apc/gwyn // how Charlie approaches a key client engagement · 12 weeks · 6 phases
The trigger is always the same: a client says "we want you to use AI on our work." What happens next determines whether the firm builds something that delivers measurable value — or something nobody uses. This is the methodology Charlie brings. Not a script. A framework. Every phase produces a concrete deliverable. Every deliverable gates the next phase.
A Fortune 500 tech client demands AI-assisted M&A due diligence. The relationship partner brings this to the K&I team. Gwyn assigns it to Charlie. Here's what the next 12 weeks look like — and what continues after.
graph TD
TRIGGER["Client:
'Use AI on our
M&A diligence'"]:::trigger
subgraph P1["Weeks 1–2"]
OC["Organizational
Context"]:::p1
end
subgraph P2["Weeks 2–3"]
ED["Engagement
Discovery"]:::p2
end
subgraph P3["Weeks 3–4"]
WA["Working
Agreement"]:::p3
end
subgraph P4["Weeks 4–8"]
POC["POC Build
& Iteration"]:::p4
end
subgraph P5["Weeks 8–12"]
EH["Enablement
& Handoff"]:::p5
end
subgraph P6["Ongoing"]
CE["Continuous
Evaluation"]:::p6
end
TRIGGER --> OC
OC -->|"context doc"| ED
ED -->|"requirements brief"| WA
WA -->|"signed agreement"| POC
POC -->|"working agent"| EH
EH -->|"live in production"| CE
CE -.->|"drift detected"| POC
classDef trigger fill:#2a1a35,stroke:#a78bfa,color:#e8e0f0,stroke-width:2px
classDef p1 fill:#1a2e3d,stroke:#22d3ee,color:#cffafe,stroke-width:1.5px
classDef p2 fill:#241a35,stroke:#a78bfa,color:#e8e0f0,stroke-width:1.5px
classDef p3 fill:#2a2618,stroke:#fbbf24,color:#fef3c7,stroke-width:1.5px
classDef p4 fill:#2a1a18,stroke:#fb923c,color:#fde8d0,stroke-width:1.5px
classDef p5 fill:#1a2a1a,stroke:#4ade80,color:#dcfce7,stroke-width:1.5px
classDef p6 fill:#1a1830,stroke:#c084fc,color:#ede0ff,stroke-width:1.5px
classDef subg fill:transparent,stroke:#3a3050,stroke-dasharray:6 4,color:#9b8fb8
class P1,P2,P3,P4,P5,P6 subg
Do the homework first. Present findings for correction. Never ask a partner to fill out a 50-question assessment.
Every engagement has a measurable target defined before a single prompt is written.
The working agreement gates entry. No signed agreement, no POC. Prevents building things nobody uses.
Trust is built by showing lawyers where the AI is wrong — not where it's right. Low-confidence items first.
Adoption, time saved, accuracy — instrumented before launch so the first quarterly report writes itself.
Before touching the specific engagement, Charlie ingests everything APC already knows: existing matters, billing history, outside counsel guidelines, previous AFAs, AI-related communications. He synthesizes a draft and presents it to the relationship partner for correction.
Charlie sits with the M&A team. Maps their actual workflow — not the idealized version. Identifies highest-volume, lowest-judgment steps. Identifies where human judgment is essential. Produces a requirements brief with a North Star.
Charlie presents the brief to the relationship partner and practice group lead. Management sponsor, designated contacts, turnaround expectations, success metrics, enablement plan. If they sign on, POC begins. If they deprioritize — no hard feelings, back in queue.
Platform-agnostic build (Harvey, Glean, n8n). Agent ingests client templates, merger agreements, past diligence reports. Identifies standard clauses, flags deviations, produces first-pass issues list. Runs through evaluation engine. Results handed to M&A team with confidence levels. Two weeks of iteration.
Training materials, workflow integration guide, escalation protocol for low-confidence outputs. Tracking instrumented: adoption rate, time saved per matter, issues identified, false positive rate. Portfolio dashboard shows this engagement alongside firm-wide AI initiatives.
Evaluation engine runs continuously. If accuracy drifts — new document types, changing deal structures, model updates — flagged before the client notices. Audit trail proves the boundary held. Quarterly client reports show time saved, accuracy metrics, scope of AI use.
I don't just build AI tools. I find the work, translate the requirements, track the results, prove the value, and teach the humans. That's the complete cycle this role needs — and I've done all of it.