Pirtek Africa focusses on being AI-ready from the inside out
While many industries have announced that they have ‘gone AI’, Pirtek Africa took the slower, less glamorous route: cleaning up its data, testing its systems against South African privacy law, and mapping out exactly what AI-readiness means, before letting any algorithm near a real customer record.
“Considerable time and effort has gone into auditing and governing that data, well before pointing a single AI agent at it.” says Lessing. IT Manager at Pirtek Africa. The logic behind this decision is simple: Pirtek Africa keeps its promise to customers, a mobile service technician on site with the right hydraulic hose before a stopped machine costs them a shift, by running on data. Which unit carries which part, which site needs which fix, how fast a job can be dispatched.
That same data, captured accurately in the field through manual docket books and mobile devices, is what allows Pirtek Africa to give customers real business intelligence rather than guesswork. For mining clients in particular, it means machine cost analysis across maintenance, breakdowns and rebuilds, pinpointing where a failure occurred on a machine and what caused it. Hydraulic hose serialisation takes it further, giving full traceability of every hose fitted: which one, when, and by whom. That is exactly the kind of information any future AI agent would eventually be asked to work with, and the precise reason the data had to be in order first.
“Everyone wants to talk about AI. Almost nobody wants to talk about what it’s reading,” says Lessing. “An AI agent is only as good as the data you point it at. Feed it a decade of messy, duplicated, ungoverned records and it won’t give insight. It’ll give you expensive nonsense, faster.”
The caution is backed by the numbers. Gartner predicts more than 40% of agentic AI projects will be cancelled before the end of 2027, and though the figure is debated, MIT research suggests as many as 95% of enterprise AI pilots deliver no measurable return. Of the thousands of vendors now selling ‘AI agents’, Gartner reckons only around 130 are the real thing.
The businesses getting it right are seeing real gains: Klarna has reported cutting service costs by $60 million, and JPMorgan credits similar tools with reclaiming 360,000 lawyer hours a year. The dividing line, Pieter Lessing argues, isn’t the tool, it’s the foundation beneath it.
“The AI revolution in most businesses is really a data-quality project wearing a robot costume,” he continues. “Before you budget for agents, ask the harder question: are you investing in the AI, or in the foundation that makes it work?”
Where does the data go?
Data quality was only half the exercise. Before allowing any AI tool near a real customer record, Pirtek Africa ran an internal approval process under the Protection of Personal Information Act (POPIA), deliberately testing with a dataset that held no personal information at all.
The company’s ERP/CRM system runs on customer information: site addresses, contact details, and hydraulic equipment and service histories, the records that get a mobile service technician to the right site with the right part. That is exactly the kind of data any future AI agent would eventually need to touch, which is why the first question wasn’t what the AI could do.
“The first thing I wanted to know wasn’t what the AI could do. It was where the data actually goes,” says Pieter Lessing. “That’s exactly why we tested with dummy data first.”
Most AI tools don’t process information on a local device. They run on servers in the United States or Europe, there is no South African processing region for this kind of work. So the moment an AI tool reads a customer record to be helpful, that personal information has effectively left the country. Under POPIA, that counts as a Section 72 cross-border transfer, lawful only under specific conditions, the strongest being the data subject’s consent.
And that, Pieter Lessing says, is the trap: “An AI tool can make hundreds of calls a day, none of which the customer ever sees. So how, exactly, do they consent?”
The stakes aren’t hypothetical. In 2023, Meta was fined a record €1.2 billion under GDPR, Europe’s equivalent of POPIA, for exactly this kind of cross-border transfer. Closer to the everyday risk, IBM’s 2026 Cost of a Data Breach report puts the global average breach at $4.99 million, roughly $1 million higher when AI is involved.
“I’m not anti-AI, it’s my job to bring it into our business,” Pieter Lessing says. “But I’ve learned one rule I won’t break: you don’t get to be excited about AI until you can answer for the data.”
Pirtek Africa is encouraging other South African businesses adopting AI to ask the same question of their own systems.
Five steps to AI readiness, none of them AI
‘Get AI-ready’ has become one of the most repeated pieces of advice in business and, according to Pieter Lessing, one of the least useful, because it almost never says what the phrase actually means. So Pirtek Africa spelled it out: five concrete steps toward AI readiness and, notably, not one of them requires switching on an AI tool.
The first is finding the data before feeding it to anything: an inventory of where personal and business information actually lives, from email archives to old shared drives and the SharePoint site nobody has opened in years.
The second is fixing permissions. “Most companies have folders set to ‘everyone can access’ that have quietly grown for a decade,” Pieter Lessing says. “A person opens one file at a time. An AI agent opens all of them at once. That’s the difference that should worry you.”
The third is deleting redundant, obsolete and trivial data: information with no reason to exist isn’t an asset, it’s a liability waiting to be processed. The fourth is giving any AI agent a sandbox rather than the keys, scoping it tightly to a single task instead of workspace-wide access “to be helpful.” The fifth is logging everything: under POPIA, automated processing still counts as processing, and a business that can’t produce an audit trail of what an AI touched can’t prove it was compliant.
“Steps one to three aren’t AI projects, they’re housekeeping,” says Pieter Lessing. “Skip them, and you’re in the group whose AI projects get quietly cancelled. Do them, and you’re not just AI-ready, you’re running a cleaner, safer, cheaper business, with or without the robot.”
Within a network of independent operators, getting that foundation right at head office level isn’t back-office IT, it hands every branch a head start it would otherwise have to build alone. Data quality, privacy compliance and basic housekeeping, in that order, is what ‘AI-ready’ actually looks like at Pirtek Africa.
