AI
Professional services: use this framework to get ready for AI
Tom Leyden · 15 June 2025
Professional services firms ask us a version of the same question: “Where should we start with AI?”
The honest answer is rarely the latest model release. It’s a structural question — how the business actually operates and where AI can compress effort or improve quality. We use a four-stage frame to map any professional services firm:
Find · Win · Deliver · Learn
| Stage | What it covers |
|---|---|
| Find | Lead generation, brand, ICP definition, outbound, partnerships |
| Win | Proposals, pitching, estimation, scoping, onboarding |
| Deliver | Project execution, collaboration, billing, status, quality |
| Learn | Retros, IP capture, training, post-mortem analysis |
The premise is simple: every firm runs all four stages, but most have invested unevenly across them. The stage you’ve under-invested in is usually where AI lands biggest.
Step 1 — Map your current state
For each of the four stages, answer three questions:
- What systems do we use today?
- How integrated are they?
- Where does friction live — manual handoffs, duplicated data, missing visibility?
You’ll usually find the same pattern: the systems are fine in isolation, the integration between stages is where time and quality bleed out.
Step 2 — Pick the AI quick wins
A non-exhaustive list of where we see professional services firms get early payback:
- Find — AI-assisted proposal search; ICP-fit scoring; outbound copy drafting
- Win — Proposal drafting against a knowledge base; deal scoring against historic outcomes; meeting briefing
- Deliver — Meeting intelligence; technical co-pilots in your stack of choice; auto-generated status updates
- Learn — Natural-language knowledge base over closed projects; pattern-finding across engagements
The point isn’t that all of these matter to your firm — it’s that you choose one or two based on where your map showed friction in Step 1.
The trap to avoid
Most firms want AI wins but haven’t mapped where their core systems sit. They buy a tool, integrate it badly into a fragmented stack, and conclude AI doesn’t work for them. It does work — but only after the foundation question is answered.
Map first. Buy second.