Payoff Reads · Cameras and loss prevention

How did installing cameras in your stores actually change customer and employee behavior?

Find out to what extent camera presence drove fewer incidents and changed how customers and employees behave, vs. similar stores without cameras, and what happened to shrink, sales and labor.

Customers: Incidents−35%Significant drop
Employees: Register exceptions (voids, refunds and no-sale drawer opens)−21%Significant drop
Shrink, sales and labor: No significant effect
Liability claims: Too few to read at one chain

Heartland Markets is a fictional chain. Every number on this page is made up to show the format.

What your numbers can't show

Cameras show you the incidents they caught. They can't show the ones that never happened.

What you have: the camera system's count of incidents caught, and a before-and-after at the stores that got cameras.
What it can't show: how many incidents those stores would have had without cameras. Cameras usually go to the worst stores first, and stores picked at their worst tend to improve the next year, even without cameras. A before-and-after gives the cameras credit for that improvement.
What the read shows:
- incidents avoided: how many fewer incidents your camera stores had vs. their twins, which is the practical way to judge what the cameras were worth.
- how employees changed: register exceptions, meaning voids, refunds and no-sale drawer opens, the transactions loss prevention teams watch for theft at the counter.
- what happened to shrink, sales and labor.

In the sample read below, a before-and-after showed incidents down 52%. Measured vs. the twins, the cameras cut them by 35%. About a third of the drop would have happened without cameras.

What lands on your desk

Cameras at twelve stores, as Seurat reads it

Heartland Markets put AI-alert cameras in 12 of its 42 stores in April 2025, starting with the 12 that logged the most incidents in 2024. Each is read for a year vs. a twin built from stores without cameras, including stores that also had a bad 2024.

the 12 camera stores vs. their twins, first year

Measure	Result	Verdict
Logged incidents	−35%	Significant drop
Before-and-after, for comparison	−52%	Overstates it
Register exceptions: voids, refunds and no-sale drawer opens	−21%	Significant drop
Shrink at inventory counts	−0.1 points of sales	No significant effect
Inside sales	+0.3%	No significant effect
Labor hours per week	−2	No significant effect
Liability claims	9, vs. about 11 expected	Too few to read at one chain

Why it's read as a group: one store logs about 15 incidents a year, too few to judge on its own, so the read pools the 12 stores and reports them together.

KeepWhat to do: the cameras cut incidents and changed what happens at the register. Judge the next wave on incidents avoided, not shrink, and add the next stores a few at a time, chosen by a rule rather than by who's worst, so the next read is sharper.

The confidence band shows the 95% confidence interval. When the band spans zero, the result isn't statistically significant. Heartland Markets is a fictional chain. Every number on this page is made up to show the format.

How the read works

Each camera store gets a twin from your stores without cameras, including stores that had a bad year too.
The twin shows what each store's incidents, register exceptions, shrink, sales and labor would have been without cameras.
The read checks itself. The twin has to have tracked the store closely before the install. Placebo tests then rerun the read on a made-up install date, when nothing changed, and on stores that never got cameras. Both should show no change. If either shows a change, the read isn't working properly, and we tell you.
Each measure gets a confidence interval and a verdict, for the camera stores together and by store type.

Side note: Why do stores picked at their worst flatter the cameras? A bad year is partly bad luck, and bad luck doesn't repeat. A twin that had the same bad year shows how much of the improvement would have come without cameras.

Can your chain be read?

Measure and setup	Smallest change it can show
Sales and labor hours, 10 of 40 stores, 26 weeks after the install	About 1.4%
Logged incidents, 25 camera stores vs. 25 stores without cameras, 10 incidents per store per year, 1 year	About 30% fewer
Register exceptions, 25 camera stores vs. 25 stores without cameras, 1 year	About 10% fewer
Shrink at inventory counts, 10 of 40 stores, 4 counts before the install and 2 after it	About a quarter of your shrink
Liability claims at one chain, the same 50 stores, 1 claim per store per year	About 80% fewer, so in practice it can't
Liability claims pooled across chains, 300 camera stores vs. 300 stores without cameras	About 30% fewer

One chain can measure incidents, register exceptions, sales and labor. Claims need many chains pooled, which is work we do with camera vendors. Rough figures from a standard formula, before we see your data.

Scope this read

What we need, and what stays with you

We need:

Install dates at each store
Incident counts by store and month, from your team's log or the camera system
Voids, refunds and no-sale drawer opens by store and week, totals only
Shrink at each inventory count, by store
Weekly sales and labor hours by store
1 to 2 years of history
Liability claims by store and month, if you want them read

We never ask for:

Camera footage
The names of anyone involved in an incident
Cashier names or IDs
Your chain's totals or your P&L

How we handle your data →

How long it takes

About 2 weeks from the day your data lands, then a 60-minute readout with whoever needs to hear it.
We scope it to your chain on a short call: which stores, which measures, which months.

Scope this read Book a call

Questions buyers ask us

Can it show our insurance claims went down? Not at one chain: claims are too rare, about 1 per store per year. Pooled across many chains, a drop of about 30% shows up within a year. We tell you up front what your data can show.
Our camera vendor already reports what the cameras caught. How is this different? Incidents caught shows what the cameras saw. Incidents avoided, the ones your camera stores would have had without cameras, is the practical way to judge what the installation was worth.
We picked our worst stores first. Does that ruin the read? No, it's the reason to do it. The twin is built from stores that had a similar bad stretch, so it shows the improvement those stores would have seen without cameras. The cameras get credit only for what they added beyond that, not for riding the natural improvement wave.
What if the read shows the cameras didn't help? A read with no significant lift is a private finding. You see every result first, and nothing goes anywhere without your approval.
Do we have to connect any systems? No. Exports from your back office and incident log, sent however your IT team prefers.
Who does the work? We do. Your team runs the exports, shares them with us, and comes back to hear our readout.

After this read

Read several changes in one request, or keep your stores measured week by week.

Scope several changes →How monitoring works →