Benchmarks Don't Drive AI Adoption. Incentives Do.

Models are improving every month. Smarter. Faster. More accurate. Yet adoption is not accelerating at the same speed. Because adoption doesn't follow capability — it follows incentives.

This is part of the Hidden Variables in AI Strategy series — examining the factors that determine whether AI initiatives deliver real organisational value or remain stuck in perpetual pilot.

The Productivity Gap

A study from the National Bureau of Economic Research showed generative AI improved customer support productivity by approximately 14%.¹ The tool worked. The data was clear. Yet rollout across industries remains uneven — not because the model was insufficient, but because the people being asked to adopt it had reasons not to.

People don't adopt tools based on intelligence. They adopt based on self-interest. And right now, for many workers outside tech, the self-interest calculation on AI is unclear at best and threatening at worst.

Where Adoption Is Happening — and Why

Some organisations are moving decisively:

• Google has integrated AI leverage into performance expectations — making AI proficiency part of what it means to perform well, not just an optional add-on.
• Microsoft has embedded Copilot into daily workflows — reducing friction and making AI the path of least resistance rather than an extra step.
• In tech broadly, AI proficiency is quickly becoming table stakes — a signal of competence, not a threat to it.

This is not hesitation. This is incentive alignment. These organisations have changed what it means to do a good job — and AI use has followed naturally.

Outside tech, the picture is different. Adoption slows — not always because models are weak, but because incentives are unclear. And underneath that? Fear.

We've Seen This Pattern Before

When computers entered Indian banks in the 1980s, protests erupted over job-loss concerns.³ Bank unions organised. Adoption stalled. Once reskilling programmes and role clarity followed — showing employees how their roles would evolve rather than disappear — digitisation accelerated rapidly.

The technology was the same throughout. What changed was the incentive picture for the people being asked to adopt it.

Anxiety first. Clarity next. Adoption after.

AI feels similar. The arc is familiar. The question is how long the anxiety phase lasts — and whether organisations accelerate or delay the clarity that follows.

The Real Unlock

The real unlock for enterprise AI adoption will not be a bigger model or a better benchmark. It will be clarity on three things:

• How performance is measured: If AI use is expected but not reflected in how success is defined, adoption will be inconsistent. People optimise for what is measured.
• How roles evolve: Workers need a credible picture of what their role looks like after AI — not just reassurance that jobs won't disappear, but an actual description of what new value they are expected to create.
• Who benefits most: If productivity gains from AI are captured entirely by the organisation while workers bear the retraining cost and performance pressure, the incentive to adopt is weak. Sharing the benefit — through compensation, autonomy, or recognition — changes the calculation.

Models improve output. Incentives drive behaviour. Behaviour drives adoption. Organisations that understand this sequence will move faster — not by buying better AI, but by designing clearer incentives around the AI they already have.

Disclaimer: The views expressed are those of the author and are for informational purposes only. They do not constitute financial, legal, or investment advice.