A fractional Chief AI Officer for a mid-market private equity firm
An industrial private equity firm with about $1.7B under management had no AI capability anywhere in the firm. As fractional Chief AI Officer, 3Pillars scored the opportunities, won an investment-committee AI charter, then built three production tools on Claude: a sourcing desk, a fundraising desk and diligence triage.
- Claude in production
- Claude API (Claude Sonnet 4.5) for natural-language-to-SQL, answer narration and investor briefs; Claude Code; Agent Skills; a self-hosted MCP server; Claude Opus in the build sessions
- Status
- Advisory April to June 2026; all three tools in daily use since June; hardening under a direct build contract July to September 2026
- Users
- All 22 employees, on per-user keys with passkey login and audit logging
- About 20 interviews and a two-by-two of difficulty against competitive advantage produced an AI charter the investment committee ratified; the framework killed low-value projects before spend.
- The sourcing desk groups about 750 proprietary targets from live CRM fields and ranks the daily outreach list; the manual re-tagging project was cancelled. The fundraising desk covers 200+ limited partners with generated meeting briefs.
- A fixed-cost build replaced quotes starting at $250,000; diligence screening runs at about $100 in API cost per data room against up to $200K in outside cost; an estimated 10 to 15 staff hours a week returned.
The problem we took
Deal teams reviewed diligence data rooms by hand, thousands of pages read too late in the deal window, with dead deals hitting the bonus pool directly. The deal CRM was stale and mislabeled: roughly 750 proprietary targets sat unworked, many tagged wrong, so outreach stalled and follow-ups slipped. Fundraising staff tracked 200+ limited partners in spreadsheets with no reliable view of who to meet next. Partners disagreed on what AI was worth, and one was openly skeptical. Outside vendors quoted $250,000 floors for builds the firm suspected were smaller.
Competing firms had announced AI programs and frontier labs were partnering with large sponsors to deploy operational AI into mid-market portfolios. The partners wanted their own capability before it became a commodity, and a defensible view of where AI pays before any spend.
What the client already had
A partner’s own diligence prototype, useful but unshared; the deal CRM with 35,000+ interaction records; placement-agent CRM and Outlook history for fundraising; and the data rooms themselves. The firm keeps the tools, the data and the governance; its partners now build their own Claude skills.
What we built
First, discovery: roughly 20 structured interviews across deal, business development, fundraising and portfolio teams, each candidate opportunity scored on a two-by-two and packaged into an AI charter the investment committee ratified. Then, build. A sourcing desk that answers plain-language questions over the deal CRM, re-anchors follow-up clocks on logged calls and ranks the daily outreach list. A fundraising desk covering 200+ investors with pipeline views, tear sheets and meeting-prep briefs. A diligence triage tool that reads every file in a deal room and flags issues by category, hardened from the partner-built screening skill. Monthly business-development reporting and intermediary one-pagers generated on demand from one verified data source.
Each tool runs on Claude through a per-user, passkey-gated web app with server-side audit logging, plus a Claude skill and a self-hosted MCP server so the firm’s own Claude sessions query live data directly. Every plain-language query runs a measurable accuracy battery against ground-truth SQL before changes ship. 3Pillars provisioned access, trained daily users, managed build-vendor selection and pricing, and coached the partners as they began building their own skills. The contract expanded from the CAIO engagement to the build, and then to a portfolio company.
What it returned
Fundraising meeting preparation that took hours now generates in minutes from live data. The ~750-row manual re-status project, days of staff work, was eliminated. Monthly reporting became on-demand, reconciled against CRM ground truth and validated against 29 top intermediaries. Combined, an estimated 10 to 15 staff hours a week across deal, business development and fundraising roles. The named skeptic now uses the sourcing desk daily and builds his own Claude skills.
The investment committee ratified an AI ambition statement, rare alignment at partner level, and every AI query now answers from one governed source, not from personal copies.
Figures are compiled by 3Pillars from server logs, the release test battery and reconciliation audits, June to September 2026. The client has not been asked to be named.
Run this inside your organization.
A 30-minute working session. We walk through what an engagement like this one would look like for you.