The plan · Composite
Industrial distributor
Three stalled pilots and a board deadline, turned into a 90-day plan in two weeks.
Read the case study →Composite case · Fractional retainer
The AI lead quit.Three pilots drifted.
A healthcare software company lost its senior AI hire after nine months. I ran the diagnostic, then stayed on the Fractional retainer until the claims workflow shipped and a permanent owner signed.
Illustrative composite, not a real client. The shape of the work is drawn from real patterns. The company is not a named client and the numbers are illustrative. For work that is real and checkable, see baxie.
01 · Before and after
Measured againstwhere they started.
40% → 58%Support tickets resolved without a human
30% → 84%Reps using the call summaries
0 → ~31%Claims routed without a triage step
HiredPermanent AI owner, signed before handoff
This composite is a healthcare SaaS company with about 180 people. The senior AI hire left after nine months. Three pilots had no owner and AI work lived nowhere.
A support copilot was deflecting tickets with no metric anyone had defined. Sales call summarization sat at low adoption. The bet that actually mattered, claims adjudication assist, was a slide deck with no prototype. The CTO was stretched thin, reluctant to hire another senior person who might quit, and had a board AI update due in 90 days.
02 · The problem
A portfoliowith no owner.
Nobody owned it
The senior AI hire left after nine months. Three pilots had no clear owner and no accountability chart.
A number nobody measured
The support copilot ran at about 40% ticket deflection, but nobody had defined or instrumented the success metric.
Built to replace the rep
Sales call summarization sat at 30% rep adoption. It was built to fully automate, reps did not trust it, and a draft-then-edit version was never built.
The real bet was a slide
Claims adjudication assist had no prototype and no owner.
A board date and a tired CTO
The CTO did not want to hire another senior person who might quit, and the board AI update was 90 days out.
03 · Step one
Two weeksto the plan.
It started with the AI Program Diagnostic, like every engagement. I mapped the workflows each pilot touched from system data and interviews with the support, sales, and claims teams. Then I measured where each one stood before anything else changed.
- A baseline for every line. The support copilot got a defined metric: 40% of tickets resolved without a human. Summarization adoption was measured at 30%. Claims triage was fully manual.
- A named owner for every line. Ownership went to the function leads through an accountability chart, so the portfolio would not depend on one hire again.
- A build type for every line. Agents where a line needed judgment, plain code where a rule did the job, and a human making the call wherever a wrong answer touched a claim.
- A kill list. One orphaned tool with no real demand got stopped. The board got a quarterly update template the CTO could defend without me in the room.
04 · The retainer
Ship the bet.Hire the owner.
The CTO kept me on as Fractional Head of AI Transformation, one to three days a week. The plan set the order of work.
| # | Workflow | Before | After | Build type |
|---|---|---|---|---|
| 01 | Support copilot | 40% of tickets resolved without a human, never instrumented | 58%. About 18% of the queue stopped reaching a person, so the team absorbed growth without hiring | Agent, with escalations to a human |
| 02 | Sales call summaries | 30% of reps used it | 84%. Enough to defend the license at renewal and plan off the notes | Agent drafts, rep edits |
| 03 | Claims adjudication assist | A slide deck. Every claim sorted by hand | About 31% routed without a triage step, with an audit trail on every decision | Code parses intake, agent suggests codes and routes, adjudicator decides |
| × | Orphaned internal tool Kill | No owner and no real demand | Stopped | None |
Agents started by recommending. The claims workflow earned more routing autonomy only as its evals held up, and the adjudicators kept the final call.
- The layer the team and the AI both read from. A domain glossary, customer archetypes, decision logs, and the automation patterns in use. The team writes to it and the agents read from it.
- The permanent owner. I wrote the job description against what I learned in the seat, screened candidates, and advanced finalists to the CTO. The full-time hire signed during the engagement. A warm handoff with office hours and regression triage meant the new owner inherited a running system.
05 · Further reading
The othertwo examples.
The build · Real
baxie
The AI system I run under my own name, guardrails and limits included.
See the build →06 · FAQ
Questions.
Our head of AI quit and the pilots are stranded. What happens first?
The diagnostic. Every stalled piece gets mapped and measured, then lands either on the 90-day plan with a named owner inside your team or on the kill list with a reason. A stranded portfolio usually has one or two things worth saving and several worth stopping, and nobody has had the authority to say so.
Can we hire a permanent head of AI while a fractional one is embedded?
That is the intended path. I work the search alongside your recruiter, write the job description against what I learn in the seat, support panel interviews, and hand off when the right person signs. These engagements usually end inside nine months for exactly that reason.
What does a fractional head of AI actually do week to week?
A weekly executive sync, availability on the calls that move the program, hands-on work on the build with the most at stake, and training running in parallel. One to three days a week. The standing rule is that every workflow shipped names a human accountable for its output.
Who should own AI at a 50-person company with no CTO?
Someone on the executive team who can stop projects as well as start them. The common failure is handing AI to whoever is most enthusiastic about it, which produces a champion with no authority. If nobody there can kill a project, the ownership is not real.
What happens to the work when the engagement ends?
Your team runs it. The portfolio, the vendor decisions and their reasoning, the standards, and the training all stay. If your team cannot take it at handoff, the engagement is not finished.
What are the terms?
Every engagement starts with the $2,500 diagnostic, credited in full if you move to the build or a retainer within 30 days. The Fractional retainer runs from $12,000 a month for one day a week, up to three days, with a 6-month minimum and quarterly renewal after. There is a 14-day mutual exit from day one, 30 days notice after month three, and days not worked are refunded.
GAT.
No owner and a board date. Start here.
If your AI work lives nowhere, the diagnostic finds what is worth saving. The retainer ships it and hires the person to own it after.
Book 30 minutes with EdwinNot ready to talk? Score your AI program first. Twelve questions, about four minutes, scored breakdown by email.