Before we told a single client how to adopt AI, we figured out what it actually takes to run on it ourselves. This is the story of what we invested in early, what we built (and rebuilt), and what we're still adjusting as the technology outpaces the plan.
Seven chapters, one firm, and every hard call in between
We had a decade of client work in our vault before AI ever showed up.
Playbooks, templates, a running record of what worked and what didn't across thousands of projects. Every AI conversation runs into the same wall: rubbish in, rubbish out. Luckily, we already had the good stuff.
Read the chapter >>
People stopped asking "is this okay?" and started asking "what else can this do?"
NH AIMS passed 90% adoption across the firm, and we knew we had to stop treating it like a project. We moved its running costs into normal SG&A, the same line as the rest of our software.
Read the chapter >>
Tokens are real money, and the bill arrives after the value shows up.
AI spend rises with usage, and vendor pricing shifts without warning. We built our own version of FinOps: live cost visibility, spend tiers by persona, and a right-tool-for-the-job matrix.
Read the chapter >>
We gave ourselves thirty days to get Claude into 1,400 people's hands.
Eight workstreams, each with a real owner, covered licensing, training, and storytelling. We stopped pretending there was one AI tool: Claude for this, Copilot for that, NH AIMS for the other thing.
Read the chapter >>
