Most AI pilots never pay off. An MIT study in 2025 found that about 95% of generative AI pilots delivered little or no measurable impact on profit.
We asked Dirren Subbiah, Head of Data Ecosystems and Data Science and Analytics at ShopriteX, the Shoprite Group’s data and innovation unit, how the group decides which AI ideas earn the right to scale…
The move: no budget without proof of value
AI has made people more willing to spend, much like a startup with AI in the pitch attracting investors. Shoprite stays pragmatic. An AI idea only gets real funding once it has shown it creates value for the business.
“You can’t unlock funding for something that doesn’t have a material part to value creation.”
How to prove an AI idea works before you scale it
1. Start as a sceptic
Dirren describes himself as a sceptic at the start of any AI idea. That isn’t a vote against AI. It’s a refusal to assume an idea works because the technology is exciting. Begin with the question of whether it creates value, not how impressive it looks.
2. Decide what value means before you build
Pick the number that would prove the idea worked: money saved, sales gained, hours freed or errors cut. Write it down before the test starts, along with where it stands today. Without a clear target, a pilot can look busy and still prove nothing.
3. Test it small first
Run the idea with a limited group, one team, one region or a slice of customers, before you go wide. The MIT research found that the pilots that succeeded tended to keep their scope narrow. A small test is cheap to stop and easy to measure.
4. Weigh the downside as seriously as the upside
At Shoprite’s scale, both sides of the bet are big. In Dirren’s words, with millions of customers, the potential upside is significant, but so is the potential for catastrophic failure. Ask what happens if the AI gets it wrong for every customer at once, and make sure the test answers that too.
5. Scale only what has proved itself
Once the test hits the value target and the risks are understood, then consider scaling. Dirren’s summary is that you have to be measured. Ideas that miss the target get dropped or reworked, which keeps the budget for the ones that earn it.
The big payoff
You stop funding AI because it sounds promising and start funding it because it has already shown results. That keeps spending tied to value, and it means the projects that do scale have far fewer surprises.
You might also be interested in
Want the full story?
Dirren’s full panel from The Open Letter Cape Town 2026 is available to members inside the Founder Collab, where he goes further than we could cover here:
Why ShopriteX started its AI work in the back office, not with customers
How Shoprite vets every AI tool before customer data goes near it
Why Pixie, its AI shopping assistant, runs mostly on classical AI
Why each team at ShopriteX uses a different AI tool
You’ll also get access to 40+ masterclasses from SA founders and operators on sales, fundraising, UX, paid media and more inside The Founder Collab.
Get more SA tech and business news, tips and business-building workflows and subscribe to The Open Letter.



