AI Shelf Photo Analysis in the GCC: What's Actually Working (and What's Just Hype)

By Sufyan · 2026-07-29 · 5 min read

Last month I sat with a merchandising lead from a beverage brand in Dubai. He pulled up 340 shelf photos his team had taken that week across Carrefour, Lulu, and a bunch of smaller groceries in Deira. His question was simple: "Sufyan, how do I actually know if my planogram is being followed?"

He couldn't. Not really. His reps were snapping photos, uploading them, and then those photos were sitting in a folder that nobody had time to open.

That's the honest state of retail execution for most FMCG brands in the GCC right now. Photos everywhere. Insights nowhere.

Why shelf photos became such a mess

Field reps have been taking shelf pictures for years. It's the easiest KPI to enforce — snap a photo before you leave the outlet, prove you were there, tick the box. But somewhere between 2018 and 2022, brands started demanding more photos. Before, after, competitor shelf, promotional display, cooler, gondola end. A rep visiting 25 outlets a day could easily generate 100+ images.

And then what?

Honestly, I used to think the answer was hiring more merchandising auditors to review them. That's what most distributors in Saudi and the UAE were doing — a small team in the office scrolling through photos, spot-checking maybe 8% of them. The other 92% just piled up.

The math never worked. A merchandising auditor in Riyadh costs around SAR 6,500-8,000 a month. To meaningfully review the shelf photography from a 40-rep team, you'd need at least three of them. And they'd still miss things because human attention drifts after the 200th photo of the day.

So the photos kept getting taken. And ignored.

What AI shelf photo analysis actually does (the useful bits)

Here's the thing — planogram compliance AI isn't magic. It's a computer vision model that's been trained to recognize your SKUs and your competitors' SKUs on a shelf. You upload a photo, it draws boxes around every product it detects, counts facings, checks positions, and compares that to what the planogram says should be there.

When it works well, you get five things back within seconds:

At Zivni we built our AI shelf photo analysis around exactly this. Not because it's trendy but because our customers in Jeddah and Muscat kept saying the same thing: "we don't need more data, we need someone to tell us what the data means."

A sales ops lead at a snacks brand in Sharjah told me their compliance rate went from 62% to 81% in about four months. Not because their reps suddenly worked harder. Because the reps got a compliance score inside the app the moment they took the photo. If a planogram was wrong, they fixed it before leaving the outlet. That's the real shift — moving the correction from the office to the shelf.

Where it still falls flat

I'll be honest about this because too many vendors won't. AI shelf photo analysis has real limits, especially in GCC retail environments.

Lighting in traditional grocery stores (baqalas) is terrible. Half the models trained on European retail photos struggle when a rep is shooting a shelf in a small shop in Salmiya with one flickering tube light. Reflections on cooler doors mess up detection. Products stacked three deep — the AI can only see the front row, so your "count" is inherently limited.

Arabic packaging adds another wrinkle. A lot of general-purpose shelf recognition models were trained mostly on Latin-script SKUs. If your product has Arabic-only front-of-pack (common for local brands), you need a model that's been specifically trained on your catalog. This isn't optional. It's the difference between 94% accuracy and 68% accuracy.

And promotional clutter in Ramadan? Forget it. When shelves get wrapped in gondola stickers and stacked with dump bins, even the best models get confused. You still need humans reviewing edge cases.

So if a vendor tells you their AI is 99% accurate across every store type in the GCC, they're either lying or they've never actually deployed in a baqala in Al Ain during peak Eid rush.

What good deployment looks like

The brands getting real value from this aren't the ones with the fanciest AI. They're the ones who did three unglamorous things:

First, they built a proper SKU library. Every product photographed from multiple angles, in different lighting, with and without promotional sleeves. This takes weeks. There's no shortcut.

Second, they set realistic KPIs. Not "100% compliance" — that's fantasy. Something like "85% planogram adherence in modern trade, 70% in traditional trade, tracked weekly." Achievable numbers that reps can actually influence.

Third, they tied the AI output to something the rep cares about. Gamification, bonuses, weekly leaderboards. Because a compliance score that just sits in a dashboard changes nothing. A compliance score that determines who wins the quarterly incentive changes everything.

One of our customers running distribution across UAE and Oman ties 22% of their field sales bonus directly to AI-verified shelf metrics. Their reps check the app after every visit to see their score. Behavior changed inside three weeks.

Where this is heading

The next 18 months in GCC retail execution FMCG are going to look different. Shelf photo analysis is becoming table stakes — if you're not doing it by end of 2025, your competitors will have a data advantage that's hard to close.

But I don't think the winners will be the brands with the best AI. They'll be the brands that figured out how to make the AI output land in the hands of a rep standing in front of a shelf, at the exact moment they can still do something about it.

That's the whole game. Everything else is dashboards.

If you're evaluating field sales automation GCC options and want to see how we've built this for FMCG merchandising software specifically, come poke around zivni.com — happy to walk you through what a real deployment looks like in a Saudi or UAE distribution environment. What's your current shelf compliance rate looking like?

© 2026 Zivni. All rights reserved.