Unfiltered with Lucinda Boshoff: AI in Operational Industries: Why Discipline Matters More Than Hype

Picture a plant hire depot on the edge of Johannesburg, 7.15 on a Tuesday morning. The hire controller has three machines due back, a telehandler that never reached site, a driver waiting on paperwork and a phone that hasn't stopped since she unlocked the office. Every decision she makes in the next hour rests on the data being right. Which asset is where. What's been serviced. What the customer actually signed for.

Nobody in that depot is asking for a chatbot. They want fewer errors, faster answers and a system that keeps pace with the morning.

That, in the end, is the test AI has to pass in operational industries. Not whether the demo impresses a boardroom, but whether it survives 7.15 on a Tuesday.

South Africa is adopting quickly. Strategy is struggling to keep up.

The appetite here is real. The South African Generative AI Roadmap 2025 found that adoption among large enterprises rose from 45% in 2024 to 67% in 2025. Yet only 14% have a formal AI strategy. As Arthur Goldstuck observed, too many organisations mistake using an AI tool for having an AI strategy.

The most successful conversations we're having focus on where AI genuinely adds value. They're about removing friction from the work people do every day, helping people find information faster and making everyday operations run more smoothly. We're already seeing businesses move beyond asking whether they should adopt AI and towards asking where it genuinely adds value. That's a far more useful conversation.

The gap between potential and reality

There is no shortage of AI announcements. New tools appear almost weekly, each promising transformation. The evidence on what happens next is sobering.

MIT's Project NANDA found that 95% of Generative AI pilots failed to deliver measurable profit-and-loss impact despite estimated enterprise spending of US$30 – 40 billion. Crucially, the failures were rarely about the AI models themselves. They stemmed from poor integration, limited understanding of workflows and an inability to adapt to the way businesses actually operate.

Gartner tells a similar story about the next wave. It predicts that more than 40% of Agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. The same research estimates that a third of enterprise software will include agentic capability by 2028, up from less than 1% in 2024.

Taken together, these findings point to a clear conclusion. AI will increasingly become part of business software. The organisations that benefit most won't necessarily be the first to adopt it, but the ones that implement it deliberately, with clear objectives and the right operational foundations.

Where AI genuinely earns its keep

In the sectors we work with, three capabilities matter more than the rest.

Generative AI helps people find answers, surface information and complete everyday tasks more efficiently. A service advisor asking a plain-English question about a vehicle's history, rather than clicking through six different screens, is not a glamorous use case. It's exactly the kind of practical improvement that saves time every single day.

Predictive analytics is the quiet workhorse, and arguably the most valuable of the three. Demand forecasting in distribution. Utilisation and maintenance planning in rental. Parts availability in the workshop. None of it makes headlines, all of it delivers value over time, week after week.

Agentic AI is the frontier, software that acts rather than simply answers. The potential is enormous and so is the temptation to deploy it prematurely. Gartner has even coined a term, agent washing, for vendors rebadging ordinary chatbots as autonomous agents. In operational environments, agency has to be earned in stages, with tight guardrails, human oversight and a clear record of every action taken.

Embedded, not bolted on

Stock, jobs, parts, assets and orders are not abstract datasets. They are the moving parts of a real business. Which is why intelligence belongs inside the systems that run those processes, respecting the same permissions, workflows and audit trails as everything else.

MIT's research points to the same conclusion from the failure side. Generic, standalone tools stall in the enterprise precisely because they sit outside the context of the work. They create duplication, confusion and, in safety-critical settings, genuine risk.

AI shouldn’t require businesses to change the way they operate. It should strengthen the processes they already trust.

Trust is the whole game

Adoption in operational industries lives or dies on trust. If a hire controller cannot see why the system flagged an asset, she will stop looking at the flags. If outputs are unpredictable, people quietly route around them, and no amount of technical capability wins them back.

That's why strong AI governance matters. AI must respect user permissions. It must be transparent about how it reached an answer. And rollouts should be phased around product readiness and genuine customer value rather than announcement calendars. Moving quickly matters in this market. Moving responsibly matters more.

Back at the depot

Now imagine that same Johannesburg hire desk twelve months from now. The overdue machines were flagged before the controller sat down. The paperwork for the waiting driver has already been prepared from the contract. The utilisation forecast already reflects last week's weather delays. Her morning is calmer. The work hasn't disappeared, the friction has.

She’ll probably never use the phrase “Agentic AI.” She'll simply notice that work flows more smoothly and the right information is there when she needs it.

The organisations that come out of this next phase of AI adoption in South Africa's rental, automotive and distribution sectors won't necessarily be the loudest adopters. They'll be the ones that embed AI deliberately into the systems people rely on every day, treating it as a long-term capability rather than the latest technology trend.

Hype fades quickly in a depot at 7.15 on a Tuesday. Dependability never does.

‍