Chapter 1: The store data foundation
Store data matches the shelf less than half the time
Store associates say the system matches the shelf less than half the time, and their trust in it barely moves until accuracy passes 70%
Every store runs on three versions of the truth.
In every store there is what the system says, what is actually on the shelf, and what the customer experiences when the two don’t match. When the versions disagree, it’s associates who close the gap by hand: correcting the record, hunting for the product, and dealing with the customers.
So we asked 2,200 of them on the frontline how often the system’s version of the truth held up at the moment they looked something up. Their answer: 46% of the time. That is accuracy as the floor and the customer meet it, not perpetual inventory accuracy. A store can report record accuracy above 90% and still fail half the lookups that matter to a customer standing in the aisle.
46%
On average, associates say the system matched the shelf 46% of the time when they looked something up.
1 in 4
Only one in four associates say system data is accurate more than 70% of the time.
:format(webp))
If I’ve learned one thing traveling around and working in stores across the country, it’s this: the store data we look at in HQ and what’s actually happening in the stores, from the products on the shelves to the processes we’ve put in place, are very, very different. To get this right is going to take some digging, more of us at HQ spending time in stores, and a feedback loop between both places that is prioritized, transparent and acted upon.
Anne Mezzenga, CEO and Founder, Retail Field Report
System vs shelf
| Accuracy | Associates |
|---|---|
| 0% | 3 |
| 1 - 10% | 11 |
| 11 - 30% | 22 |
| 31 - 50% | 20 |
| 51 - 70% | 18 |
| 71 - 90% | 17 |
| 91 - 99% | 7 |
| 100% | 1 |
| Country | Percent |
|---|---|
| Canada | 50 |
| France | 47 |
| Italy | 47 |
| UK | 47 |
| Global Mean | 46 |
| US | 46 |
| Poland | 45 |
| Germany | 44 |
Survey question: "Over a typical week when looking up stock data, what percentage of the time did the system accurately reflect what was on the shelf, in top stock or in the backroom?" Left: green bars represent half the time or less.
The 46% accuracy is an average. Most associates put the system at half right or worse. Only a small minority, around 8%, say it is accurate more than 90% of the time, which is where trust begins.
What's striking is how little this accuracy as reported by associates varies. Across the seven markets the average runs from 44% to 50%, and across the six sectors from 44% to 53%. This isn't one brand's problem, one country's or one sector's.
This is what associates experience every shift, but it isn’t where it ends. Every other system that relies on the inventory record inherits the same errors, and each one compounds the cost of the gap. The question covered the whole lookup, including whether the product was where it should be: on the shelf, in top stock, or in the backroom.
Associates asked what they’d change about their store
Get the stock counts actually right, so I’m not chasing ghost inventory.
Store Associate, US, Department Store
Improve inventory accuracy so the system more reliably reflects what is actually on the shelf and in the back room.
Store Associate, Canada, Grocery Supermarket
:format(webp))
Retailers have spent 30 years building systems that record what should have happened: what was received, sold, and ordered. Very few can see what actually happened on the shelf. That gap is where inventory distortion lives, in out-of-stocks and overstocks alike. If the record is right less than 90% of the time, every forecast, replenishment order, and AI agent built on it starts from the wrong number. Inventory accuracy isn’t a back-office metric anymore. It is the foundation for everything retailers want AI to do.
Greg Buzek, President, IHL Group
What Scandit has seen in stores
A good average hides empty aisles. In stores reporting on-shelf availability of around 95%, individual categories were as low as 70 to 84% once the shelf itself was captured.
The record was wrong where it mattered most: around 150 products on the shelf that the system showed as zero. A further 50 the system showed in stock weren’t on the shelf, most of them waiting in the backroom. Capturing the shelf directly put the expected availability gain at 1.3 points per store, most of it in the categories head office couldn’t see.
The only reason it isn’t worse
Only 1% of associates say the system always matches the shelf. When it doesn’t, most of the rest fix it. Seven in ten (72%) put it right in the system, either by correcting the record themselves (49%) or by raising a stock adjustment (44%), and many do both.
The 46% is the record as associates find it, after whatever fixes have already been made, not the raw system output. So the record that every report, replenishment order, and AI model depends on is being maintained by the associates on the floor: unasked, unmeasured, and often without the training to do it (more on that in chapter 5).
Not everyone catches it. Nearly a quarter (23%) work around the mismatch and carry on, rising to 31% in the US: they find the product or tell the customer it’s gone, and leave the record saying what it said. Every one of those is an error that stays in the system, and the next person to look it up inherits it.
49%
Associates correct the record themselves when it doesn’t match the shelf.
23%
Associates work around incorrect records and carry on.
Associates asked what they’d change about their store
Correct the stock myself straight away, instead of filling in slips of paper first.
Store Associate, Germany, DIY
Trust is a threshold, not a slope
Associates aren’t cynical about their data. They’ve taught themselves how to act on it based on real-world experience.
Associate trust in stock data, by accuracy
| Accuracy | Very or completely trust |
|---|---|
| Rarely | 27 |
| Sometimes | 25 |
| Under half | 25 |
| Over half | 34 |
| Mostly | 45 |
| Accurate 91-99% of the time | 75 |
Survey question: "How much do you trust the stock information provided by your store's systems?" Share of associates who trust their stock data very much or completely, grouped by how accurate they say it is.
Only a third (35%) trust it very much or completely, while 22% trust it slightly or not at all.
Set that against the accuracy they report and a pattern appears. Among associates who say the system is accurate half the time or less, around one in four trust it. That barely moves until accuracy passes 70%, then climbs sharply: above 90%, three in four trust it.
In other words, associate trust in data barely moves through the first half of the climb, then jumps. A retailer that lifts accuracy from 46% to 60% may see almost no change in what the floor believes, and it’s the floor’s belief that decides whether anyone acts on an alert. Chapter 6 returns to what this means for the business case.
The view from HQ. One thing that they would fix
:format(webp))
Improve data accuracy and consistency across systems, so that store operations and decision-making can rely on the same source of truth.
Supply Chain Lead, DIY
The customer pays for the gap
:format(webp))
The three versions meet at the shelf edge, in front of a customer, and when the record and the shelf disagree, the customer is often the one who finds out.
Associates spend around four hours a week answering customer questions they say store technology should have handled. That’s a cost in customer experience that no system records. The customer’s version of the truth is the one this research can’t see directly, and the one the other two exist to serve.
:format(webp))
AI is only as good as the data it learns from. If the starting point is store data that isn’t trusted by the people closest to it, retailers are simply amplifying inaccuracies. Imagine asking a doctor to treat patients using records that are only accurate 46% of the time. It wouldn’t be accepted; the doctor would find a workaround, and that is exactly what’s happening on the shop floor. This should be a wake-up call: retailers cannot afford cracks in the foundations.
Natalie Berg, Retail Analyst & Host, Retail Disrupted, NBK Retail
What this means for AI in store
On the floor’s account, the inventory record matches the shelf less than half the time, and it’s associates who correct it by hand. Any AI built on that record starts from the same place. See the store as it is before you ask a model to act on it.