When the shelf says one price and the register says another, the customer says fraud.

The Problem

A typical morning in a store: a manager receives an update with twenty new prices. A store associate has to find each product, print a new price tag, remove the old one, put up the new one. Add a dozen other tasks competing for attention, and price-tag updates fall behind. Customers in the store choose products based on the prices they see on the shelf, and discover a different price at checkout.

To customers this is fraud. To regulators, this is a fineable offence. To store managers, it’s a battle they cannot win at scale: a typical supermarket carries 3,000-5,000 products, and even at one tag per minute (which is unrealistically fast), verifying every tag fills a workday.

Manual checks compound the problem. Employees get tired, lose focus, rush, and make more mistakes than they catch. When a discrepancy turns up, the employee logs it, finds the responsible person, and follows up on the fix. That workflow takes longer than the original task.

Network-level, the data is even worse: no aggregated picture, no analytics by category or location, no ability to spot the systemic problems behind individual discrepancies.

What the Solution Does

A monitoring system that reads every visible price tag in your store, matches it to the price master, and flags every discrepancy in real time. One camera setup also detects promo material compliance: is the promo poster where it should be? does the discount on the tag match the campaign?

1. Capture, cameras photograph shelves on a schedule (often the cameras already running for shelf monitoring).
2. Locate, detect products and price tags in each frame.
3. Read, OCR reads the price (and any promo markings) from each tag.
4. Match, link each tag to its product (not always trivial, tags shift, products move).
5. Validate, compare tag price against the price master and promo schedule.
6. Alert, discrepancies become real-time alerts to store staff.
7. Analyze, patterns across stores, categories, and time inform process improvement.

Where It Fits

This makes sense if you…

• Operate a multi-store retail chain with frequent price changes (weekly or daily)
• Have promo campaigns with specific in-store material requirements
• Have measurable cost from customer complaints or regulator fines on price compliance
• Want network-level visibility into which stores have which compliance issues
• Already operate camera infrastructure or are deploying it for adjacent use cases

This is probably not the right time if you…

• Have very stable prices that change rarely, manual checks suffice
• Use electronic shelf labels (ESL) for most of your assortment, the price-tag problem mostly disappears
• Don’t keep a single price master that reflects what should be on the shelf (fix that first)
• Operate at small scale where the cost of CV exceeds the cost of one extra audit walk

Business Value

Audit time savings. Manual price audits in a 3,000-SKU store take a full workday. Automated monitoring removes that walk and surfaces only the discrepancies that need action. Manual audit time typically drops by 70-90%.

Customer complaint reduction. When the shelf and the register agree, customers stop complaining about pricing. Price-mismatch complaints typically fall by 40-70%. The range varies sharply with your baseline: a chain with a complaint-driven culture sees larger gains than one already running tight.

Regulator fine reduction. Some jurisdictions enforce price-tag fidelity by inspection: most of the EU, several US states, several Asian markets. The fines there are meaningful. Continuous monitoring catches violations before inspectors do. The fine-reduction range depends on your starting compliance rate and the local regulator’s posture, so we measure it during pilot.

Network analytics. Which stores have the most price-tag issues? Which categories see the most discrepancies? Which days of the week have the most violations? Surface-level numbers come from existing audit data; the systemic patterns only appear with continuous, complete monitoring.

How It Works

The architecture sits on top of the camera and CV stack already used for shelf monitoring. It adds two layers: an OCR layer that reads price tags, and a validation layer that compares tags against the price master.

1. Capture and detection

Cameras photograph shelves on a schedule. The system detects both products and price tags in each frame. Detecting price tags is its own problem: they can be hanging, glued, shelf-edged, electronic, or hand-written. The base model handles common formats. Unusual formats need a short data-collection sprint.

2. OCR, reading the price

Once tags are located, an OCR component reads the price, any unit indicators (“/kg”, “/each”, “/100g”), promo markings, and other text. We combine a general OCR backbone (Tesseract or a fine-tuned variant) with a custom model trained on your specific tag templates. On standardized formats accuracy reaches 95%+. On irregular or hand-written tags it runs lower.

3. Tag-to-product matching

The non-trivial part. Tags shift relative to products, products get moved, new tags go up before old ones come down. To assign each tag to its most likely product, the system uses spatial logic, shelf-edge templates, and the product recognition output from the shelf-monitoring layer. Ambiguous cases get flagged for review. The system does not guess them.

4. Validation

Tag price compared to price master. Promo markings cross-checked against the active promo schedule. Discrepancies typed by severity (wrong price, missing tag, expired promo material still displayed, etc.).

5. Alert and analyze

Real-time alerts ship to whatever channel store staff already monitor: Slack, Teams, or the store-management app. Analytics roll up to network-level dashboards for ops and category teams.

Stack

YOLO-family detection (shared with shelf monitoring). Tesseract with custom OCR fine-tuning. Datapipe for retraining. Python and FastAPI for the validation service. Integration with PIM or ERP for the price master. Metabase for analytics.

What You Need to Make This Work

Data. Price master that’s actually maintained, promo schedule with start/end dates and required materials. SKU catalogue with reference images (shared with shelf monitoring). Reference photos of your typical tag formats (we collect these during the test phase if you don’t have them).

Integrations. Read access to your PIM / ERP for the price master. The shelf-monitoring infrastructure (cameras, CV stack). Alert delivery (Slack / store app / etc.).

Hardware. Existing CCTV often works, but price-tag OCR puts higher demands on resolution and angle than product recognition. Some shelves may need additional or repositioned cameras. We assess during the test phase.

Team. A store-ops lead for alert response. A pricing-ops lead who works the analytics. A merchandising contact for promo material verification. PIM or ERP integration takes about 15-25 hours during pilot.

Implementation Roadmap

1. Test (2-4 weeks)

The test phase often piggybacks on shelf monitoring. We pick 1-2 stores and validate tag detection, OCR accuracy, and matching logic on your specific tag formats. We measure baseline price-tag accuracy against the master. Output: a written report with per-format OCR accuracy and recommendations for camera positioning or tag format standardization.

2. Pilot (1-2 months)

Roll out across 3-10 stores. Tune the validation rules with pricing-ops. Wire up the alert workflow. Build the analytics dashboards. Output: a production deployment, a measured complaint and audit-time reduction, and a go/no-go on full rollout.

3. Scale (3-6 months)

Network-wide rollout. Quarterly OCR retraining tied to tag format changes. Periodic recalibration as your promo cadence evolves. Your team owns the day-to-day; we stay on for retraining and edge cases.

Keep in Mind

Limits:

• OCR on unusual tags is harder. Hand-written, partially obscured, or weathered tags drop accuracy. The system flags low-confidence reads and leaves them for a human to read. You keep a manual workflow for those cases.
• Tag-to-product matching can be ambiguous. When tags shift or products move, the system has to make a judgement call. We surface uncertain cases for review. The system does not fire a false alert on them.
• The price master is the source of truth. If your PIM or ERP master isn’t current, the system flags the master, and the shelf is fine. This is sometimes the actual finding: the pricing process is broken upstream.
• Promos with creative material need explicit definitions. “There should be a poster on this end-cap” is verifiable. “The promo should look festive” is not. We work through these definitions during pilot.
• Electronic shelf labels (ESL) are an alternative. If your chain has ESL coverage, the price-tag problem is largely solved upstream. CV monitoring still helps for the non-ESL portion: promo materials, hand-written tags, edge cases. The ROI shifts, so we work through it during the pilot scope conversation.

FAQ

Can we use this for promo material compliance as well as prices?

Yes. The one detection layer can verify promo posters, end-cap displays, sale stickers, and other in-store campaign materials. Validation rules are configurable per campaign.

How accurate is OCR on price tags?

On clean, standardized, well-lit tags: 95%+. On hand-written, partial, or unusual formats: typically 80-90%. We measure per-format during the test phase and surface explicit numbers.

What about electronic shelf labels (ESL)?

If you have ESL for most SKUs, the price-tag problem is mostly solved by the ESL system itself. CV monitoring still helps for promo material verification and for non-ESL items. We work through the ROI when ESL is in play.

How does this integrate with my existing audit process?

The system replaces the manual audit walk for price-tag verification. Your existing audit process for other checks (cleanliness, staff presence, etc.) remains. Many chains keep a reduced manual audit as a quality check on the CV system itself.

Ready to Discuss?

If you operate a multi-store chain with frequent price changes, and price-mismatch complaints or audit overhead are a real cost line, this is a high-ROI deployment. We’ll walk through your current audit process, your price-master quality, and your tag formats, then tell you what to expect from a pilot.

Discuss your project