A plain checklist for retail teams comparing video AI providers, from camera support and accuracy to privacy, pilots, and the first 90 days.
—
3 min read
The short answer
Judge a provider on six things: whether it works with your current cameras, how accurate its alerts are in real stores, how it handles privacy and security, how fast an alert reaches a person, how easily it rolls out across locations, and what the first 90 days look like. Ask for proof in writing, and run a short pilot in a few stores before you commit to all of them.
Where should you start?
Start with the loss you most want to cut. Checkout losses, employee theft, and safety incidents each call for different checks. A vague goal leads to vague results, so pick one or two.
What are the 10 questions?
Does it work with my current cameras? New cameras are often the biggest cost in a project.
What is your alert accuracy in real stores? Ask for the number and how it was measured.
How many false alerts will my team get? Too many wear a team out.
How fast does an alert reach a person? Speed decides how much a team can do.
Who sees the video, and where is it stored? Get a clear, specific answer.
Which security audits do you hold? Look for SOC 2 Type II and ISO 27001, and ask to see the documents.
Do you use facial recognition? Know the answer before you sign.
How does it roll out across many locations? Ask for the plan, not a promise.
What will my team need to do each day? A tool that adds work will get ignored.
What happens in the first 90 days? Ask for a week by week outline.
How do you run a good pilot?
Pick 3 to 5 stores with different layouts and traffic
Set targets before you start, such as alert accuracy, response time, and losses
Run it long enough to see patterns, often 60 to 90 days
Review results every week with the people who use the alerts
Decide with the numbers, not the demo
What are the red flags?
The provider won’t share accuracy numbers
New cameras are required and nobody can explain why
Answers about privacy are vague
There is no named person for support
Someone promises to stop all theft
Where does Spacture AI stand?
Spacture AI works with a retailer’s existing cameras, with no camera replacement required. Across 500+ US retail locations, it delivers 87% alert accuracy, a 74% reduction in false alerts, and a 43 second response time. It is SOC 2 Type II and ISO 27001 certified, GDPR and CCPA compliant, and does not use facial recognition.
Frequently asked questions
How long should a pilot last?
Often 60 to 90 days. That is long enough to see patterns beyond a single good or bad week.
How many stores should be in a pilot?
Three to five, and make them different from each other so you see how the tool handles variety.
How do I compare providers fairly?
Use the same measures for each one: alert accuracy, false alerts, response time, and the effort required from your team.
Key takeaways
Know your goal, ask all ten questions, get answers in writing, and test in real stores before a full rollout.
