Four in five US-based private equity firms have appointed a chief artificial intelligence (AI) officer to oversee internal adoption of the technology. Two-thirds expect to allocate at least 25% of their business budget to AI initiatives in 2026 compared to less than half last year.

These findings are from an EY survey published at the end of last year. Accompanying commentary noted that debate over adoption is over; investors must now show how they use AI to “create differentiation and not just ‘get to average.’”

It’s a message that has been taken to heart, but execution is not straightforward. Industry participants are under pressure to understand how AI can make a difference at the sharp end of their business while the technology itself continues to evolve. Ultimately, they must move from experimentation to institutionalisation – from trying everything to a recognition that less can be more.

“Sustainable competitive advantage comes not from adopting every new tool, but from deploying AI in a controlled and measurable way that delivers tangible business value while maintaining appropriate risk management and human oversight,” said Karen Sands, COO for private equity at Federated Hermes.

Global sponsors appear to be leading the charge, with AI increasingly playing a role at investment committee (IC) level – though not so far as to replace humans as the ultimate decision makers.

EQT’s AI engine, known as Motherbrain, has been elevated from deal-sourcing to IC sounding board. Advent is using AIEdge, a tool trained on 13 years of investment memos, including those for deals the firm passed on, to challenge assumptions underpinning every new opportunity put before the IC.

Warburg Pincus claims to have embedded AI across daily workflows, deploying an engine that analyses decades of investment memos, board materials, and presentations. This digital manifestation of 60 years of institutional knowledge provides strategic insights and informs new investment decisions.

Jesse Thomas, head of the global data science team at Advent, believes competitive strength derives from the combination of experience, resources, and scale. Previously, Advent sought an edge by investing in better secondary sources. Now, it embeds those sources into large language models (LLMs) and invests yet more in sourcing proprietary primary data.

“We can afford – and we do – run more surveys, make more customer calls and expert calls, and leverage a larger network of high-quality advisors,” he said. “That primary research muscle still benefits from our scale in a way that smaller GPs may find difficult replicate. So I think AI doesn’t close the gap.”

Middle market managers, meanwhile, argue that nimbleness counts for more than scale. Japan-focused D Capital – which has around USD 700m in assets under management to Advent’s USD 94bn – believes that commercially available IT systems are good enough for most purposes. It emphasizes flexibility, running AI adoption as bottom-up experimentation and top-down governance.

“As long as we control how we use these tools, we can do that quickly because of our size. That’s our strength. Large funds have to manage risks globally and localize across regions, so there’s a cap on how fast they can adapt,” said Megumi Matsutani, a partner and chief data officer at D Capital.

Starting small

Private equity’s most eye-catching announcements regarding AI often involve tie-ups with the likes of OpenAI, Anthropic, and Google Cloud. However, these are largely aimed at the portfolio company level. Internal adoption typically has humbler starting points: automation of process-oriented functions such as summarising information, project management, and drafting documents.

At Federated Hermes, for example, the quick wins have come in the back and middle offices and in investor relations. Sands described AI as an enablement tool that allows IR teams to prioritise “engaging with investors, responding to bespoke requests and delivering a higher-quality client experience.”

Adoption becomes more complicated when use cases move into the deal-making sphere, such as due diligence, deal evaluation, and investment memo drafting. Yet decision-support capabilities remain nascent and comparatively shallow because they need more advanced systems and richer data.

Daniel Angelucci, a managing director in Alvarez & Marsal’s performance improvement practice, warns that huge investment is required in agentic AI and automation over the next 10 years just to maintain baseline competitiveness. Managers must then spend still more on decision support initiatives that are potentially higher return but come with greater risk.

“We have these interesting decision support capabilities and we don’t really know quite what they look like either from a use-case perspective or a technology deployment perspective,” said Angelucci. “It’s not as certain as the automation piece. So, you’ll probably spend tens of millions of dollars there, and hopefully you’ll get some value out of it. That’s what a lot of people are facing.”

The systems developed by EQT, Advent and Warburg Pincus represent a step in this direction, consolidating resources so that investment professionals reach faster, more substantiated conclusions.

Other tools are scattered across internal ecosystems. Advent, for example, has dedicated AI agents that produce initial reports based on company details and industry comparables, help analysts compile and contextualise deal documents, conduct daily searches for updated information, and review PowerPoint presentations and Excel spreadsheets for accuracy before sending to senior partners.

According to Thomas, the first 70% of any due diligence analysis contains so many commonalities within the target sector that Advent has “always run” outputs that can be produced at the push of a button. These are run against a proprietary codebase to ensure consistency and guard against hallucinations.

“The real differentiator is the remaining 30% – you can go back and forth with the chatbot, follow your nose, and do analyses that weren’t in the template. It’s always a blend: consistent work upfront, leaving time for the bespoke stuff specific to each deal and each individual’s instincts,” Thomas said, adding that the existing set of AI tools saves at least one iteration loop.

The reality is that many of these initiatives are the product of grassroots experimentation, especially at smaller managers. Employees are equipped with licenses for OpenAI’s ChatGPT, Anthropic’s Claude and Google Gemini, given guidelines to follow on data privacy, and encouraged to find ways to do their jobs more efficiently. They share learnings and look to implement what works firmwide.

At emerging markets-focused mid-market private equity firm Affirma Capital, this approach prompted an India-based investment professional to build a proprietary pipeline tool that tracks financials, contacts, engagement history, and deal interest for potential target companies.

“We now have it in beta testing. If that works, we will roll it out to the rest of the platform and maybe develop it further as a CRM system,” said Ivo Philipps, a founding partner and COO at Affirma.

Data dividends

Other industry participants observe that operational workflows are easy enough to standardise; the obstacle to implementing common AI protocols is data.

“Data is not being standardised, and files are scattered in different places and personal laptops. Important knowledge is trapped inside people’s brains and needs to be put into a file format,” said an investment professional at a China-based private equity firm.

“Building an in-house system will cost a lot of money. The more important question is whether operational efficiency translates into fund performance.”

The problem is not unique to smaller managers. Leigh Coney, an AI engineer turned consultant at WorkWise Solutions, who has private equity clients, larger players are often still “going through Salesforce implementation.” In his experience, AI pilot programmes are most likely to fail because of poor data quality rather than insufficient AI capabilities.

While data preparation used to be a time-consuming process, now it can be achieved swiftly by any firm, Coney added. Moreover, early movers have demonstrated how getting it right can pay off.

BlackRock claims to have spent more than a decade developing proprietary, predictive models under a partnership between its private equity team and BlackRock Systematic, a business unit that applies AI, machine learning and alternative data to investment decisions. Johnathan Seeg, global co-head of BlackRock Private Equity Partners, suggests PE can learn from other asset classes in how to use data.

“We believe the growth of digital data around private companies has fundamentally changed private investing,” he said. “Many of the same techniques historically applied in public markets can now be extended to private markets to improve sourcing, diligence and monitoring.”

An ecosystem of providers now offer massive datasets tagged with metadata suitable for LLM consumption, enabling firms to run analytics against them. Alvarez & Marsal’s Angelucci believes PE investors should recognise the strategic value of this information rather than treat data as a commodity. If datasets cannot be generated internally, purchasing them can be worthwhile.

D Capital, founded as recently as 2021, made data a priority from the outset. Matsutani described a process whereby the firm studied how to store and structure data so it could flow through investment activities. This established the foundations for AI-driven analysis. On a practical level, D Capital records all internal meetings and keeps data in raw, flexible formats to minimize loss.

Jolt Capital, a Paris-based growth PE firm, underwent a similar process with AI platform Jolt Ninja, which launched in 2016. It captured five million companies and 10 million patents globally, using them to quantify target quality and competitiveness while benchmarking regional valuations.

For example, Jolt Ninja flagged better risk-return in Japan and South Korea compared to the US, which shaped the firm’s expansion priorities. According to Jean Schmitt, a managing partner at Jolt, the data is sufficient to identify a 25-company shortlist in Japan despite the firm not having a complete local team.

“With AI supporting our investment professionals, we do not need 20 people on the ground to begin investing in a new country. A small team can launch a fund, source opportunities, and provide meaningful value creation,” said Schmitt.

Organisational change

The shift from experimentation to institutionalisation of AI has broader implications for how private equity firms are organised. Having been tasked with assessing the impact of AI on different aspects of their jobs, investment professionals must adapt to a new normal as implementation gathers pace.

Bain Capital emphasized the need to rethink how work gets done rather than layering AI onto existing processes. Teams were asked how a business would operate if it were built today with AI in its DNA and an “AI twin” was developed as a digital representation of existing workflows.

Jordi Diaz, head of data science and AI at the firm, said it helped teams “identify where AI can fundamentally redesign workflows rather than automate existing tasks,” which in turn helped guide resources to the areas where they would have the most impact.

Senior-level buy-in is also essential, with WorkWise’s Coney observing that many leaders push AI without experimenting personally, creating a sincerity gap that undermines adoption. This is easier for those with technology backgrounds. For Jolt’s Schmitt, who has previously served as CEO of AI companies, conviction wasn’t a problem. It felt like a natural extension of personal experience.

“We are AI-native. Ninja came first, and the team came after. I hired 50 people after the AI was already in place. It was never intended to be a productivity tool layered on top of existing processes. Instead, it became part of the firm’s foundation,” he said.

“On average, employees spend around two hours a day using Ninja. Personally, I cannot imagine attending a board meeting without it because I constantly rely on it to answer questions and provide insights.”

Much as job types can be graded based on extent of AI impact, it is possible to categorise employees in terms of AI adoption. Advent created four personas: sophisticated users and enthusiastic learners present few problems; the challenges come with defensive sceptics worried about AI replacement and overconfident experimenters, who push the boundaries of approved use cases.

Advent’s response comprises education and guardrails. On one hand, it runs firm-wide training to demonstrate how AI frees time for more human-centric work. On the other, IT security and compliance teams promote responsible AI adoption. “We want to steer them in the right direction,” Thomas said.

Despite having a digital-native team with engineering and data science backgrounds, D Capital’s top-down approach to governance means it does not advocate firmwide adoption. There are two reasons. First, not every investment professional understands technological tools in depth. Second, technical expertise alone cannot address the strategic, governance, and security challenges.

Geography can present another obstacle to adoption, especially in China where access to foreign-developed LLMs is restricted. Several sources noted that local managers have purchased Claude licenses through overseas offices and employees use them via virtual personal networks (VPN) while in China. Others rely on personal accounts, although it raises data security and compliance concerns.

Angelucci observed that a lot of clients in China insist their data is not processed in the US, which inevitably limits access to some of the commercial tools.

“The solution is to find a provider in a jurisdiction that’s more appealing or to build your own model which is really expensive,” he added. “There are no magic bullets. People talk about how they can adjust their data processing and data sovereignty policies in contracts. That feels like a band-aid. You must solve it specifically around your company and what kind of risk you’re willing to tolerate.”

Fire the people?

The longer-term human resources question is whether AI adoption will lead to reduced headcount. Hypothetical scenarios abound in which private equity firms cut back on analyst and associate-level recruitment, building their organisations around a few rainmakers and a cluster of LLMs.

However, no firm interviewed for this story said they expected hiring rationalisation at any level of seniority in response to AI. Rather, they highlighted the limitations of the technology when it comes to matters of culture and human interaction.

It remains to be seen how this plays out in terms of middle and back office roles – companies across many industries are trimming administrators – but the notion of unpicking the apprenticeship model that sees investment professionals rise through the ranks is fiercely opposed.

“While AI is automating some research, document review and reporting tasks, we continue to view junior talent as essential to developing future investment and operational leaders,” said Sands of Federated Hermes.

“What is changing is the nature of the work. Junior professionals are spending less time on information gathering and more time interpreting outputs, challenging assumptions and applying judgement.”

For all the enthusiasm for what AI can do in terms of facilitating the investment process, there is still a keen appreciation of where it falls short. It’s not just a matter of verifying findings and strategic thinking but also bringing the personal touch.

“Someone still needs to meet with companies, build relationships with management teams, and evaluate opportunities in person. That’s difficult to replace with AI,” said Jason Raats, a director in the operations division of Europe-focused software investor Main Capital Partners.

“We’ll see what’s possible in five or 10 years, but today those activities still require people. It’s a relationship-driven business.”