Chapter 6: What good looks like
The technology most stores run today gets you to half. Readiness gets you the rest
More devices lift accuracy to about half, then stop. Trust only arrives once the data is close to right. Asking the floor, and testing with them, moves both. Stopping halfway is the most expensive place to stop.
Three things in this research point the same way.
Stores with more technology report more accurate data, but only up to about half. Associates’ trust barely moves until accuracy is well past that, so a retailer improving from 46% to 60% may see almost no change in what the floor believes. And trust and adoption rise with every step a retailer takes toward its associates: higher when they are asked, highest when they were part of the testing process.
The retailers starting to close the gap between the record and the store are doing two things. They are giving associates a true picture of the store, captured at the shelf rather than inferred from the record. And they are introducing it with their people rather than to them.
~50%
Reported data accuracy in stores running four or five of the store technologies. It stops climbing after that.
17%
Work in a store with technology that captures the shelf itself: shelf-scanning robots or fixed cameras.
Technology alone won’t close the gap
Devices help, then stop helping.
More store technology lifts accuracy, but only to about half
| Technologies in use | Accuracy |
|---|---|
| None | 30 |
| One | 42 |
| Two | 44 |
| Three | 48 |
| Four | 50 |
| Five | 52 |
| Six | 48 |
| Seven | 50 |
Average of how often associates say the system matched the shelf when they checked, by how many of seven store technologies they say their store uses.
We asked associates which of seven store technologies their store runs: handhelds, work apps on their own phones, shared devices at fixed points, electronic shelf labels, self-checkout and smart carts, shelf-scanning robots or cameras, and RFID.
The more of them a store has, the more accurate associates say the data is, up to about half. Then the line flattens.
More of the same technology does not get a store past half, and only 8% of associates put their data above 90%, which chapter 1 showed is where trust arrives. The reason is in what that technology is pointed at. A handheld, a shelf label, or a shared terminal reads and writes the inventory record. If the record is wrong, the device carries the error faithfully, and faster.
Of the seven, only shelf cameras and shelf-scanning robots do autonomous checks of the shelf rather than the record, and only 17% of associates work in a store that has them.
The same device in an associate’s hand becomes a different tool the moment it is pointed at the shelf: it stops inheriting the record and starts correcting it. That is the step the stores at the top of the accuracy range have taken, and the one most of the technology in this survey has not.
Associates asked what they’d change
An easier way to do inventory check than counting each item and putting the number in the system to see if they match.
Store Associate, Canada, Grocery Supermarket
What Scandit has seen in stores
In one retail store deployment, once the daily list held only the gaps associates could actually fill, with stock waiting in the backroom, the share of those gaps closed each day rose from around 56% to 78% with no change in effort.
Most gaps were one-offs: 71% of flagged products appeared on only one day. A small core kept returning, mostly slow-moving non-food, and that’s where a facing change or a one-time fix paid off rather than another walk.
The trust threshold
Chapter 1 showed that trust is a threshold, not a slope.
Flat at about one in four until accuracy passes half
| Reported accuracy | Trust |
|---|---|
| 1 - 10% | 27 |
| 11 - 30% | 25 |
| 31 - 50% | 25 |
| 51 - 70% | 34 |
| 71 - 90% | 45 |
| 91 - 99% | 75 |
Survey question: “How much do you trust the stock information provided by your store’s systems?” Share answering very or completely, by how often associates say the system matched the shelf. Associates who said 'never' are not shown.
Around one in four associates trust the data until accuracy passes 70%, and three in four trust it above 90%.
This chapter is about what that means for the business case. Getting better hurts before it helps. A retailer that lifts accuracy from 46% to 60% has paid for the improvement and the floor still behaves as if nothing has changed, so the hours spent reconciling by hand don’t come back yet.
That is where pilots stall. It is also why half right is the most expensive kind of wrong: the retailer is paying for the technology and still paying for the reconciliation.
The case for the last part of the climb is easier than the case for the first, and the retailers who get there plan for both from the start.
Ask the floor, then scale what the pilot taught.
Trust rises when associates are asked and rises most when they are part of the test; take what the pilot stores learned to every store that didn’t get to try it.
Trust feeds the record too. Associates who trust their stock data (very or completely) are more likely to put it right in the system when it is wrong: 76% correct the record or raise an adjustment, against 66% of those who trust it slightly or not at all.
They also use more of what they are given, with 66% still using most or all of their new tools against 52%. So a more accurate record earns trust, and trust keeps the record accurate.
76%
Associates who trust their stock data (very much or completely) correct the record or raise adjustments 76% of the time
66%
Associates who trust their stock data (slightly or not at all) correct the record or raise adjustments 66% of the time
:format(webp))
I am not surprised by how slowly trust moves. Until the data is right more than half the time, only about one in four associates trust it, no matter how hard the team has worked to get it there. Most people don't believe it until it's right nine times in ten. So a retailer can do a lot of good work, go from mostly wrong to half right, and the floor won't feel much different yet. That's usually when a pilot gets quietly parked, and it's exactly when I'd suggest you keep going, with the people on the floor inside the test rather than watching it from afar and waiting for the results you had in that fancy presentation to come true.
Andrew Smith, Managing Partner and co-founder, ThinkUncommon
Scandit Insight
The stores at the top of the accuracy range in this survey are not running more technology than the rest.
They are looking at a different thing. In the deployments Scandit has observed, three habits separate them.
The operators whose availability stays high are the ones who can tell those cases apart before anyone sets off down the aisle.
One: They measure availability at the shelf and by category, not as a store average that hides an empty aisle.
Two: They sort gaps into the few that keep recurring, usually slow-moving non-food where a facing change or a one-time fix pays off, and the many that are one-offs.
Three: And they treat phantom stock, where the system shows units but the shelves and backroom are empty, as a data correction rather than another walk.
Associates and HQ both want a picture of the store that is captured where the stock is, and arrives before the decision is made.
Not just working through lists, but the actual gaps.
Store Associate, Germany, Grocery Supermarket
Camera systems that detect gaps on the shelves.
Store Associate, Germany, Grocery Supermarket
Make capture real-time and sync with distribution centres.
Retail IT Director, Grocery
:format(webp))
The retailers that win with AI over the next three years won't be the ones that buy the most models. They'll be the ones that fixed their data first. Agentic AI will make decisions faster than any person can review them, and that only helps if the data underneath is right. Get a true, current picture of the shelf, put it in your associates' hands, and then let AI act on it. Do it in the other order and you've just automated your mistakes.
Greg Buzek, President, IHL Group
Five actions before you lock your AI budget
1.
Measure accuracy where it matters
Measure at the shelf, at the moment of lookup, not only in the record.
2.
Cost the reconciliation.
Put the hours associates spend making up for the data on a line item, so the investment case starts from the real baseline.
3.
Bring the shadow store into the record
Give associates one way to capture what they already know, and make it faster than the notebook.
4.
Pilot with associates, don’t just consult them
Being part of the test is what moves trust and adoption. Ask the floor and scale with the lessons from the pilot.
5.
Fix the data before the model
An AI tool is only as good as the record it reads, and associates say that record matches the shelf less than half the time.
In their words
None of what associates want is a request for an AI model. But it is a request for a record the shelf agrees with, and for people with the time, the tools, and the reason to keep it that way.
What associates would change about shelf and inventory management
| Change | Percent |
|---|---|
| A better organized store | 22.5 |
| Accurate, up-to-date stock information | 19.3 |
| Better technology and equipment | 17.1 |
| More staff, time and training | 14.2 |
| The right amount of stock on shelf | 11.9 |
| Better communication and management | 10.9 |
| Clear, correct prices and labels | 7.5 |
Survey question: “If you could change three things in your store, what would they be in relation to shelf and inventory management?” Open answers grouped into themes by Scandit. Share of all 2,200 associates who mentioned each theme; associates could mention more than one.
The second half of that is readiness, and it is where the AI budget should start.
Every finding in this report came from a question we chose to ask. The last word goes to one we left open.
Asked what three things they would change about shelf and inventory management, 2,200 associates gave 4,912 answers.
What they asked for, in every market, was a store that makes sense and stock information they can believe: a better organized store (22.5%), accurate and up-to-date stock information (19.3%), technology that works (17.1%) and more staff, time and training (14.2%).
Fewer than 4% mentioned automation or AI at all and that is the point. Associates describe the outcome, not the mechanism. A store that makes sense, stock data that is right, and tools that work is what store AI exists to deliver.
The floor is asking for it in its own words, and the first three asks are the ones a system that can see the shelf answers directly.
Associates asked what they’d change
Automatic notification on the handheld device when an item on the shelf has run out.
Store Associate, US, Department Store
I’d like to make sure people who are doing stocktake actually care about getting the numbers right. When they do a bad job it causes so many issues.
Store Associate, UK, DIY
Stock to be updated when sold or scanned at tills.
Store Associate, UK, Department Store
That we can have a view of the stock in real time.
Store Associate, France, Department Store
Introduce a way to change stock levels quickly and efficiently.
Store Associate, Poland, Grocery Hypermarket
I would replace the poorly working devices used to do the inventory.
Store Associate, Italy, Grocery Supermarket
How ready are you?
The survey measured readiness as associates experience it: whether they were shown why the work matters, trained on the tools, and part of the test. Andrew Smith’s readiness assessment measures the same three things as an organization sets them up: purpose, people and process. It names your pattern and the one move to start with.