Your Merchandiser Spent 40 Minutes Counting Facings. A Photo Does It in 6 Seconds.

A merchandiser in Sharjah once told me she spent 40 minutes at a single Carrefour hypermarket counting facings, checking planogram compliance, and scribbling stock levels into a paper form. Forty minutes. And when she got back to the car, she couldn't remember if the Nestlé block had 7 facings or 8, so she guessed.

That guess went into a report. That report went into a dashboard. Somebody in head office made a decision based on it.

This is the dirty secret of manual shelf audits. They're slow, and worse, they're wrong more often than anyone admits. When we ran a quiet test with one distributor in Karachi, we found that manual facing counts were off by an average of 23% compared to what was actually on the shelf. Not because reps are lazy. Because counting 60 SKUs across a 12-foot gondola while a customer is reaching past you is genuinely hard.

What computer vision actually does on a shelf

Here's the short version. Your rep takes a photo of the shelf. That's it. The AI reads it.

Within seconds it tells you how many facings each brand has, whether your product is where the planogram says it should be, what your competitors are running, and whether there's an out-of-stock gap. No counting. No paper. No guessing in the car park.

The technical name for this is computer vision merchandising, and honestly the term makes it sound more intimidating than it is. The model has been trained on thousands of shelf images. It recognizes a Lay's packet the same way you recognize a friend's face across a room — not by reading the label letter by letter, but by pattern.

When we built shelf audit automation into Zivni, I assumed the hard part would be the AI. It wasn't. The hard part was photo quality. Reps hold the phone at weird angles, the lighting in a Lulu freezer aisle is different from a corner dukaan in Lahore, and sometimes a customer's shoulder is blocking half the shelf. We spent months just teaching the system to handle bad photos gracefully instead of throwing an error. I got that priority wrong at first — I thought accuracy on perfect photos was the win. Turns out nobody takes perfect photos.

What the AI gives you that a human never could is consistency. Every audit measured the same way, every time, in every store. A rep in Muscat and a rep in Manchester counting the same shelf will disagree. The camera doesn't disagree with itself.

The numbers that make ops heads care

Let me be practical about why this matters beyond "it's faster."

An AI planogram check turns a 40-minute audit into something closer to 6 seconds of photo capture plus a few minutes of the rep confirming what they see. That's not a rounding error in time savings. If a rep does 15 outlets a day and each shelf audit drops from 40 minutes to 5, you've just handed them back a huge chunk of their route. More outlets covered. Or the same outlets covered properly instead of rushed.

Then there's the data. With manual audits you get numbers weekly, maybe. With photo-based capture you get a live view. One CPG client in Riyadh spotted a distributor systematically under-stocking their premium line within days, something the old monthly report had been hiding for a full quarter. That's real money — shelf space they were paying for and not getting.

And the out-of-stock detection is where I've seen the fastest ROI. Every hour your product is missing from a shelf is a sale walking to a competitor. When the system flags a gap in real time, the rep can fix it before they leave the store instead of finding out three weeks later in a spreadsheet.

Look, I'm not going to pretend the technology is magic. It isn't. It misreads new packaging until it's retrained. It struggles with products stacked behind each other. And in stores with genuinely chaotic shelves — and I've seen some in smaller markets that would break any system — you still need a human eye. The AI handles the 80% that's repetitive and error-prone. The human handles the judgment calls. That split is the whole point.

Where this actually goes next

The part I'm most excited about isn't better counting. It's what happens when the shelf data connects to everything else.

Imagine the audit photo automatically triggering a reorder because it saw an out-of-stock. Or the AI planogram result feeding straight into a rep's incentive — you hit your compliance target this week, here's the bonus, no manager reviewing photos by hand. We're building toward that at Zivni, where the shelf photo isn't a report you file, it's a trigger that moves the next thing.

Here's the thing I keep telling ops leaders who are nervous about this. Computer vision isn't replacing your merchandisers. It's replacing the boring, error-filled clerical part of their job so they can do the part that actually needs a person — talking to the store owner, negotiating that extra shelf, fixing the display. The counting was never the valuable bit. We just didn't have anything better until recently.

The distributors who figure this out early are going to have a data advantage that compounds. Every photo makes the model smarter about their specific SKUs, their specific stores, their specific competitors. A year in, they'll know their shelf reality better than the retailer does. Two years in, that gap gets embarrassing for whoever's still counting facings with a clipboard.

My honest bet? Within three years, asking a rep to manually count facings will feel as dated as asking them to fax the order in. The tech is already good enough. The only real question is who moves first — and whether your competitor is reading this same idea right now, wondering the same thing about you.