Most AI adoption doesn't fail because the technology doesn't work. It fails because nobody did the design work first. Here's the framework I built — and what's running because of it.
A delivery motion built around heroics and tribal knowledge. Each implementation ran differently depending on who owned it. Time-to-value was slow, quality was inconsistent, and AI tooling was being bolted onto a process that had never been clearly designed. The technology wasn't the problem — the foundation was.
I rebuilt the implementation methodology from the customer outcome back. Defined what each phase was responsible for delivering. Shifted from "customer builds, team reviews" to "team recommends, customer validates." Removed steps that only existed because a better approach hadn't been available. Established ownership at every handoff.
Only after the process was clear. AI was applied to specific synthesis inflection points — pre-kickoff briefing, account update generation, audience analysis, handover documentation. Each tool targets a real friction point in a defined phase. None of them are compensating for unclear process design.
What is each step in the process actually responsible for delivering — and does the customer know that too?
Shift from "customer builds, team reviews" to "team recommends, customer validates." The team should be the expert in the room, not a facilitator.
Remove unnecessary steps first. Then apply AI at the points where synthesis is genuinely the bottleneck — not to mask process problems.
AI ingests account notes and call recordings to produce a structured customer brief before every kickoff. Replaces manual document hunting.
Takes call summaries and produces formatted stakeholder updates after every implementation touchpoint.
Accepts multiple audience files, identifies overlap and redundancy, produces a consolidation recommendation.
AI drafts the end-of-implementation record for clean handoff — platform config, decisions made, training status, research pipeline.
Produces customer-specific structured spreadsheets built around their platform setup and program design.
Custom agents that connect call recordings, manage project trackers, and surface to-dos — giving each implementation its own persistent AI layer that knows the account.
Identifies coverage gaps across training sessions and surfaces what still needs to be addressed before handoff.
Customer-specific deck generation from a templated base.
Most AI adoption asks: how do we support this process with AI? That's the wrong place to stop. The better question: which of our current practices only exist because a better approach was too expensive to build? AI doesn't just make existing processes faster. Sometimes it makes them avoidable. I'm working out which defaults that invalidates — and I want to think through it with people doing this work.
I think out loud on LinkedIn — follow the thinking there.
Ask Merlin anything
Usually responds instantly