The ROI of AI Shelf Photo Analysis: What the Numbers Actually Look Like
Last month a distributor in Sharjah sent me a spreadsheet. He'd been running AI shelf photo analysis across 340 outlets for 11 weeks. His question was simple: "Am I actually making money on this, or am I just making my reports look prettier?"
Fair question. And honestly, one I don't think enough founders in this space answer straight.
So let's actually do the math. Not the marketing-deck math where every KPI moves 40% and everyone lives happily ever after. The real math — the kind where you have to explain to your CFO why you're spending $2 per outlet per month on a computer vision tool.
What you're actually paying for (and what it costs to skip)
Before ROI, you need a baseline. Most FMCG brands I talk to have no idea what their share of shelf actually is. They think they know. They have a monthly report from a merchandising agency that says "92% planogram compliance" and everyone nods.
Then we deploy AI shelf recognition at Zivni and the real number comes back at 61%.
That 31-point gap is where the money lives.
Here's a rough framework I use with prospects. Take a mid-sized snacks brand doing $8M annually through 1,200 outlets. If planogram compliance is running at 61% instead of the assumed 92%, and if every 10-point improvement in shelf compliance drives roughly 3-4% in sell-through (this is Nielsen's number, not mine), you're looking at somewhere between $750K and $1M in leaked revenue. Per year. From one category.
Now the cost side. AI shelf photo analysis, depending on vendor, runs anywhere from $1.50 to $4 per outlet per month. Call it $3 average. For 1,200 outlets that's $43,200 a year. Add merchandiser time (they still need to fix what the AI flags) and you're maybe at $70K all-in.
So the math is roughly 10x return if you actually act on the data. That last part is where 80% of deployments fall apart, which I'll get to.
The four numbers you should actually track
I got this wrong at first. When we built our shelf analysis module, I thought brands would want a single "shelf score" that goes up and to the right. Turns out ops leaders don't trust composite scores. They want specifics.
After about two years of watching customers use the tool, here are the four numbers that actually predict merchandising ROI:
Share of shelf (SOS) vs. contract. If you paid the retailer for 40% facings and you're getting 27%, that's not a merchandising problem, that's a trade spend recovery problem. One dairy brand in Riyadh clawed back $180K in a single quarter just by showing photo evidence to modern trade buyers.
Out-of-stock rate at shelf level. Not warehouse OOS. Shelf OOS. These are wildly different numbers. Warehouse says you have stock. The shelf says the SKU has been missing for six days. AI catches this in near real-time; humans catch it on the next visit cycle, which might be two weeks out.
Planogram compliance by SKU, not by outlet. Averages lie. An outlet at 85% compliance might be missing your two highest-margin SKUs. That's worse than an outlet at 60% that has your hero products facing correctly.
Time-to-fix. How many hours between the photo being flagged and the issue being resolved? If your answer is "we don't know," your ROI is probably zero. I mean that.
Where the ROI story falls apart
Here's the thing nobody wants to say out loud: AI shelf photo analysis doesn't fix shelves. Reps and merchandisers fix shelves.
I've watched customers deploy beautiful dashboards, generate thousands of compliance alerts per week, and then… nothing happens. The alerts pile up. The rep still gets to the outlet on Thursday because that's when his beat says he goes. The merchandiser doesn't have authority to re-negotiate facings with the store manager. The regional sales manager looks at the dashboard once a month.
And everyone wonders why the ROI didn't materialize.
Look, the tech is the easy part now. Computer vision on packaged goods is basically a solved problem — accuracy is above 95% for most categories in decent lighting. The hard part is workflow. Who gets the alert? What are they supposed to do in the next 24 hours? What happens if they don't do it?
The deployments I've seen actually hit 8-12x ROI have three things in common. One, alerts go directly to the rep's mobile within minutes, not to a manager's email overnight. Two, there's a named person accountable for time-to-fix as a KPI. Three, monthly business reviews with retailers include the shelf data — which changes the entire negotiation dynamic.
A soft drinks distributor we work with in Karachi rebuilt their KAM meetings around AI shelf data. They walk in with photos, dates, compliance percentages by store. The conversation stopped being "give us more space" and became "here's what your competitor is doing in your own store." Their contracted facings grew 22% in two quarters.
A quick way to size your own number
If you're trying to build a business case, don't overthink it. Grab these five inputs:
- Annual revenue for the category you'd deploy on
- Number of outlets covered
- Current planogram compliance (guess if you have to — you'll be wrong, which is the point)
- Estimated share-of-shelf gap vs. contract
- Your gross margin
Assume a 3% revenue lift per 10 points of compliance improvement. Assume you'll close half the SOS gap in year one. Multiply. That's your upside.
Then subtract the tool cost, integration time, and the cost of whoever's job it is to actually chase down the alerts. If the ratio is under 4x, don't buy it — the workflow won't survive contact with reality. If it's above 6x, you probably have a real case.
And if your merchandising agency is telling you compliance is already at 90%+ across the board, ask them how they measured it. That answer alone will tell you whether you need this.
What's your current shelf compliance number, and how confident are you in it?