AI in Water Treatment: NICP Tests Where the Technology Works — and Where It Doesn’t
By Kris van der Bijl
It is rare to find an innovative voice in AI who is truly sceptical about the limitation of the machines and technology they build. But a founder of a Johannesburg firm developing AI for water treatment plants wants its first live pilots to show where the technology fails.
Unathi Pretty Mbengwana founded New Innovations Consulting and Projects (NICP), which is building Water Treatment Expert AI (WTE AI) for water and wastewater professionals.
“A successful pilot should not simply confirm that our idea works. It should identify what works, what does not work, what needs improvement and where AI should not be used,” she says.
Mbengwana built the concept on more than a decade of engineering, maintenance, reliability and failure analysis work in the wastewater sector. Her reliability work involves analysing equipment failures from maintenance records that vary in consistency and detail.
That experience led her to ask whether AI could compare maintenance history with operational information and plant conditions to flag risks for investigation.
Water treatment AI pilots
NICP started with a customised minimum viable product (MVP) to validate the need before building a complex standalone platform.
The MVP has recorded more than 700 interactions. Those interactions show NICP the technical and operational questions professionals bring to the platform.
Water Treatment Expert AI was among South Africa’s top three nominees for the 2025 BRICS Young Innovator Prize representing South Africa in Brazil.
Mbengwana separates proof of demand from proof of a product inside a treatment plant. NICP’s next phase is a standalone platform and its first formal pilots in operating plants.
The pilots will test the platform against real operational workflows. Her questions cover access to information, consistency of troubleshooting, the platform’s mistakes and the design of human oversight.
Two more questions concern structured failure analysis and the information the platform requires. A final question asks which capabilities create measurable operational value.
“Only after generating that evidence should we make strong claims about operational performance and move towards wider commercialisation,” she says.
AI decision support in treatment plants
NICP designed the platform to make technical and organisational knowledge accessible when professionals need it. Treatment facilities depend on experienced personnel, according to Mbengwana. Their knowledge can leave with them when they retire or resign.
Newer employees can wait for senior colleagues before receiving guidance on unfamiliar problems, she says. A less-experienced professional could draw on structured technical guidance from the platform.
The MVP covers troubleshooting, technical guidance, structured root-cause analysis and compliance guidance. It stands apart from plant equipment, SCADA, PLCs and live plant data.
PLCs and control systems execute control logic, and SCADA provides monitoring and alarms. Mbengwana places AI in a separate knowledge and decision-support layer.
She pictures the platform telling an engineer that a pattern or inconsistency deserves attention, along with the information behind that flag. Future versions could compare present conditions with plant history and flag emerging risks.
She describes that capability as a development vision for testing in pilots. In her view, incomplete job cards, unreliable instrumentation, miscalibrated sensors and missing operational context all limit such analysis.
Human oversight of plant decisions
Mbengwana keeps consequential decisions with qualified professionals. Those decisions cover safety, environmental compliance, equipment integrity, process stability and public health.
She notes that a system records “only part of what is happening in the physical plant”. An experienced operator hears a change in a machine’s sound or notices unusual vibration.
“AI should strengthen engineering judgement, not replace it,” says Mbengwana.
She sees the biggest opportunity in strengthening the people who operate and maintain South Africa’s treatment infrastructure.
She advises South African plants to start with the problem. Her four questions cover the problem, the available information, its reliability and the decisions AI could improve. Facilities can begin at their current level of digital maturity.
Her long-term vision is a plant where automation controls processes and AI helps professionals interpret plant information.
NICP is seeking utilities, treatment facilities, industrial operators, research organisations and technology partners for the pilots.
