Companies rolling out AI should set realistic expectations, start with small pilots, and prove the technology works before spending big on it, tech executives advised at the Cairo ICT conference this month.
Why it matters: Just because you use AI doesn’t mean you’ll make (or save) money with it. Nearly nine in ten people McKinsey surveyed say their companies use AI regularly, but only 37% say it’s done anything for profits, the consultancy found. Research firm Gartner predicts more than 40% of agentic AI projects — AI that does tasks on its own — will be scrapped by the end of 2027.
Expectations are the first trap: Companies often build a custom AI tool to do one specific job, then judge it against ChatGPT, a tool that’s been mass-tested for years, says Mohamed Ashraf (LinkedIn), who heads AI and business software development at Zone. Unsurprisingly, the first reaction is often disappointment. But that usually isn’t the AI’s fault, because the company often picked the wrong task to turn over to AI in the first place, Ashraf says.
Start small: GPUs, the powerful chips that AI runs on, don’t come cheap. So if you ask your company for a big budget to buy them upfront, the money folks “will remain skeptical,” says Dr. Hatem Elattar (LinkedIn), head of industrial AI at Elsewedy Electric. His advice is to rent computing power from a regional cloud provider for a small pilot and show results first. “Once they see tangible value delivering an operational impact, the business will be ready to invest,” he says.
Other advice:
- Keep the sensitive stuff at home: Elattar favors a hybrid setup. Data that’s especially sensitive stays on a company’s own servers, while everything else can run in the cloud.
- Use the front door: AI shouldn’t act “like the company owner’s privileged relative who enters unchecked, claims full access, and demands visibility over every database,” Ashraf says. It should follow the same access rules as any other system, he says. Companies should also start off with AI only being able to read data, not change anything, he says.