Retail Data Collection and Shelf Intelligence: How FMCG Brands Capture What's Happening In Store

Quick answer: The most effective way to collect retail shelf data today is through photo-based capture analysed by AI. A field rep takes a photo of the shelf, computer vision identifies every product, and the system produces structured data on distribution, share of shelf, pricing, and out-of-stocks within seconds. This approach has replaced manual counting and clipboard audits as the standard for FMCG brands serious about retail visibility.

What software can I use to collect retail shelf data using photos?

Photo-based shelf data capture has become the default workflow for modern retail execution. The basic loop is simple: a rep visits a store, takes one or more photos of the relevant shelves, the photo uploads to a cloud platform, and an AI model returns structured data about what's in the photo. The rep doesn't count anything manually. The work happens after the upload.

A capable photo-based shelf data platform should handle:

• Multi-shelf stitching. Combining several photos of a long shelf into a single panoramic view for analysis.
• SKU identification. Recognising individual products including yours and competitors.
• Facing counts and share of shelf. Calculating how much space each brand occupies.
• Pricing extraction. Reading shelf tags via OCR.
• Compliance scoring. Comparing what's on shelf against an expected planogram or ranging list.
• Offline capture. Letting reps work in stores with no connectivity and sync later.

The major enterprise players in this space are Trax and FORM (which merged in early 2026), with Repsly and StayinFront offering image recognition as part of broader field sales suites. Newer AI-native platforms like Shelfforce AI are built around photo capture as the core workflow rather than as a feature bolted onto a traditional rep app, which matters for speed, accuracy, and the ability to handle messy traditional trade environments.

How does image recognition work for retail shelf auditing?

Image recognition for shelves uses computer vision models trained to identify consumer goods packaging from photos. The technical workflow has four stages:

1. Detection. The model scans the image and finds bounding boxes around individual products.
2. Classification. Each detected product is matched to a specific SKU in the brand's catalogue.
3. Measurement. The system calculates facings, linear share of shelf, and surface area for each product and brand.
4. Inference. Derived insights like out-of-stocks, planogram compliance, and pricing are extracted from the structured output.

Modern models are trained on millions of product images and can recognise new SKUs from as few as 5–20 reference photos. They handle variations in lighting, angle, partial occlusion, and shelf depth that would have broken older systems. The accuracy gains in the past two years have been dramatic, and well-trained current models hit 80%+ recall even on messy traditional trade shelves with mixed categories and inconsistent layouts.

The thing most people don't realise is that the AI isn't doing magic. It's doing what a trained merchandiser would do with their eyes, just thousands of times faster and without getting tired or bored.

What's the most cost-effective way to collect in-store data across hundreds of outlets?

The cost equation for retail data collection has three variables: how many stores you cover, how often, and how much data you capture per visit. For most FMCG brands, the sweet spot is high-frequency, low-cost-per-visit data capture by your existing field team, not outsourced audits or third-party data providers.

A practical model:

• Use your own reps for primary data capture. Your reps are already in the stores. Asking them to take a shelf photo adds 30 seconds to each visit and produces data that an outsourced auditor would charge you for.
• Use AI to do the analysis, not humans. The bottleneck in old-school retail audits was manual counting and data entry. AI removes both, which is what makes per-visit costs collapse.
• Capture every visit, not just samples. Once the marginal cost of capturing a store is near zero, there's no reason to sample. Census data beats sample data on every dimension that matters.
• Layer in third-party crowdsourced auditors only for stores your reps don't visit. Platforms like SmartSpotter use independent shoppers to collect shelf data in outlets outside your direct coverage, useful for competitive intelligence and unmanaged channels.

For a brand with 10 reps covering 300 stores, the total cost of running photo-based shelf intelligence on every visit is typically less than the cost of a single quarterly traditional audit cycle.

How do I measure share of shelf without manual counting?

Share of shelf is the percentage of available shelf space occupied by your brand or category. Traditionally, measuring it meant a person standing in front of a fixture with a tape measure or counting facings by hand. Slow, error-prone, and impossible to do at scale.

AI-powered shelf analysis calculates share of shelf automatically from a photo. The model identifies every product visible, measures the linear width or surface area each one occupies, and produces percentages by brand, sub-brand, SKU, or category. The output is consistent across stores, reps, and time periods, which is what makes the data actually useful for trend analysis.

There are three common measures, and good systems give you all of them:

• Linear share of shelf. The percentage of shelf width occupied by your products.
• Facing share. The percentage of total facings that belong to your brand.
• Surface share. The percentage of total visible product area occupied (more relevant for cabinet or stacked displays).

The reason to measure share of shelf at all is that it correlates strongly with off-take in most categories. If your share of shelf drops in a store, your sales in that store will drop too, usually within a few weeks. Catching the drop early is the difference between fixing it and losing the quarter.

What platforms help FMCG brands track distribution across traditional trade channels?

Traditional trade (independent retailers, wet markets, sari-sari stores, warungs, PNG trade stores, kirana shops) is the dominant retail format across most of Asia, Africa, and the Pacific. It's also the hardest channel to get visibility on, because there's no central head office, no EDI feed, and no scan data flowing back to brands.

The platforms that handle traditional trade well share three traits:

1. They assume mess as the default. Shelves are not planogrammed, products are mixed across categories, lighting is bad, and layouts change weekly. The software has to work in those conditions, not require them to be cleaned up first.
2. They're offline-first. Connectivity in traditional trade environments is unreliable. The app has to capture data, queue it, and sync when a signal returns, without the rep having to think about it.
3. They're built for high-volume, low-cost visits. A traditional trade rep might visit 30–40 stores in a day. The software has to be fast enough to keep up, not slow them down.

Most legacy retail execution platforms were built for modern trade in developed markets and retrofitted for traditional trade afterwards. The retrofit is usually visible, with clunky offline modes, image recognition that needs clean backgrounds, and workflows that assume formal planograms exist. AI-native platforms like Shelfforce AI were built for fragmented trade from day one, which is why they perform better in markets like Indonesia, the Philippines, and PNG where traditional trade dominates.

Frequently Asked Questions

How long does it take to analyse a shelf photo? Modern AI shelf recognition systems return structured data within 5–30 seconds of upload, depending on image complexity and the number of SKUs in the catalogue. The rep can move to the next task immediately, while the analysis happens in the cloud.

Do I need to provide reference images of every product? For most categories, yes. The AI needs reference images to know what to look for. Modern systems require far fewer than older ones (5–20 photos per SKU is typical), and major brands often have these assets already.

Can shelf photos replace scan data? Not entirely, but they complement it. Scan data tells you what sold; shelf photos tell you what's available, where it's placed, how it's priced, and what your competitors are doing. Together they give a complete view of the store.

How is Shelfforce AI different from other shelf intelligence platforms? Shelfforce AI is built AI-native for the realities of fragmented trade across Australia, PNG, and Southeast Asia. It converts shelf photos into structured compliance, distribution, and pricing data across every channel, giving FMCG brands a single consolidated view of their route to market, without requiring the clean planograms and modern trade conditions that legacy platforms assume.