Most businesses start with an AI tool and go looking for somewhere to use it. That’s backwards, and it’s why so many AI projects end up as demos rather than results.
We asked McElory Hoffmann, co-founder and CEO of Praelexis AI, which has built AI for South African and European companies since 2013, how he finds the right problems to work on…
The move: start with the problem, not the technology
McElory first saw an AI algorithm in 2002, when facial recognition was taught in applied maths. Two decades later, his starting point hasn’t changed. AI is a way to get somewhere, not the destination.
“What is the problem I want to solve, and how can I use this technology to solve it?”
How to find the problems AI can actually solve
1. Ask the people closest to the work
People often already know what’s broken and what they wish they could see. McElory’s example is a farmer who says: if I knew next year’s yield on this land, I could sell better. That’s a clear problem with a clear payoff, and it points straight at a prediction AI can make.
2. Map your processes and read off the bottlenecks
Write down each stage of a key process, and where possible, attach a number to each: how long it takes, how often it fails, how much it costs. McElory says some organisations have stats for every stage, so you can almost read off where the bottlenecks are. This matters more now that AI agents can take over whole stages of a process.
3. Show people what the technology can do
This is the hardest route, and McElory uses a hammer to explain it. With the first 2 routes, you start with the nails, the problems people can see. But if you’ve never seen a hammer, how would you know it can hit something? Teach your team what AI can actually do, with real examples, so they can recognise uses they couldn’t imagine before.
4. Ask what’s possible now that wasn’t before
Once people understand the technology, ask them one question: with this new knowledge, what can I do that I couldn’t do before? The answers often point to work nobody tried, because until recently it simply wasn’t possible at a sensible cost.
5. Write each problem down with its payoff
For every problem you find, note what solving it would be worth, even roughly: time saved, sales gained, errors cut. McElory says defining that value is often harder than finding the use case, so start early. It turns a long list of ideas into something you can compare and rank.
The big payoff
You end up with a list of real problems, each with a rough value attached, instead of a tool looking for a job. That makes it far easier to pick the AI projects worth doing and skip the ones that would only make a good demo.
You might also be interested in
Want the full story?
McElory’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:
How to rank AI projects on value against risk
How to keep an AI system working long after launch
How to decide where your AI processes your data
How to use AI without handing over what makes your business valuable
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.



