2.5×
More precise on cash theft
Of the cash-theft cases the team opened, two and a half times as many were confirmed as real.
StoreTwin · Store analytics & optimisation
Turn every store into living data
StoreTwin harnesses data across the infrastructure you already own into a live data ecosystems and recommend how you can optimise sales, planogram, and stock management.
Estate view17:42:08
- 94%
- Bays in plan
- 17-19h
- Peak hours
- 12
- Low SKUs
Footfall by hour · store 1091Rainy season · idx 1.12
0810121416182022
Customer read · todayGroup level only
312shoppers browsed and left without buying — ฿46,800 unconverted
Members38% · 71% conv
Non-members62% · 44% conv
- F 57 · M 43
- Age 25-34 · 38%
- Office workers · 41%
- Bay 04 · beverages
- SKU out of plan
- Bay 07 · snacks
- High traffic, low conversion
- Students, 16-19h
- +14% vs last week
One data layer, three ways to use it
Stock, shoppers and shelf plans become one data set per store, then analytics you can act on. Everything here reads at group level and identifies nobody.
- Planogram compliance, bay by bay
- Empty slots surfaced while staff are still on shift
- Misplaced products, named by bay and SKU
- SKU-level availability and share of shelf against plan
- Fast and slow movers by bay, so range decisions have evidence
No two branches share an hour, or a season
Footfall is read per branch, per hour and across the year, so staffing, promotions and shelf depth follow the pattern of that store instead of a group average.
Footfall by branch & hour
Weekday average| 08 | 10 | 12 | 14 | 16 | 18 | 20 | 22 | |
|---|---|---|---|---|---|---|---|---|
| 1091 Talat | 08h: 22 | 10h: 34 | 12h: 62 | 14h: 44 | 16h: 51 | 18h: 88 | 20h: 96 | 22h: 40 |
| 0447 Sathon | 08h: 48 | 10h: 72 | 12h: 94 | 14h: 55 | 16h: 42 | 18h: 60 | 20h: 52 | 22h: 20 |
| 1220 Rangsit | 08h: 16 | 10h: 24 | 12h: 40 | 14h: 36 | 16h: 58 | 18h: 82 | 20h: 90 | 22h: 62 |
| 0863 Hua Hin | 08h: 30 | 10h: 42 | 12h: 50 | 14h: 62 | 16h: 74 | 18h: 70 | 20h: 55 | 22h: 28 |
| 1502 Khon Kaen | 08h: 20 | 10h: 30 | 12h: 46 | 14h: 34 | 16h: 40 | 18h: 66 | 20h: 72 | 22h: 34 |
- Quiet
- Steady
- Peak
Seasonal index · store 1091
JFMAMJJASOND
December runs 1.6x January here, and the peak hour shifts two hours later in the rainy season. Range and roster follow it.
One store, read three ways
Illustrative example: the same plan of store 1091, with a different read laid over it. Pick one.
SHELF BAYSBACK ROOMENTRANCESKU OUT OF PLAN2 SKUS BELOW MIN
Plan drawn from the store's own camera coverage.
Planogram & SKU
Every bay, every SKU, against plan
Each bay is read against your planogram at SKU level, so a product in the wrong place, a line below its minimum or a bay losing share of shelf shows up without anyone walking the aisle.
- Bays in plan
- 94%
- SKUs below min
- 12
- Share of shelf
- -6%
One data layer per store, built on the cameras you already have and the same engine as Z-Fraud.
See the case studyWhat this looks like across a store network
One live deployment, in the client's own numbers: where they started, what we put in, and what changed on the shop floor.
Live deployment · convenience retail · Thailand
How a 2,300-store chain halved its shrinkage
A major Thai convenience store chain knew roughly what it was losing every year. What it could not do was find those losses inside the footage: thousands of stores, a security team of a few dozen, and no way to tell which minute of which camera mattered.
Z-Fraud is now expanding store by store across the chain, reading every transaction against the video of that transaction. In each store that goes live, reviewers open a ranked list each morning instead of searching tape.
- Sector
- Convenience retail
- Estate
- 2,300 stores, nationwide
- Product
- Z-Fraud
- Deployment
- On-premise
What changed after go-live
50%
Projected annual loss cut
5×
More confirmed fraud found
10×
Faster to act on an event
Measured against the same security team's own case list before deployment.
11×
More precise on return fraud
Refund and return cases are where the gap between video and receipt is widest, and where the models gained the most.
Our process
Coverage check first
We read their existing cameras and told them which angles gave a usable read before anything was committed. Some of those cameras were over a decade old.
Retailer-specific incidents
There is no one-size-fits-all. Our forward-deployed team worked with their loss prevention leads to define the key incidents, then built the data foundation and train the models against proprietary footage.
A morning list, on-premise
Everything runs inside their own estate. The security team opens each flagged event with its footage and its receipt already lined up, confirms it, and closes it.
More detail on the estate, the rollout order and the internal reporting is being added to this page. Ask us for the full write-up in the meantime.