The 3 Things Every Organisation Gets Wrong When They Start With AI

I've spent the last two months building products with Claude, Gemini, Grok, and OpenAI every day. Before that, I led enterprise technology programmes in defence, aerospace, and aviation. From both perspectives, I see the same three mistakes organisations make when they start with AI.

Mistake 1: Starting with the technology

Most organisations begin their AI journey by picking a tool. They get a ChatGPT Enterprise licence, run a few demos, and then try to figure out where to use it. This is backwards. You should start with your highest-friction workflows and then find the AI tool that addresses them. Technology-first adoption leads to expensive experiments that don't stick.

Mistake 2: Running pilots that go nowhere

The pilot programme is where AI adoption goes to die. A small team experiments with AI for 6 weeks, produces a report, and then nothing happens. The problem isn't the pilot — it's that most pilots are designed to evaluate the technology rather than to deliver a working workflow change. A good AI implementation should deliver measurable time or cost savings within 2-4 weeks, not a PowerPoint deck.

Mistake 3: Treating AI as a tool instead of a workflow shift

Adding AI to an existing workflow is like putting a turbo engine in a car that still has a horse harness attached. The real gains come from redesigning workflows around AI's capabilities — not from automating individual tasks within broken processes. This requires someone who understands both the technology and the organisational dynamics. That's rare.

The path forward

Start with your most painful workflows. Pick one. Implement AI-augmented processes that deliver measurable results in weeks. Use that success to build momentum for broader adoption. That's the playbook that actually works.