The First 90 Days
Inside a Portco.
What a Fractional Chief AI Officer actually does, and what to expect on the cap table by day 91.
Mid-market portfolios need
one operator, not a lab partnership.
The largest funds have the budget to embed an AI lab inside the family of funds. Mid-market PE doesn't. And the portfolio companies inside a mid-market fund don't need one either.
What they need is one operator who can sit on the board, pick the right stack, and ship the agents that move the CFO metric, across the whole portfolio. One playbook applied to a dozen companies, instead of a dozen vendor relationships managed by an over-stretched operating partner.
That's the role of a Fractional Chief AI Officer. Below is what that role actually does in the first 90 days inside a portfolio company, not the deck version, the operational version.
Four phases. Ninety days.
Surface area mapping
- 30-minute interviews with the CEO, COO, CFO, and the head of revenue. The CAIO listens for the workflows that hurt, not the buzzwords that sound exciting.
- Quick read of the customer-success queue, billing exceptions, AR aging, and the SaaS stack inventory. The agent surface is where work is reactive, repetitive, and human.
- A short memo for the PE operating partner: 'These three workflows. This is the order. Here's why the obvious one is not first.'
Strategy gate. Build vs. buy.
- Run the build-vs-buy decision tree on each candidate workflow. Off-the-shelf AI tools win when the workflow is generic (transcription, basic customer FAQ). Custom agents win when it's the company's actual moat.
- Pick the vendors that will still be here in 18 months. Bias toward foundation-model neutrality so the portco isn't trapped if the AI lab landscape shifts.
- Define the one success metric the CFO will sign, usually hours-per-week reclaimed, exceptions-handled-without-human, or working-capital impact. Not 'accuracy.'
- Deliverable: a one-page Strategy Memo with the agent roadmap, the CFO metric, and the named risks (data, security, change management).
First agent in production
- Ship the first operational agent end-to-end. Not a chatbot, not a copilot, an agent that owns a workflow and reports back like a teammate.
- Every agent ships with two documents: an Agent Handbook (the rules the agent follows, scope, tool inventory, escalation, tone) and an Operations Manual (the rules the human supervisors follow, daily inspection, HITL gates, kill switches).
- First HITL review with the workflow owner before the agent runs unsupervised. The agent learns from the first 50,100 exceptions. The team learns the agent.
- Deliverable: one production agent, two manuals, and a weekly dashboard the operating partner can read in 90 seconds.
Security, governance, next two agents
- Security audit before scale: lethal-trifecta split (no single agent has private data + untrusted content + outbound network), RAG access controls, agent action sandboxing, API-key hygiene, audit log review.
- Governance mapping to EU AI Act, NIST AI RMF, and ISO 42001. The receipt trail the LP report needs.
- Spec out the next two agents based on what the first one taught the team. Start the build on agent two before the operating-partner update.
- Day 90: a board-ready brief, what shipped, what it cost, what the CFO metric did, what's next.
Five things mid-market PE
consistently gets wrong.
Buying the platform before the strategy
Fix: Pick the workflow first. Pick the agent second. The platform is the last decision, not the first.
Hiring a full-time CAIO at the portco before there's a strategy
Fix: Senior comp before there's a roadmap means the new CAIO spends six months writing the roadmap. Hire fractional until there's enough surface area to justify full-time, usually agent #4 or #5.
Letting the CTO own AI strategy as a side responsibility
Fix: Different muscle. The CTO is paid to keep the system up; an AI strategy is about deciding which work goes away. Conflict of interest, not malice.
'Pilot' that quietly never reaches production
Fix: Define the production criteria up front. If it's not in production by day 60, kill it. Pilots are not the deliverable, agents in front of customers are.
Treating governance as a Q4 deck instead of runtime code
Fix: Audit trails, kill switches, HITL gates have to be part of the agent build, not bolted on after the LP report. Security precedes governance. The lock comes before the receipt.
Seven questions for the
next CAIO candidate.
These work for fractional and full-time. If the candidate can't answer any one of them with a specific story, keep looking.
- Q01Show me an agent you've shipped that's been in production for at least 90 days. What does it do? What did it replace?
- Q02Walk me through the Agent Handbook from one of your builds. Don't show me the template, show me what's actually in it.
- Q03What's your stance on the lethal trifecta? What would you not let an agent do?
- Q04How do you map governance to EU AI Act / NIST AI RMF? Show me a real audit log, not a deck slide.
- Q05Who actually writes the code, you or a subcontractor? If subcontractor, who owns the agent handbook?
- Q06What's your kill-switch SLA? When the agent goes sideways at 2am, who responds and how fast?
- Q07What does month 4 look like if I sign for a year? Walk me through agents 2, 3, and 4.
If this brief reads like the work
you need done in your portfolio.
We work with mid-market PE firms and their portfolio companies on operational AI strategy and shipped agents. Same operator advises and builds. One Fractional CAIO can cover multiple portcos across a fund.