AI Shelf Photo Analysis in the GCC: What's Actually Working for FMCG Brands

Last month I sat in a distributor's warehouse in Sharjah watching a supervisor scroll through 340 shelf photos on his laptop. He'd been at it for two hours. His job? Check if the merchandisers actually placed products correctly at 47 outlets that morning. He was 12% through the pile.

That's the problem AI shelf photo analysis is quietly solving across the GCC. And honestly, it's the feature at Zivni that surprised me most — not because it's flashy, but because of how quickly brand teams stopped treating it as a nice-to-have.

Let me explain what's actually happening on the ground.

The old way was theatre, not execution

Here's how retail execution used to work for most FMCG brands in the region. A rep visits an outlet in Riyadh or Muscat. He fills a checklist. He ticks boxes: shelf share okay, planogram followed, competitor pricing noted. He moves on.

The supervisor trusts the checklist. The brand manager trusts the supervisor. Somewhere in that chain, 30-40% of the data is either guessed, copy-pasted from yesterday, or filled in from the car park outside the store.

I'm not saying reps are dishonest. Most aren't. But when you're covering 22 outlets a day in Dubai traffic, corners get cut. That's human.

The fix isn't more checklists. It's a photo — and software that can actually read it.

What AI shelf photo analysis actually does

A rep opens the Zivni app at an outlet. Snaps a photo of the shelf. Within about 8 seconds, the system tells them:

• Which SKUs are present, by count
• Which competitor products are eating your facings
• Whether the planogram matches head office guidelines
• Out-of-stock alerts on your priority SKUs
• Share of shelf as a percentage

The rep sees this on their screen before they leave the store. So does the supervisor. So does the brand manager sitting in a Jeddah office watching a dashboard.

And here's the part I got wrong at first — I used to think the killer feature was the analytics dashboard. It's not. The killer feature is that the rep sees the result immediately and can fix the shelf right there. Restock. Rearrange. Talk to the store owner. Before walking out.

That one shift — from reporting after the fact to correcting in the moment — is what changes retail execution in FMCG.

Why the GCC is a weird, wonderful market for this

GCC retail isn't like the US or UK. You've got hypermarkets like Lulu and Carrefour sitting next to tiny baqalas run by one guy who's been there 23 years. Modern trade and traditional trade in the same distributor route.

AI shelf photo analysis works differently across these. In a Lulu, the planogram is strict, the shelves are tall, and compliance is measurable to the millimeter. In a baqala in Ajman, there's no planogram — the goal is just visibility, price-tag accuracy, and beating the competitor who's slipping cash to the shopkeeper for prime shelf space.

Good merchandising software has to handle both. That's something FieldAssist and BeatRoute understood early for the Indian market, but the GCC has quirks — Arabic-language pack recognition, seasonal SKUs during Ramadan that spike and vanish, and distributor structures where one company might handle 6 brands in Oman and just 2 in Bahrain.

We learned this the hard way. Our first version of shelf recognition worked beautifully on English packaging and struggled with certain Arabic script rotations. Took us four months and a lot of retraining data from a beverage client in Kuwait to fix it properly.

The brands winning at retail execution in the GCC right now aren't the ones with the biggest field teams. They're the ones who figured out that a rep with a camera and decent AI beats a rep with a clipboard every single time.

The numbers that actually matter

I'll skip the marketing stats. Here's what we see from real deployments:

One dairy client in the UAE went from checking planogram compliance at 18% of visits (through manual audits) to 94% of visits (through automatic photo capture at every stop). Their out-of-stock detection time dropped from 3 days to same-day.

A snacks brand in Saudi cut their trade marketing spend by 21% in the first quarter — not because they spent less, but because they finally knew which displays were actually being executed versus which ones were being paid for but ignored.

A personal care company in Pakistan (yes, technically not GCC but same regional playbook) found that 34% of their premium SKUs were being placed on lower shelves in traditional trade, well below the eye-level agreements they'd paid the retailer for. They renegotiated. Sales in that segment went up 11 points in five months.

That last one still surprises me. The brand had suspected it for years. Nobody could prove it. Photos and AI proved it in three weeks.

What to watch out for before you buy

A few honest warnings if you're evaluating AI shelf photo analysis tools — whether it's us, or someone else.

First, ask about training data for your specific SKUs. Generic shelf recognition is close to useless. If a vendor can't recognize your top 40 SKUs with 90%+ accuracy in a pilot, walk away. We usually spend the first two weeks of any deployment just training the model on client packaging, including regional variants.

Second, check offline mode. Half the outlets in Oman and interior Saudi have patchy connectivity. If the app can't capture photos, run analysis locally, and sync when back online, you'll lose data every single day.

Third — and this is where most implementations die — align the KPIs before you deploy. If you tell your reps "take shelf photos" without tying it to their commission, gamification points, or supervisor scorecards, adoption crashes to 20% within six weeks. I've watched it happen.

The tech is the easy part now. The change management is where the real work is.

Anyone want to talk through what this looks like for your brand specifically? Drop me a note through zivni.com — happy to show you what a real deployment looks like, warts and all.