Computer vision in retail explained: how it works, seven real use cases, and the honest challenges. Plus a simple way to start with your current cameras.
Have you ever wondered how Amazon Go lets customers walk out without stopping at a checkout? Or why your supermarket sometimes still runs out of bread on a Friday afternoon even though the system says there are 40 loaves in stock? Or how large retailers can tell exactly which shelf is empty, in which store, at 3 p.m. on a Tuesday?
The answer to all three questions is the same: computer vision. It is one of the fastest-growing branches of AI in retail, and it is already running in stores you have probably visited this month.
Computer vision is a type of AI that teaches computers to understand images the way humans do. A camera captures a picture, and software analyzes what is in that picture. It can count products, spot gaps on a shelf, recognize a customer waiting too long in a queue, or notice that a display does not match the agreed layout. It does not need new cameras. Most systems work with the CCTV your store already has.
This guide walks through how the technology actually works, seven use cases you will see across modern retail, and the honest challenges that come with it.
How Computer Vision Actually Works in a Store
The pipeline is simpler than most people think. There are three steps.
First, cameras capture images or video, either from fixed positions or from moving devices like drones or shelf-scanning robots. Second, deep learning models (a branch of machine learning that works well with images) analyze those images and identify what they see. This is where the system decides “that shelf has an empty spot” or “this customer has been standing near the wine section for four minutes.” Third, the output triggers an action, which could be a notification to a staff member, an update in the inventory system, or a row added to a daily report.
The interesting part is that the same cameras can run multiple use cases at once. One camera over an aisle can track shelf gaps, count customer traffic, and flag suspicious behavior, all from the same video feed.
A Quick Word on Hardware
You do not always need fancy equipment, and in some cases you do not need cameras at all. Computer vision in retail runs on two types of hardware, depending on the use case. Fixed cameras — whether standard HD, specialized 3D, or compact edge devices — are suited for continuous monitoring throughout the day. They can track shelf gaps, count foot traffic, and flag suspicious behavior from a single feed. For shelf monitoring specifically, a mobile app used by store staff during a structured scanning routine can be just as effective, and far simpler to deploy. Staff walk the aisles, point the device at each shelf, and the app detects issues in real time. No cabling, no installation, no ongoing hardware maintenance. For retailers running large store networks, this flexibility matters. The right choice depends on what you are trying to monitor and how often you need to check it.
Seven Use Cases You Will Recognize
Below are the seven applications that show up most often in real retail deployments, along with which parts of a store they affect.
1. Shelf Monitoring
Knowing what is on the shelf — and what is missing — is the starting point for every other retail decision. The challenge is doing it consistently, across every aisle, every day.
The On Shelf Availability (OSA) solution from Digitalplace.ai approaches this through a structured morning scanning routine. Every day, store staff use a mobile app to scan the shelves before the store opens. The app analyzes the images in real time and flags three conditions:
- Out of stock — products missing from their shelf position, triggering an immediate replenishment prompt to staff.
- Misplaced products — items found outside their designated planogram position, whether moved by shoppers or mishandled during restocking.
- Unknown products — new SKUs or recently repackaged items whose barcodes are not yet registered in the system, flagged automatically for data team follow-up.
- Extra products — items displayed in quantities beyond the approved planogram, taking up space intended for other SKUs and weakening the intended shelf strategy.
The unknown product detection is particularly valuable. New launches and packaging refreshes happen constantly, and the gap between a product arriving in-store and its master data being updated can cause silent blind spots in reporting. The OSA app surfaces these before they distort inventory figures.
Because the scan happens each morning, store managers begin every shift with a clear picture of exactly which shelves need attention — not after sales have been lost, but before the first customer walks in. The app runs on standard mobile devices, which makes rollout practical even across large store networks without additional hardware investment.
For retailers who want continuous monitoring throughout the day, fixed-camera computer vision can complement the morning scan by catching gaps that open during peak hours. The two approaches address different parts of the same problem and work well together.
2. Planogram Compliance
A planogram is the layout agreement between a retailer and a brand. Computer vision can check whether products are placed according to that layout, how much facing space each brand gets, and whether promotional displays match the approved design.
A planogram is more than a shelf layout. It is the operational blueprint that determines where each product should appear, how many facings it should receive, how categories should be organized, and how promotional displays should look in-store. In practice, however, maintaining planogram compliance across dozens or even thousands of stores is difficult. Products are often placed in the wrong shelf position, displays become inconsistent, and stores gradually drift away from the approved layout without anyone noticing immediately.
This matters because planogram non-compliance is not just a visual issue. It directly affects product visibility, category performance, promotional execution, and the overall shopping experience. A product placed in the wrong position may appear unavailable to the customer even when stock is physically present in the store. A promotional item given too little space may underperform. On the other hand, products displayed with more facings than planned can take space away from other SKUs and distort the intended assortment strategy.
Digitalplace.ai approaches this problem through the On Shelf Availability solution’s shelf scanning workflow. As store teams scan shelves using a mobile app, the system can check whether products are positioned according to the approved planogram, identify misplaced items, and flag display quantities that exceed planogram rules. This gives retailers a practical way to monitor compliance as part of routine store operations, rather than relying only on occasional manual audits.
One of the biggest advantages of this approach is consistency. Traditional planogram checks depend heavily on staff walking the aisles, visually comparing shelves with printed guidelines, and documenting issues manually. That process is time-consuming and difficult to standardize across locations. By contrast, app-based image analysis helps stores run more structured morning inspections, surface deviations earlier, and document findings in a more repeatable way.
The value becomes even clearer in large store networks. Even when every branch receives the same merchandising instructions, local conditions, staffing limitations, and day-to-day restocking decisions often create variation in execution. Over time, those small differences accumulate into inconsistent customer experiences across stores. Automated planogram checking helps retailers maintain more uniform retail standards while giving field teams better visibility into where compliance is weakening.
This also improves speed of corrective action. When a planogram issue is detected early in the day, store teams can fix shelf gaps, reposition misplaced items, and rebalance display quantities before those issues begin affecting sales. Instead of discovering execution problems during a delayed audit or after a campaign underperforms, managers can respond while the issue is still small and operationally manageable.
In that sense, planogram compliance is not just about keeping shelves tidy or satisfying brand guidelines. It is a way to protect visibility, preserve the intended customer journey, and make sure merchandising strategy is actually executed on the store floor. When monitored consistently, it supports stronger product availability, more disciplined shelf space usage, and more reliable in-store execution across the retail network.
The system can give actionable recommendations based on the scan results, this actionable recommendations later will be used for the store man as the compliance tasks for current scanned shelf/rack.
3. Customer Counting and Heatmaps
Knowing how many people enter a store is useful, but it only tells part of the story. The more valuable question is what customers actually do once they are inside: which paths they take, which aisles attract attention, where they pause, and which areas they consistently ignore. Computer vision makes this measurable by turning foot traffic into spatial data rather than rough observation or anecdotal staff feedback.
That visibility helps retailers make better layout decisions with much less guesswork. If one section consistently receives strong traffic but low engagement, the issue may be assortment or display quality. If another area gets little traffic at all, the problem may be placement, signage, or adjacency to stronger categories. Heatmaps help teams see these patterns clearly, so store design becomes an ongoing performance decision rather than a one-time setup.
4. Queue Management
Long checkout lines are one of the fastest ways to damage the in-store experience. Customers who have already made their shopping decisions can still abandon baskets if the final step feels slow, crowded, or disorganized. Computer vision helps by monitoring how many people are waiting at each lane, how quickly queues are growing, and when intervention is needed before frustration builds.
This allows store teams to respond earlier instead of relying only on staff intuition or customer complaints. A system can alert managers to open additional registers, redirect staff, or rebalance checkout resources during peak periods. The value is not just operational efficiency, but protecting conversion at the final moment when a sale is closest to being lost.
5. Cashierless Checkout
Cashierless checkout is the most visible example of computer vision in retail because it changes the shopping journey itself. Instead of scanning items one by one at a traditional register, the system tracks which products a customer picks up and associates those actions with a virtual basket, allowing payment to happen automatically when the customer exits the store.
What makes this model important is not only speed, but friction reduction. Removing the checkout step can shorten trips, reduce queue dependency, and create a smoother convenience-led experience, especially in smaller formats or high-frequency shopping missions. At the same time, it requires strong coordination between product detection, identity tracking, and transaction accuracy, which is why execution quality matters as much as the concept itself.
6. Theft and Fraud Detection
Shrink rarely comes from a single source, and that is why it is difficult to manage with manual supervision alone. Some losses come from shoplifting, while others come from operational errors or cashier fraud, such as items being passed across the counter without being properly scanned or transactions being handled in ways that do not match normal patterns. Computer vision helps by narrowing attention to the moments that deserve review.
The goal is not to replace people with constant automated judgment. The real value is that the system can surface unusual behavior faster, reduce the amount of footage or activity that managers need to inspect, and make exception handling more targeted. That creates a more practical workflow for loss prevention teams, who often struggle with too much video and too little time.
7. Heat Analysis for Visual Merchandising
Visual merchandising has always aimed to shape attention, but without measurement it often depends too heavily on instinct. Retailers may know what display they intended to create, yet still have limited visibility into whether shoppers actually noticed it, paused in front of it, interacted with the featured products, or moved on without engagement. Computer vision closes that gap by making customer attention observable.
This matters because display effectiveness is not just about aesthetics. A fixture may look excellent from a brand perspective and still fail to influence behavior on the floor. By measuring dwell time, product interaction, and return-to-shelf behavior, retailers gain a clearer view of whether a display is driving curiosity, consideration, or conversion. That turns merchandising from a matter of taste into something much closer to measurable store execution.
The Market Is Growing Fast
The numbers give a sense of how quickly retailers are adopting this. Grand View Research estimates that the computer vision AI in retail market was around USD 1.66 billion in 2024 and will reach USD 12.56 billion by 2033.
A Deloitte survey in 2024 found that 68% of US retailers were either piloting or actively implementing computer vision. That is not an early-adopter number anymore. It is mainstream.
Comparing the Seven Use Cases
If you are trying to figure out where to start, a side-by-side view helps. The table below summarizes the seven use cases by complexity and where the payoff usually lands.
| Use case | Where it runs | Complexity | Main payoff |
|---|---|---|---|
| Shelf monitoring | Store aisles | Medium | Fewer out-of-stocks, faster restocking |
| Planogram compliance | Store aisles | Medium | Better brand deal execution |
| Customer counting | Entrance and floor | Low | Better layout and staffing decisions |
| Queue management | Checkout area | Low | Less cart abandonment |
| Cashierless checkout | Whole store | High | Labor savings, faster experience |
| Theft and fraud detection | Checkout and aisles | Medium | Lower shrinkage |
| Visual merchandising | Displays and endcaps | Medium | Better promotion ROI |
Most retailers start with customer counting or shelf monitoring because the complexity is manageable and the results show up in weeks, not months. Cashierless checkout is a much bigger commitment and usually comes later, if at all.
The Honest Challenges
Not everything about computer vision is smooth. Here are the issues we see retailers run into most often.
Lighting matters more than you think
Dim corners, reflective floors, and seasonal displays can confuse even well-trained models. A pilot that works in one store may need tuning for another.
Data privacy is a real concern
Even if you never identify individual customers, video analytics touches sensitive ground. You need clear policies and ideally on-premise or edge processing.
Integration takes effort
Getting alerts out of a camera is easy. Connecting those alerts to your replenishment system, your planogram tool, and your staff workflow is where most of the work happens.
Staff training is underrated
If your team does not trust the alerts, they will ignore them. Training and clear escalation rules are essential.
Upfront cost can look scary
Hardware, software, and integration add up. Most retailers see payback within 12 to 18 months, but the first bill is never small.
How We Approach It at Digitalplace.ai
We have been building AI solutions for retail since our R&D started in 2022, and shelf-level problems have been one of the first areas where our partners asked for help.
On Shelf Availability (OSA) gives store teams a structured daily scanning workflow via mobile app, detecting out-of-stock products, misplaced items, and unknown SKUs in a single morning routine — without requiring fixed cameras or additional hardware.
Inventory Tracking helps warehouse teams scan moved locations, confirm empty spots, and run full facility scans far faster than manual counts.
Interactive Kiosks use visual and voice recognition to create customizable avatar experiences that connect directly to client systems.
Every solution is designed to run on low-spec devices and in offline mode, because retail environments are rarely ideal.
What we have learned from working with partners running more than 20,000 minimarket outlets, 2,000 supermarket outlets, and 15,000 quick commerce locations is that the technology matters less than the implementation. The retailers who succeed are the ones with a clear problem, honest measurement, and a team willing to adjust based on what the data shows.
Where to Start If You Are Curious
A good first step does not require a big budget. Try this:
- Walk through one of your stores with a notebook. Write down the three biggest visible problems. Empty shelves. Long queues. Products in wrong spots. Whatever you see.
- Check your existing camera coverage in those areas. Most stores already have more cameras than they use.
- Pick one of the three problems and ask two or three vendors for a three-month pilot. Make sure you agree on what success looks like before starting.
- Measure honestly. Compare the pilot store with a similar store that did not get the system. If the numbers move, scale up. If they do not, ask why and adjust.
The best time to start is when you can define the problem in one sentence. Computer vision is not magic. It is a faster, more reliable way to see what is happening in your stores, and for most retailers that is already a huge upgrade on the current view.
If you want a second opinion on whether your stores are ready for this kind of pilot, feel free to reach out. We have walked through this question with enough retail teams to know where the quick wins usually hide.
References
- Grand View Research — Computer Vision AI in Retail Market Report
- commercetools — Computer Vision in Retail
- NVIDIA — AI in Retail and CPG Survey