An anonymised retailer. Name withheld under confidentiality.
People counting and heatmaps built on existing cameras, without face recognition. The system counts entries, estimates occupancy and shows where people spend time. It does not identify anyone and does not store identifiable images.
- 1No data on store traffic.
- 2Managers decided staffing from gut feel and till receipts, which show who bought, not who walked in and left.
- 3They could not tell when the store was busy, which zones attracted people, or how long customers waited at checkout.
- 4Staff were often short at peaks and idle at quiet times.
02 · WHAT WENT WRONG
What went wrong, and how we fixed it.
Every engagement hits problems. These were the ones that mattered here, and what we changed so they did not come back.
- PROBLEM 01Early door counts were too high at one store
A staff member regularly walked back and forth across the entry line, and a mirror near the door created ghost detections. We added a staff-exclusion approach based on zones and timing rather than identity, masked the mirror area and added a short debounce on door crossings. We then checked the counts against manual counts again before reporting to the business.
- PROBLEM 02We then checked the counts against manual counts again before reporting to the business
- PROBLEM 03A second hurdle was reporting trust
Store managers doubted the numbers at first, because they could see the counts differ slightly from what they remembered. We ran a one-week side-by-side check with a person counting at the door, shared the comparison openly and explained where small differences come from. Once managers saw the counts were close and consistent, they began using them for rosters.
- PROBLEM 04Consistency mattered more than perfection, since decisions depend on patterns by hour, not on exact headcounts
03 · THE PROCESS
How we ran it.
- 01Weeks 1 to 2
audited the existing cameras in a few pilot stores, checked angles, coverage and image quality, and agreed the questions the store managers wanted answered.
- 02Weeks 3 to 5
built and tuned counting and zone logic on pilot footage, and agreed privacy rules with the retailer's legal and store teams.
- 03Weeks 6 to 8
ran the system in pilot stores, compared its counts with manual counts on sample days, and corrected zone boundaries.
- 04Weeks 9 onward
weekly reporting to store managers and regional leads, with a monthly review.
04 · RESULTS
What changed.
Measured against the way the work ran before the engagement began.
- NOW 01
Rosters are built from hourly traffic patterns, with extra cover at the busiest times.
- NOW 02
Layout changes are tested: the team compares zone activity before and after moving a display.
- NOW 03
Queue length triggers a prompt to open another till, rather than waiting for complaints.
- NOW 04
The retailer has a clear privacy statement about what the system does and does not record, which store staff can explain to customers.
Traffic by hour and zone reported weekly
Staffing adjusted to peaks
- 15%
Checkout queue time down 15%
05 · WHAT WE BUILT
How it works.
- Person detection and tracking running at the edge, on a small gateway in each store.
- Only anonymous metadata leaves the store: counts, zone IDs, timestamps and queue length.Video stays local and is discarded after processing.
- Zones drawn per camera by the store team, so heatmaps match their floor plan.
- Entry and exit lines for door counts, with rules to avoid double counting people who linger at the door.
- A central dashboard with traffic by hour and zone, and a weekly summary by email.
- Face recognition was deliberately left out of the design, and the privacy approach was documented.
- Monitoring and handover
- We added alerts for a camera going offline or a count dropping to zero in opening hours, because a dead camera otherwise looks like an empty store. The retailer's analytics lead received training on adjusting zones, and a short guide on what to do after a camera is moved. Rolling the system out to further stores follows a fixed checklist: audit, zone set-up, side-by-side check, then reporting.
06 · QUICK ANSWERS
The short version.
- What problem did the client have?
- No data on store traffic. Managers decided staffing from gut feel and till receipts, which show who bought, not who walked in and left. They could not tell when the store was busy, which zones attracted people, or how long customers waited at checkout. Staff were often short at peaks and idle at quiet times.
- What did Brandsmashers build?
- People counting and heatmaps built on existing cameras, without face recognition. The system counts entries, estimates occupancy and shows where people spend time. It does not identify anyone and does not store identifiable images.
- What were the results?
- Traffic by hour and zone reported weekly; staffing adjusted to peaks; checkout queue time down 15%.
YOUR TURN
Have a similar requirement?
Tell us what is stuck. We will come back with a plan and matched, pre-vetted engineers.
