How PE firms are building AI capability across their portfolio
Most private equity firms have accepted, at least in principle, that AI matters. The harder question - the one that separates firms generating real value from those generating slide decks - is how you actually build AI capability across a portfolio of ten, fifteen or twenty businesses with different sectors, different data maturity levels and different management teams.
The mistake most firms make is treating AI as an asset-level problem. They wait for portfolio companies to raise it, or they respond to it when a management team asks. A few firms have gone the other direction and mandated a single enterprise platform across the portfolio, only to find that what works in a £200m logistics business does not translate to a £600m professional services firm.
Neither approach is working. What is working is a deliberate, tiered operating model that sits between those two extremes - one that creates portfolio-wide leverage without imposing one-size-fits-all solutions.
This article explains what that model looks like, where the value is actually being captured and what questions operating partners should be asking right now.
The capability gap is wider than the hype suggests
There is a significant difference between a portfolio company that has deployed a copilot tool for its marketing team and one that has restructured a core business process around AI-driven decision-making. Most portfolio companies sit firmly in the first category and believe, incorrectly, that they are in the second.
The gap tends to manifest in three ways. First, AI adoption is happening at the edge of the business - in functions like marketing, HR and IT - rather than in the processes that drive revenue or margin. Second, there is no data infrastructure capable of supporting anything more sophisticated than basic reporting. Third, management teams are making AI investment decisions without a clear thesis on where value will actually come from.
A useful diagnostic exercise: ask the CFO of any portfolio company to show you what data they use to make their three most important operational decisions each month. In the majority of mid-market businesses, that data sits in spreadsheets, is assembled manually and arrives too late to influence the decision it was meant to inform. That is not a technology problem. It is a data infrastructure problem, and no AI tool fixes it.
Operating partners who understand this distinction move faster. They are not asking "which AI tools should we deploy?" They are asking "what decisions do we need to make better, and what does it take to make them better?"
Where portfolio-wide leverage actually comes from
The firms building durable AI capability across their portfolios are not doing it by installing software. They are doing it by building three things centrally: frameworks, talent access and data standards.
Frameworks means having a consistent methodology for assessing AI readiness at acquisition, identifying high-value use cases quickly and sequencing investment sensibly. A good framework lets an operating partner walk into a new portfolio company and within two weeks understand whether the data infrastructure can support meaningful AI work, where the highest-value intervention points are and what the realistic 12-month roadmap looks like.
Talent access means maintaining relationships with external specialists who can be deployed across the portfolio without each company bearing the cost of a full-time hire. A fractional Chief Data Officer engaged across three or four portfolio companies costs a fraction of three or four separate hires, and brings cross-portfolio pattern recognition that no single-company hire can match.
Data standards means establishing baseline expectations at acquisition - what data needs to be captured, how it should be structured and where it should live. This is unglamorous work, but it is the difference between a portfolio company that can deploy AI in year two of ownership and one that spends year two still trying to get clean data out of its ERP system.
A mid-market consumer goods business acquired by a London-based PE firm illustrates the point. At acquisition, the business had reasonable commercial data but no coherent data infrastructure. The operating partner spent the first six months establishing data pipelines and a reporting layer. By month nine, the business was using automated commercial reporting to identify margin leakage by SKU in near real time - something that had previously required three days of manual analysis. The AI capability was not the hard part. The foundation was.
The due diligence question most firms are not asking
Technical due diligence on AI and data has lagged behind financial and operational diligence. Most deal teams are still treating it as a subset of IT diligence, which means they are asking the wrong questions.
IT diligence asks: is the infrastructure stable and secure? AI and data diligence asks something different: is this business capable of generating the data assets that will drive value, and how much investment will it take to get there?
The distinction matters because data infrastructure debt is expensive and slow to pay down. A business with fragmented, siloed data across three legacy systems may look operationally sound but will require 12 to 18 months of foundational work before any meaningful AI capability can be deployed. That is a material consideration for a four or five-year hold.
The questions that should be in every deal team's process:
- What are the three to five decisions that drive most of the value in this business, and what data currently informs them?
- Where does that data live, how is it captured and how reliable is it?
- What is the current state of reporting and analytics, and who owns it?
- Has the management team articulated a view on AI, and does that view reflect commercial reality or vendor marketing?
- What would it cost and take how long to get this business to a point where AI could meaningfully improve those core decisions?
Firms that embed this into their diligence process arrive at completion with a clearer picture of the investment required and a faster path to value creation.
Building the operating model: a tiered approach
The firms that are furthest ahead have moved beyond ad hoc AI initiatives and built what amounts to an AI operating model for the portfolio. The architecture typically looks like this.
At the portfolio level, the firm maintains a small central capability - often one or two senior practitioners, supplemented by external advisors - responsible for setting standards, running diligence and identifying cross-portfolio opportunities. This team does not execute; it enables.
At the asset level, capability is built through a combination of upskilled internal talent and targeted external deployment. Not every portfolio company needs a Chief Data Officer, but every portfolio company above a certain revenue threshold should have someone with clear ownership of data and analytics. Where that person does not exist internally, a fractional or interim engagement fills the gap during the critical early period.
The sequencing matters. The most common mistake is attempting to deploy sophisticated AI tooling before the data foundation is in place. The correct sequence is: establish data infrastructure, build reporting and visibility, identify the highest-value AI use cases, deploy and measure.
A useful test: if a portfolio company cannot produce a reliable, automated weekly commercial report, it is not ready to deploy AI in any meaningful sense. Fix the reporting layer first.
Tools like natural language business intelligence platforms - which allow commercial teams to query their data without SQL or analyst support - are often the right first step. They reduce the time between data and decision, build internal confidence in data-driven approaches and surface the gaps in data quality that need to be addressed before going further.
What good looks like in 24 months
The operating partners who are executing this well share a few characteristics. They have a clear taxonomy of AI use cases by business type and maturity level. They are not chasing every new model release; they are focused on a small number of high-value, measurable interventions. And they have accepted that the constraint is not technology - it is data infrastructure, change management and sequencing.
Firms that continue to treat AI as an asset-level problem, responding to it only when management teams raise it, will find themselves in a difficult position at exit. Buyers are increasingly sophisticated about data and AI capability, and a business that has spent five years deploying copilot tools without building a coherent data foundation is not going to command the premium that a genuinely data-driven business will.
The cost of inaction is not visible in year one. It becomes very visible in year four.
If you are an operating partner building or refining your approach to AI across a portfolio, the right starting point is a structured diagnostic - one that maps the current state of data infrastructure and AI capability across your assets and produces a prioritised investment roadmap. Rodan runs portfolio-level AI diagnostics designed specifically for PE firms at this stage of the problem. Request a diagnostic conversation at rodan.io.
