Operational Excellence

"AI-driven Kaizen" is a misnomer — but there's a real story underneath

The phrase is everywhere in vendor decks right now. It also does quiet violence to what Kaizen actually is. A practitioner's look at where artificial intelligence genuinely fits in continuous improvement — and where it never will.

Bharat Kumar4 min readAugust 2026
Operator in safety gear reviewing data on a tablet in an industrial facility
The tablet is new. The judgement holding it is the part that was never the bottleneck.

Kaizen (改善) translates loosely as "change for better." But the word has never really meant technology, or scale, or speed. Its core is small, human, and cultural — incremental improvements, owned by the people closest to the work, made continuously because improvement has become a habit rather than a project. It is one half of the Toyota Production System's foundation, and the other half is "respect for people." That is not incidental. It is the point.

So when a vendor deck promises "AI-driven Kaizen," a Lean practitioner's ear should twitch. You cannot automate a philosophy of human ownership any more than you can automate mentorship. What the tools are actually doing is something real and valuable — it is just not Kaizen. Naming it correctly matters, because it changes where you look for the return.

The shift is real, even if the label is wrong

Gartner's 2026 Manufacturing Predicts report frames it clearly. Today, semi-autonomous artificial intelligence agents orchestrate roughly 2 per cent of key production, quality, and maintenance use cases worldwide. By 2030, Gartner projects 10 per cent — a five-times increase in four years. Something is genuinely changing on factory floors. The question is only what to call it, and therefore what to expect from it.

35 – 50%
Reduction in unplanned downtime for factories using AI-driven predictive maintenance — vs 15–30% for traditional approaches
15 – 20%
Higher Overall Equipment Effectiveness at factories using AI predictive maintenance vs calendar-based preventive maintenance
80 – 97%
Accuracy range of modern AI predictive maintenance systems in forecasting equipment failures
66%
Of manufacturers expected to adopt a hybrid model in 2026 — AI for critical assets, traditional programs elsewhere

Those are strong numbers. But notice what they measure: downtime, equipment effectiveness, failure prediction. That is the language of predictive maintenance, statistical process control, and Six Sigma analytics — not the language of Kaizen. Which is exactly the point worth making.

Where the line actually falls

The clearest way to think about it is a division of labour. Artificial intelligence is extraordinary at the mechanical, high-volume, always-on work that used to consume a continuous-improvement team's attention. It is incapable of the human work that was Kaizen's real purpose in the first place.

What AI genuinely does well
  • Continuous waste scanningWatches real-time data for deviations no human huddle would catch in time.
  • Computer-vision defect detectionFlags defects and motion waste the moment they occur, not at end-of-line.
  • Automated root-cause suggestionMines historical records to propose likely causes — a faster first draft of an A3.
  • Sustainment monitoringWatches for regression after a gain is made — where improvements usually quietly fade.
What only Kaizen can do
  • Ownership on the floorAn operator who improves their own station because they see it as theirs.
  • Small daily change as habitThe cultural reflex to improve without being told, prompted, or measured.
  • Respect for peopleBuilding capability in a person, not extracting throughput from a process.
  • Judgement in contextKnowing which improvement matters this week, on this line, for these people.

The mechanics can be automated. The philosophy cannot. "AI-driven Kaizen" collapses the two — and in doing so, quietly promises something it cannot deliver.

Traditional maintenance vs AI-driven — the honest comparison

Unplanned downtime reduction — traditional vs AI-driven
Best-case reduction ranges reported across manufacturing sectors.
AI-driven
predictive maintenance
35 – 50%
Traditional preventive
& reactive maintenance
15 – 30%
Roughly double the downtime reduction — but only on assets where sensor investment is cost-justified. This is predictive-maintenance performance, correctly named. It is not a measure of Kaizen.

A concrete example — ENGIE, January 2026

ENGIE, a global energy company, shifted from time-based maintenance schedules to artificial-intelligence-driven condition monitoring across 10,000 connected assets in January 2026, reporting $870,000 in annual savings. That is a genuine win — and it is a predictive-maintenance win. Calling it "AI Kaizen" would obscure what actually happened: sensors and models replaced a calendar. Valuable, measurable, and entirely distinct from an operator suggesting a small fix to their own workstation.

The caveat that ties it together

Here is where the two threads meet. Artificial intelligence only produces reliable insight when it sits on top of a plant that already has operational discipline — functional Total Productive Maintenance, operator-led autonomous care, honest preventive-maintenance compliance, a maintenance system that reflects reality. Without those, AI is measuring chaos, which is expensive noise rather than insight.

And building that discipline is Kaizen work. Small, human, cultural, continuous. So the real relationship is almost the reverse of the vendor framing: Kaizen is not something AI runs. Kaizen is the foundation that makes the AI worth buying. The 66 per cent of manufacturers moving to a hybrid model are, whether they say so or not, acknowledging exactly this.

Is your CI programme confusing the tools with the practice?

A brief self-assessment. Nothing tracked, nothing saved.

Eight honest signals
Check what is true today, not what appears on the roadmap.
Operators still suggest and own small improvements — the AI tools have not replaced that habit.
Preventive-maintenance compliance is above 90 per cent, and that number is honest.
The maintenance system reflects what actually happens on the floor, so any AI has clean data to learn from.
A previous improvement gain has been sustained for at least twelve months.
The continuous-improvement team, not only IT, has a defined role in the digital programme.
Leadership can articulate the difference between a predictive-maintenance win and a Kaizen win.
Governance for AI-driven decisions is defined — who overrides, who audits, who escalates.
The plant would still improve if every dashboard went dark tomorrow.

What to watch next

2026 – 2027
The hybrid model becomes the default
The 66 per cent figure is a projection. The follow-on data will show whether AI-plus-traditional discipline settles in as the operational norm — and whether the industry starts naming the pieces correctly.
2027 – 2028
Agentic AI moves from suggestion to action
Today's tools mostly suggest. Tomorrow's agents will schedule and adjust autonomously — which makes the human-judgement layer more important, not less.
By 2030
Semi-autonomous agents at 10 per cent of use cases
Gartner's projection. The plants that benefit will be the ones that kept their Kaizen culture intact while adding the tools — not the ones that mistook the tools for the culture.

None of this is an argument against the technology. The downtime numbers are real, the return on investment is real, and any plant that can justify the sensors should be looking hard at predictive maintenance and computer-vision quality. It is an argument for precision. Call predictive maintenance what it is. Call statistical process control what it is. And keep the word Kaizen for the thing it has always described — the small, human, continuous improvement that no model has ever made, and that remains the foundation the entire toolkit is built on.

Sorting the tools from the practice in your own CI programme? Happy to compare notes.

Get in touch →

Where these numbers come from

Bharat Kumar · Manufacturing Transformation & Operational Excellence