Investment committees are starting to ask a question most operating partners are not prepared to answer: how AI actually improves your portfolio company’s financials?
Not “what tools are deployed.” That is the easy answer, and most teams have a version ready. The harder question is whether AI has a positive impact on performance. Whether revenue per employee is improving because more is achieved with less. Whether net revenue retention is holding as AI changes how customers are serviced. Whether the portfolio companies absorbing the most AI investment are building toward better exits, or just toward better updates at the next LP meeting.
I got a sharp reframe on this from a senior operating partner at a growth-equity fund. His argument: the traditional PE metrics such as MOIC, EBITDA margin, ARR growth, IRR, are still important, but they are lagging indicators when it comes to AI. By the time they show the effect, good or bad, you’ve already lost response time. The metrics that will matter as leading indicators are Revenue per Employee, Net Revenue Retention (NRR), and how quickly we can attain both.
Here is why that reframe holds up, and what to do about it.
The traditional metrics are still useful. They just show you last quarter, not next year.
EBITDA and ARR are not the wrong metrics. But in the context of AI utilization, they are lagging metrics. If AI is improving your team’s productivity, it will eventually show up in EBITDA margin as headcount costs shift. If AI is helping your team win and retain more business, it will eventually show up in ARR. But “eventually” is the operating word. By the time you see the delta in EBITDA, you may have already been six months into the wrong trajectory.
Revenue per employee and net revenue retention, on the other hand, provide an earlier signal. They measure the cause, not the downstream effect. If the AI investment is doing what it promised, revenue per employee should improve before improvements to EBITDA margin are detected. If the operational changes are preserving customer quality, NRR will either remain unchanged or improve..
The practical risk of relying only on lagging indicators came up directly at a roundtable of PE operating partners. The conversation had turned to AI-assisted financial analysis, and someone put it plainly: “You still have nightmares. You just get to the scary parts faster.”
That line is worth sitting with. AI does not eliminate bad data or bad judgment. It simply accelerates data and judgement. A revenue forecast assembled from inaccurate Salesforce data gets fed into an LLM-assisted tool. The model produces a number. Nobody validates the inputs because the output looked credible. The CFO sees an inflated forecast, decides headcount needs to grow, and hiring decisions follow. The revenue never materializes. The team over-hires. People blame the AI.
The AI did not get it wrong. The workflow upstream was broken, and the speed of the output made it harder to catch before it caused damage. EBITDA will look fine through all of this, right up until it doesn’t. That is the case for measuring something closer to the work.
Revenue per employee: the goal is a positive change, year over year.
Revenue per employee is both a target and a directional measure. You are not chasing a specific dollar-per-head benchmark against an industry standard. You are pushing toward a positive change in the ratio compared to your own prior period. A portfolio company with 80 employees and $12M in revenue this year should want that number to be better next year. That means growing revenue with the same team or sustaining revenue with a leaner, higher-capability one.
There are two levers. Grow revenue through better GTM execution: sharper segmentation, faster proposal cycles, tighter pipeline analysis. Or reduce cost by redefining who you hire and what the role requires. AI enables both, but only if someone is driving the integration toward a measurable outcome rather than hoping adoption happens on its own.
A concrete version of the cost lever: a CFO with the right prompts and a capable AI account can run financial modeling and produce reports in a fraction of the time previously spent by a junior FP&A analyst. That position either goes unfilled, or the budget shifts toward a higher-capability role. Applied to RevOps, the same math holds. Two to three analysts become one, with the right tooling and proper workflow design.
The hiring implication is direct. New employees at AI-enabled companies need to be AI-capable, not just AI-adjacent. They need to know how to clean data before feeding it to a model. They need to understand what accurate output looks like so they can push back when the model surfaces something wrong. There is a real difference between a human in the loop and a knowledgeable, experienced human in the loop. That distinction is what separates companies where AI moves the ratio from companies where AI produces confident-sounding errors at higher speed.
The forecast story above makes the mechanism concrete. If the right person had owned data quality and validation in that workflow, the bad inputs would have been caught. Revenue per employee would have flagged the problem early: headcount growing while revenue holds flat. The AI investment would have had a measurable outcome to point to. Instead it became a cautionary story about trusting the model too quickly.
Net revenue retention tells you whether effectiveness held while you made the change.
Revenue per employee measures efficiency. Net revenue retention measures the other side: did quality hold while the team changed how it operates? The two metrics work as a pair. Neither tells the full story without the other.
When AI accelerates customer health analysis, it only creates value if the people running the analysis know what they are looking for. An account manager using a tool to flag churn signals needs to understand which signals actually predict churn before the renewal conversation, not just which ones the model surfaced. If the team is following AI-generated recommendations without understanding the logic, customer relationships erode in ways that will not appear until a renewal conversation goes badly.
NRR answers the question EBITDA cannot: did customers stay through the transition? Did the change in how the company operates preserve the quality of service they purchased?
For most PE-backed businesses, it is the most defensible proof point with the investment committee. A portfolio company that improved its revenue per employee and held NRR above 90% through an AI-enabled operational change has an exit-relevant story that stands on its own. Not “we deployed tools.” We improved productivity and kept the customers. That combination is what PE firms are starting to look for.
Two out of seventeen. That is where most portfolios stand today.
At two operating partner roundtables over the past twelve months, with seventeen PE firms represented, only two said they had successfully stood up an AI organization inside their portfolio. Not launched a pilot. Stood up, with a named owner, a governance structure, and cross-functional teams working toward measurable outcomes.
The two who got it right assigned AI leads at each portfolio company. They run monthly cross-fund learning sessions: every problem solved at one portco gets shared across the rest of the portfolio within thirty days. The structure exists. The measurement loop exists.
The other fifteen tried organically. Individual employees adopted tools. Teams ran their own experiments. Nobody owns the outcome. One operating partner described it as everyone solving their own side of a Rubik’s Cube, with the rest of the cube becoming more chaotic as each person works independently.
The pattern that shows up consistently in that group: tools were treated as a productivity initiative rather than an organizational design question. Someone in IT or operations was handed the rollout. No one built a link between tool adoption and the leading indicators that matter at exit. Leadership checked the box. Employees used the tools however made sense to them individually. The improvement in revenue per employee never materialized because nobody was building toward it.
That year is recoverable. But the two who are ahead are not standing still. The gap is not about tools access. It is about who owns the measurement and whether the organization is designed around improving the ratio or just deploying the software.
Three things to do before the next board meeting.
Start with what you can measure today. No projections, no roadmap slides. Just the current state.
Step 1: Take the snapshot.
- What is revenue per employee at each portfolio company right now?
- What is net revenue retention at each?
- Can you pull those numbers without digging? If not, that is the first signal.
Step 2: Survey the portfolio.
A focused, nine-question instrument a CEO or RevOps lead can complete in under fifteen minutes. What you are looking for:
- Which LLMs are in use, officially and otherwise?
- What does the RevOps tech stack look like at each company?
- Are there AI tools being used that leadership does not know about?
- Which workflows are AI-supported for revenue decisions, and who owns the quality of those decisions?
- Is there a named person responsible for AI adoption at each company?
Step 3: Find the champion.
This person already exists at most portfolio companies. They have been running their own experiments, building prompts, figuring out what works in their specific workflow, without waiting for a mandate from above. They just do not have authority to replicate what they have built. Surface them. Give them standing. Pull them into a cross-portfolio task force with the champions from the other companies.
That is the model the two-of-seventeen are running. Not a top-down rollout. A diamond-identification exercise: find who is already doing it right, give them resources, and let the replication happen from the people who already solved the problem.
Start with the baseline survey.
The nine-question portfolio AI readiness survey is built to be sent directly to a portco CEO or RevOps lead. It covers the LLMs in use across the business, the RevOps tech stack, additional AI tools beyond chatbots, use cases by workflow, the broken processes costing revenue, and whether there is an identified AI champion with real authority. The output is a clear baseline across the portfolio, and it tells you which companies are on track and which ones are still hoping the tools sort themselves out.
