AI for private equity portfolio companies: a repeatable playbook from diligence to exit
Operating partners keep getting the same question from their LPs: what is the AI plan for the portfolio? The honest answer for most funds is a collection of one-off pilots. A better answer is a playbook that runs the same way in every company, shows up in the value creation plan, and produces numbers you can defend at exit.
the short version
- Treat AI as a line in the value creation plan with an owner, a budget, and a target, not as an innovation side project.
- Ask AI readiness questions in diligence: data access, system ownership, process documentation, and who would run the work after close.
- Run the same sequence in every portco: process inventory, value sizing, one production system per quarter, measured against a baseline.
- Count only what finance can verify: hours removed from a queue, roles not backfilled, revenue from faster response. Pilots without baselines do not count.
01why portfolio AI usually disappoints
I see the same pattern across funds. Each portfolio company runs its own experiments. A CFO tries a chatbot, a sales leader buys an email tool, the operating team commissions a strategy deck. A year later there is a lot of activity and nothing anyone can put in a board pack.
The fix is not more pilots. It is treating AI like any other value creation lever: a defined process, an owner in each company, a target in the plan, and a measurement method agreed before money is spent. Funds already know how to do this for pricing, procurement, and sales effectiveness. AI fits the same mold.
02start in diligence
The cheapest time to understand AI potential and AI risk is before close. Add these questions to your operational diligence:
- Data access. Does the company own its data, and can it get it out of its core systems without a vendor's permission or a large fee?
- System landscape. How many systems run the business (ERP, CRM, billing, field service), and do they have usable APIs?
- Process documentation. Are the high-volume processes written down, or do they live in a few people's heads? Undocumented processes are harder to automate and riskier to lose.
- Current AI use. What tools do staff already use, with what data, under what policy? Unmanaged use is a data risk to price in.
- Labor concentration. Which teams are large, repetitive, and rule based? That is where the value is.
- Ownership. Who in management could run an AI program after close?
The answers shape the value creation plan and sometimes the price.
03the use cases that repeat across portcos
Industries differ, but the back office looks similar across most lower and middle market companies. These are the use cases I would rank first, in order:
- Order entry and invoicing. Reading purchase orders, emails, and PDFs into the ERP, matching to contracts, and flagging exceptions. High volume, rules heavy, easy to measure.
- Collections and cash application. Matching remittances, drafting follow-ups, prioritizing accounts. Shows up directly in working capital.
- Sales response and quoting. Faster first response to inbound leads and faster quote turnaround, built from the company's own pricing and past quotes.
- Customer support deflection. An assistant grounded in the company's own documentation through retrieval, with a clean handoff to people.
- Management reporting. Pulling from several systems into the weekly operating review without an analyst rebuilding a spreadsheet each Monday.
Each needs the same things: access to the system of record, a review step for exceptions, and a baseline measured before launch.
04the first 100 days
- Days 1 to 20: process inventory. List the twenty highest-volume processes. For each, capture volume, hours, cycle time, error cost, and the systems involved.
- Days 20 to 35: size and choose. Put a dollar value on each and pick one or two with high value and accessible data. Write a one-page scope for each. Our scoping guide covers what goes on that page.
- Days 35 to 50: governance. Acceptable use policy for staff, approved tools, data rules, and a named owner on the management team.
- Days 50 to 100: ship one system. Into production with real users, with working increments every two weeks so management sees progress, and measurement wired in from the first day.
The output at day 100 is one system running, a measured baseline, a policy staff follow, and a ranked list for the next two quarters.
05making it repeatable across the portfolio
The fund's advantage is pattern reuse. The second invoice automation should be faster and cheaper than the first because the playbook, the evaluation approach, and often much of the code carry over. To get that:
- Keep a shared playbook. Process inventory template, sizing model, scope template, policy template, measurement template.
- Keep a short approved vendor list. Models, platforms, and build partners vetted once.
- Own the code per company. Each portco must own its systems outright. That matters at exit, when a buyer asks what happens if the vendor goes away.
- Run a quarterly review across companies. What shipped, what it returned, what to copy next.
06measuring EBITDA impact without fooling yourself
AI value claims are easy to inflate. A buyer's diligence team will discount anything that cannot be traced to the P&L. Keep the method strict:
| Counts | Does not count |
|---|---|
| Reduced overtime or contractor spend on the automated queue | "Hours saved" that were never removed from payroll or redeployed |
| Roles not backfilled after attrition, documented | Headcount avoided in a hypothetical future |
| Revenue from faster response, measured against a baseline period | Revenue during a period when pricing or market also changed, unadjusted |
| Working capital improvement from faster collections | Pilot results on a hand-picked sample |
| All of the above, net of model usage and maintenance costs | Gross savings with run costs left out |
A case on record: on Pinned Golf, AI-augmented development reduced the engineering need from five engineers to one. That is the kind of outcome that holds up, because it is a payroll line that was never created, not an estimate.
07what to avoid
- Mandating a single platform for every company before you know their systems.
- Strategy without shipping. A roadmap with no system in production by the end of the first quarter is a warning.
- Vendor lock-in at the portco level. It reduces what a buyer will pay for the capability.
- Ignoring the people. Staff who fear automation route around it. Be clear about redeployment plans.
08where Insomnia Club fits
Our clients are typically companies with revenue of ten million dollars a year and up, which is where many portfolio companies sit. We can act as the build partner across a portfolio: running the inventory and scoping, then shipping systems through AI workflow automation and custom AI development, each priced as a fixed budget before work starts so the number fits the value creation plan. For portcos without an AI leader, our fractional AI officer engagement owns the program until management can. See the private equity page for how we work with operating partners.
common questions
How are private equity firms using AI in portfolio companies?
The practical uses are operational: automating back office queues such as invoicing, collections, and order entry, speeding up sales response and quoting, improving customer support, and giving management better reporting. The firms that get value run a consistent playbook across companies rather than separate experiments.
How do you measure the EBITDA impact of AI?
Establish a baseline for the specific process before anything is built: volume, labor hours, error cost, and cycle time. After launch, measure the same things. Count savings only when they show up as reduced overtime, roles not backfilled, lower vendor spend, or revenue that finance can attribute, and report run costs against them.
What should AI due diligence include?
Whether the company owns and can access its own data, how many core systems it runs and whether they have usable APIs, how well its key processes are documented, what AI tools staff already use and under what policy, and who in management could own an AI program after close.
Should a PE firm build a central AI team or work company by company?
Often a hybrid works best: a small central capability or partner that owns the playbook, the vendor list, and the guardrails, with execution inside each company owned by its management. Purely central teams lose context, and purely local efforts repeat the same mistakes.
How long does it take to see results from AI in a portfolio company?
A well-scoped first system can be in production within a quarter if the data is accessible. Results then need enough time in production to measure against the baseline, so plan for value to show in the following quarter's numbers rather than at launch.