AI Shelf Photo Analysis in the GCC: What Actually Changes on the Ground

Last month I sat with a trade marketing head in Dubai who told me his reps had submitted 14,000 shelf photos in Q3. Nobody had looked at more than maybe 400 of them. The rest sat in a folder somewhere, quietly rotting.

That's the dirty secret of retail execution in the GCC right now. Everyone's collecting shelf photos. Almost nobody is actually using them.

And honestly, I get it. If you're a regional sales manager covering 600 outlets across Dubai, Sharjah, and the Northern Emirates, when exactly are you supposed to review 3,000 photos a week? You can't. So the photos become theatre — proof that the rep visited, nothing more.

This is the gap AI shelf photo analysis is finally closing. Not in some futuristic 2030 way. Right now, this quarter, in stores from Lulu Hypermarket in Abu Dhabi to Panda in Riyadh to imtiaz in Karachi.

What the AI is actually doing (and what it isn't)

Let me be honest about what this technology does well and where it still stumbles.

When a rep takes a shelf photo at a Carrefour in Riyadh, the AI does a few things in under 8 seconds. It counts your SKUs. It counts competitor SKUs. It measures your share of shelf as a percentage. It checks whether your planogram matches what head office agreed with the retailer. It flags out-of-stocks. It reads price tags if the lighting is decent.

What it doesn't do well yet? Weird angles. Reflective glass coolers where you can see the rep's reflection more than the products. Very small SKUs stacked on top shelves. We've gotten our accuracy at Zivni to around 94% on standard shelf setups, but drop it to a chest freezer with frost on the glass and we're closer to 78%. Anyone claiming 99% across every scenario is selling you something.

Here's what changed for one juice brand we work with in Oman. Before AI, their supervisor manually audited maybe 6% of visits. After turning on shelf analysis, every single visit got scored automatically. They found that 31% of their outlets in Muscat had planogram violations they didn't know about. Wrong facings. Competitor products in their allocated space. Missing SKUs the distributor claimed were stocked.

Fixing that gave them a 9.4% sales lift in three months. Not because they added outlets. Because the outlets they already had started actually selling what was supposed to be on shelf.

Why the GCC is a weirdly perfect market for this

I've spent time working with FMCG teams in the UK and US too, and there's something specific about the Gulf that makes AI shelf analysis land harder here.

First — the density. A Riyadh sales route might hit 22 outlets in one day. A London route, maybe 8. More outlets means more photos, means more chaos, means more value from automation.

Second, the retail mix is brutally mixed. You've got modern trade giants like LuLu and Carrefour with strict planogram rules, sitting next to bakalas and small groceries where merchandising is basically wherever the shopkeeper felt like putting your product that morning. Manual auditing across that spread is impossible. AI doesn't care — it scores both the same way.

Third, and this is the one nobody talks about: the language piece. Price tags in Arabic, English, sometimes Urdu in Sharjah, Malayalam in Muscat. Older OCR tools choked on this constantly. The newer vision models handle multilingual shelf tags much better, which matters more here than it does in, say, Manchester.

Fourth — WhatsApp culture. Reps here are used to snapping photos and sending them. The behaviour already exists. You're not asking them to learn a new habit. You're just plugging intelligence into a habit they've had since 2016.

The uncomfortable truths nobody in the vendor space will tell you

Look, I run Zivni. I want you to buy our AI shelf photo analysis. But I've watched enough deployments go sideways to tell you the truth about where this fails.

It fails when leadership doesn't decide what to do with the data. You'll get share-of-shelf reports, planogram compliance scores, out-of-stock alerts — and if nobody's job is redesigned around acting on those alerts within 24 hours, you've just built a very expensive dashboard.

It fails when incentives don't change. If your reps are still paid purely on orders booked, they'll still take blurry photos on purpose to avoid getting flagged. I've seen it. Reps are smart. They'll game whatever system you build unless the compensation math is aligned with merchandising compliance, not just volume.

It fails when the AI is walled off from the rest of the field sales automation. A compliance score means nothing if it doesn't trigger a task, adjust a rep's next-day route, or feed into the KPI shown to their supervisor. This is why we built it inside the same platform as beat planning, order capture, and gamification instead of as a bolt-on. Standalone image analysis tools tend to die in adoption because they don't connect to what the rep does next.

The brands winning with this in the GCC right now aren't the ones with the fanciest AI. They're the ones who redesigned their weekly ops rhythm around it. Monday morning: supervisor reviews compliance heat map. Tuesday: bottom 20% outlets get corrective visits. Friday: trade marketing sees which SKUs are consistently under-facing at which chain.

That's the real transformation. The AI is the easy part. The organisational change is where 8 out of 10 deployments live or die.

If you're evaluating this for your brand — whether you're in Jeddah or Kuwait City or Manama — the question isn't "which vendor has the best image recognition." The question is: who on your team owns the response to what the AI finds? Until you can name that person, don't sign anything.