Where Do You Start with AI?

You've decided AI is strategic for your company. Good. Now you face the first real decision: who gets it first?

Most companies get this wrong by giving licenses to everyone, or by waiting to see who asks for them. Both approaches fail. The data is clear on this: deliberate deployment to specific teams for specific work produces adoption rates three to four times higher than giving AI to everyone at once.

So where should you start? Look for teams doing high-volume, information-intensive, repeatable work. Customer support, where the same types of questions come in hundreds of times per day. Engineering, where code reviews and documentation tasks follow predictable patterns. Finance, where forecasting and reconciliation are necessary but time-consuming. Legal, where contract review involves scanning the same provisions over and over. These roles have something in common: the work is pattern-based and information-heavy. AI can actually help here.

What you shouldn't do is start with your executive team or your board. These roles involve judgment-heavy decisions, complex relationships, and political context that AI can't touch. AI can research options, but it can't tell you which decision feels right for your culture. It can draft a presentation, but it can't read the room. The people who benefit most from AI are the ones doing the repeatable knowledge work that fills their days — your ICs, not your executives.

Once you've picked your first team, you need training. Not the one-hour overview that most companies provide. Real training has three parts: how to use the tool effectively (mechanics), where it fits in existing workflows (integration), and what boundaries exist around data and escalation (awareness). Most companies stop at mechanics. They teach people how to prompt and call it done. But the gains come from integration and awareness — understanding when to use AI, when to step in yourself, and what information should never leave your company.

The final piece of your starting foundation is governance. Not a thick policy document that no one reads — you need the minimum viable version. Which tools are approved, for what use cases. What data can and can't go into AI systems. When to escalate to humans. That's it. Write it on one page. Anything more elaborate is theater.

What we covered: Where to start with AI — picking the right team, training properly, and establishing minimum governance.

Coming next: How do you measure whether AI is working? We'll look at what metrics actually matter, and why most companies track the wrong things.