We Said Yes to AI Before We Knew Why, and It Worked Anyway

We don't recommend doing what we did. But it's also the reason we're still here.

In late 2024, our team had no AI strategy. None. We had three different leadership offsites that were supposed to produce one, and each one ended with a deck that read like ChatGPT had written it about itself. The CTO was waiting for the CEO. The CEO was waiting for the board. The board was waiting for "more clarity from leadership."

What broke the loop was one of our customer support engineers. She quietly wired up a private Claude instance to her ticket queue, started drafting replies with it, and didn't tell anyone for six weeks. By the time anyone noticed, her resolution time was down 40%, her queue was the smallest in the org, and she had a Notion doc full of opinions about which prompts worked.

The accidental strategy

What we eventually realized was that she had skipped every single thing the offsites had told us we needed:

• No risk assessment
• No ROI projection
• No vendor RFP
• No "AI center of excellence"

What she had instead was an actual problem (the queue), an actual constraint (her time), and the freedom to try something cheap without asking for permission. That turned out to be a better strategy than the strategy.

What we did next

We stopped trying to write the strategy. We started looking for the next person like her. We made it explicit: if you can find a problem in your own workflow that an AI tool might help with, you have a $200/month budget and 3 days to prototype. No approvals, no governance review, no slide deck.

In four months we had 31 internal prototypes. About 8 of them turned into permanent tools. About 3 of those turned into actual product features. The rest got abandoned, which is how it should work. The cost of abandoning a $200 experiment is roughly nothing.

The lesson, if there is one

Most enterprise AI advice tells you to start with a strategy and work down to use cases. Ours was the opposite, by accident. Start with one person who has a real problem and the freedom to try something. Multiply that. Eventually you have a strategy, and it's better than the one you would have written, because it survived contact with reality.

The engineer who started it all, by the way, is now leading our internal automation team. She still doesn't have a deck.