Two Years of Advising on AI: The Patterns I Keep Seeing

I've spent the last two years working with 50+ professional services firms on AI strategy. Law firms, consulting shops, accounting practices, design agencies. Different industries, different models, similar patterns. The firms that are winning with AI do very specific things consistently. The firms that are struggling do something different. Here's what I've seen.

The Winners' Pattern

1. They picked one problem and solved it completely. Not five workflows. One. Legal document review, proposal generation, client intake, research summarization. They picked the one that hurt the most and focused on it. Pilot for 6 weeks. Integration for 8 weeks. Adoption for 4 weeks. By month 5, one complete, functional, adopted workflow. Then they moved to the next problem.

2. They measured from day one. Before they built anything, they timed the old process. Counted how many hours were spent. Identified who was doing the work and what it cost. Then, after deployment, they measured the same things. "We saved 12 hours per week" is a powerful statement. "The AI works well" is vague.

3. They involved the people doing the work. Not as an afterthought. As part of the design process. If your account managers are entering client data manually, they should help design the intake system. They know the pain points that engineers don't.

4. They built abstraction layers from the start. They didn't say "let's use Claude" and build everything on Claude. They built a system that could route to Claude, to GPT-4, to whatever came next. This costs 20% more upfront. It saves 100% later.

5. They budgeted for failure. Not all projects work. Some get 6 months in and realize the ROI is lower than expected or the adoption is slower than hoped. Winners budgeted for this. Three projects, expect two to work. Three teams trying, expect two to adopt. They built that into their expectations.

The Strugglers' Pattern

1. They tried to do everything at once. "Let's automate document review, proposal generation, client intake, scheduling, research, and billing." All five in the first year. They end up with nothing working well, no clear success, no momentum.

2. They didn't measure. Six months in, I ask "how much time is this saving?" and the answer is "I don't know, but it feels faster." Without metrics, you can't justify the investment to skeptics. Without metrics, you can't know if something is actually working or if you're just getting lucky.

3. They built for themselves, not for the users. An engineer designs an AI workflow. It's technically cool. It solves a theoretical problem. The people who actually do the work don't use it because it doesn't fit their reality.

4. They married a single vendor.** "We're an OpenAI shop" or "We're all in on Claude." Then a year later, pricing changes or a new model is better, and they're locked in. No flexibility.