The three biggest mistakes organisations make when adopting AI

I've now run AI adoption programs at Eucalyptus and through Meridian across several industries. The failures cluster into three patterns, almost without exception.

1. Confusing access with adoption

Buying licences and telling people to use AI is not an adoption strategy. It's an access strategy. The assumption is that if you build it, they'll come — and that the tool's value is self-evident.

In practice, most knowledge workers don't know what to use AI for beyond generic tasks. They need structured use cases, workflow integration, and someone to show them concretely how the tool changes their day.

Adoption comes from designing specific workflows, not from a Slack announcement.

2. Under-investing in context

This is the biggest one. Most organisations deploy AI with zero investment in context — no custom instructions, no domain knowledge baked in, no structured way for the tool to understand the organisation.

The result is that every user interacts with a generic assistant. The tool is technically capable, but it doesn't know your terminology, your processes, or your constraints. It gives generic answers.

Context engineering — building a layered system of persistent knowledge and live data access — is the single highest-leverage investment in AI adoption.

3. No measurement strategy

If you don't measure it, you can't improve it. Most adoption programs lack a telemetry strategy, so they can't answer: who's using what, which use cases are generating value, and which teams are falling behind.

You can't fix what you can't see. Build observability into the rollout from day one, not as an afterthought.