Why Your AI KPIs Need a Human Side (Before You Cut Headcount)

The Question Every Business Owner Asked Me on Friday

Every training has a moment where the real concern in the room surfaces. Today it was headcount. Not asked directly at first — more like “how do we know which tasks to automate” — but underneath it, the same fear every leader in that room is quietly carrying: if AI can do the work, do I still need the person?

It’s the wrong frame, and it’s costing businesses more than they realise. Not because automation is bad — it isn’t — but because “automate and reduce” treats AI adoption as a cost line instead of a capability decision. A business that cuts its way to efficiency gets smaller and more fragile. A business that uses AI to free its people for higher-judgment work gets stronger, and so do the people in it.

The AI KPI Most Businesses Get Wrong

Most companies I work with are already tracking Activity KPIs (are people using the tools?) and, if they’re a step ahead, Efficiency KPIs (is it saving time and cost?). Almost none are tracking the third tier: Amplification KPIs — whether AI adoption is actually making the business and its people measurably stronger.

That gap matters because of how Kai-Fu Lee’s own automation curve is built. The tasks AI reaches first are Repetitive (fixed steps, no branching — data entry, basic order-taking) and Routine (procedural, but with situational judgment inside known rules — scheduling, standard troubleshooting). What AI does not reach for a long time is Optimizing, Complex and Creative work — the judgment calls, the relationship repair, the strategic read on a customer or a market. If your KPIs only measure how much of the Repetitive and Routine layer got automated, you will never see whether your people are actually growing into the layer above it. You’ll just see activity going up while capability quietly goes untracked.

That untracked space is where businesses get into trouble. Not loudly. Quietly.

The Human Amplification Matrix — And Where Kopi & Co. Actually Sits

This is the framework I built the whole session around: plot every team on two axes — AI/Tool Adoption on one, Human Capability & Engagement on the other. Capability & Engagement isn’t a vibe; it’s a measurable average of four scored components: Critical Thinking, Empathy & Relationship, Trust, and Engagement, each rated 0–100. That’s what turns “our people are fine” from an assumption into a KPI.

To make this real this morning, I used our running case study, Kopi & Co. — a 4-outlet Malaysian coffee chain, 32 staff across Kitchen, Ops, Sales, Service and Admin. It’s a useful case precisely because it’s ordinary: no unlimited training budget, no dedicated AI team, just a business trying to grow without breaking its people. Some of their departments sit in Amplification. At least one sits closer to Quiet Implosion than the manager realised — fast tool rollout, but capability scores that hadn’t moved in months. Naming that on a matrix, instead of leaving it as a feeling, is what makes it fixable.

The engine behind moving a team toward Amplification is the OWN Framework: Objectives, Weekly Coaching, Numbers System. AI’s only job in this framework is the third letter — building the Numbers System. It doesn’t set the objectives and it doesn’t run the coaching conversation. Leaders and managers still own the human parts. AI just makes sure the numbers exist to have that conversation with, instead of relying on gut feel. If you want the mechanics of how that Numbers System actually gets built — without a data team — I walked through it here: Turn Raw Marketing Data Into a Live AI Dashboard.

Not sure which quadrant your own team is sitting in? Take the free DigitalAI Readiness Assessment →

The Role Mastery Index: Catching the Pattern One Person at a Time

The Human Amplification Matrix works at the team level. This morning I also introduced something new: the Role Mastery Index — a per-role, per-person scorecard that puts two numbers side by side for each team member: AI Mastery % (how well they’re using the tools available to them) and Human Mastery % (their composite score across the same four capability components).

Here’s why that pairing matters more than either number alone. In the Kopi & Co. case, one Kitchen team member’s Critical Thinking score made him the standout example of judgment holding steady even as AI took over routine prep-scheduling. A Front-of-House team member showed the same pattern through Empathy — her AI Mastery climbed, and so did her ability to read and de-escalate a difficult customer moment, because the tool freed up her attention rather than replacing her judgment. But a third case, in Admin, was the one that made the room go quiet: AI Mastery had climbed fast, while the underlying human score hadn’t kept pace. That gap — AI Mastery outrunning Human Mastery for one specific person — is Quiet Implosion showing up at the individual level, early enough for a manager to have a coaching conversation about it instead of a resignation letter later.

That’s the entire point of measuring this at all. Not to judge people. To catch the moment before it becomes a cost.

What AI’s Own Insiders Are Telling Us

The same week I was building this training, a story broke that put a much bigger version of the same pattern in front of the whole world. Jacob Coxon, a researcher who had worked on pretraining at both OpenAI and Anthropic, resigned publicly, warning that AI labs are “racing straight to self-improving superintelligence and gambling with our lives.” He wasn’t describing a Hollywood scenario. The mechanism he pointed to is recursive self-improvement — AI systems building increasingly capable successor systems faster than humans can meaningfully supervise the process. A colleague of his at Anthropic has put the odds of an AI-caused catastrophic outcome within the next decade at over 10%. CNN’s Anderson Cooper interviewed him about it this month.

To be clear-eyed about it: the AI safety community is genuinely split on both the timeline and the probability Coxon cites, and this is a live, unresolved debate among serious researchers, not a settled fact. I’m not sharing this to alarm anyone about the AI tools your business uses day to day — a booking chatbot is not a self-improving superintelligence. I’m sharing it because the underlying pattern he’s describing at the scale of an entire industry is exactly the same pattern the Human Amplification Matrix is built to catch at the scale of one business: capability racing ahead of the ability to govern and guide it responsibly.

If the people building the most advanced AI in the world are worried they’ve let capability outpace oversight, that’s not a reason to slow down your own AI adoption. It’s a reason to make sure your business isn’t running the exact same unmanaged gap internally — tools moving faster than the coaching, trust, and judgment needed to guide them well.

Why This Is Bigger Than Any One Business

This is the part I care about most, and it’s the reason I do this work the way I do it. AI education that only teaches cost reduction is building a smaller, more fragile economy one business at a time — because an economy is made of people who need jobs and income to be customers for someone else’s business. If every SME’s AI strategy is “use it to need fewer people,” the businesses down the street lose their customers too. It’s a shared ecosystem, not a private efficiency game.

The alternative isn’t slower AI adoption. It’s AI adoption that develops people into judgment work — Optimizing, Complex, Creative — faster than it automates them out. That’s what makes a business genuinely stronger instead of just leaner, and it’s what keeps the wider ecosystem healthy enough for everyone’s growth to actually mean something.

What to Do Differently Starting Monday

You don’t need a 32-person team or a data platform to start. Three moves, in order:

1. Name your Activity, Efficiency and Amplification KPIs separately. Most businesses only have the first two. Add the third, even if it starts as a simple quarterly rating of Critical Thinking, Empathy, Trust and Engagement per department.
2. Plot your teams on the matrix, honestly. The quadrant that matters most isn’t Amplification — it’s finding anyone sitting in Quiet Implosion before their next performance review does it for you.
3. Use the OWN Framework to close the gap, not headcount cuts. Set the Objective, run the Weekly Coaching conversation, and let AI’s Numbers System tell you who needs it most.

The businesses that come out of this decade stronger won’t be the ones with the most AI tools. They’ll be the ones who used AI to make their people more valuable, not less necessary — and who understood that a business only stays strong for as long as the people around it can afford to be its customers.

Frequently Asked Questions