AI in Private Equity Portfolio Companies: What Changes After the Deal

AI in private equity portfolio companies is the work that starts once a deal closes. The owner writes AI into each company's value creation plan, supplies people to build it, and judges it the way it judges any other operating improvement: by what it does to revenue and margin.

Vista Equity Partners' Agentic AI Factory, an in-house team of AI engineers that builds products alongside the companies Vista owns, is the clearest example. Apollo gets there through APPS, a team of specialists and operators that works with each company's management on its plan.

For an associate or an operations hire, the change is concrete: the plan now carries an AI line, and someone has to turn it into a number and prove the number is real.

June 2025

Vista launches its Agentic AI Factory

built for its enterprise software portfolio

85+

Vista portfolio companies

as of June 30, 2026, per Vista

35

Full-time professionals in Apollo APPS

as stated in July 2025

What happens to a portfolio company after the deal closes?

Closing the purchase ends one job and starts the one that decides the return. The fund now owns the company, and the question moves from what to pay to how to make the business worth more by the time it is sold. Cathy Cai, who heads KKR Capstone in Asia-Pacific, puts the stakes plainly: "In today's environment, operational improvements are a necessity to achieve our target investment returns."

The value creation plan

That improvement runs through a written plan. KKR Capstone, KKR's in-house operating team, describes its job as working with "investment teams, board members, and portfolio company management teams" to "align on concrete plans to create sustainable value during KKR's investment period."

Its menu of levers runs from sales-force effectiveness and procurement to back-office design. Under digital value creation, "AI strategy and execution" sits beside data strategy and cybersecurity. A plan picks the few levers that matter for one company, names who owns each, and sets the numbers that will show progress. Management runs it; the owner's people track it at board meetings and in the monthly results.

The deal team and the operating team

Two groups do that tracking:

• The deal team. The associates who worked on the purchase stay with the company: reviewing its results against the forecast, updating the model when it reports and analysing bolt-on acquisitions. EQT tells associate applicants they will be "supporting value creation across our portfolio alongside experienced deal teams."
• The operating team. Specialists who work inside the companies. KKR Capstone has grown from a small team in 2000 to about 100 full-time operating professionals, and says more than 70% of its work is partnering with management teams to put changes in place.

The associate's side of this is measurement: is the company doing what the plan said it would? That is the part AI changes most.

What does AI change in portfolio value creation?

AI changes the plan in three ways: what goes into it, who builds it, and how it is judged.

AI gets its own line, with a number

Bain & Company's 2025 private equity report described Vista, as part of its annual operational planning, "requiring each of its portfolio companies to submit goals and quantified benefits from generative AI initiatives." Vista's specialists then worked with management teams, "defining and monitoring relevant metrics along the way." AI stops being a side project and becomes a target a chief executive signs up to, reviewed like any other line in the plan.

The owner brings the builders

Rather than leave each company to hire an AI team of its own, several of the largest owners now keep engineers who work inside the companies they own, or bring in a cloud provider's. A portfolio company can ship an AI product before it has built the team to make one.

The test is the income statement

Rodney Zemmel, global head of Blackstone's operating team, put the test bluntly in May 2026: "Too many companies run performative pilots that never show up in EBITDA." Vista frames it as a pair of measures, tracking AI across "two dimensions of value creation: revenue growth and margin expansion." An AI project either raises what the company sells or lowers what it costs to run, and the plan has to say which, and by how much.

What it means for the associate

For a junior, the monthly check of results against the plan gains new lines. The harder new job is sizing an AI initiative before it goes into the plan at all. When a chief executive says an AI support tool will save $3 million, someone has to ask which costs fall, in which quarter, and what the trial actually showed.

The associate's month before

• Check results against the forecast
• Update the model when the company reports
• Chase the cost and sales projects in the plan

With an AI line in the plan

• Check AI revenue and cost ratios against their targets
• Ask whether a trial saving has reached the accounts
• Size the next AI initiative before it enters the plan

What is Vista's Agentic AI Factory?

Vista, the firm behind those planning targets, also built a team to help deliver them. The enterprise software investor describes its Agentic AI Factory as an in-house "team of AI engineers and specialists that rapidly designs, builds, and monetizes Agentic AI products alongside our portfolio companies."

It launched in June 2025 as "a platform purpose-built to scale Agentic AI across our enterprise software portfolio," drawing on Vista's value creation team and on partnerships with Microsoft, Google and AWS. By Vista's count, the portfolio it serves ran to 85+ companies on June 30, 2026, with more than 250 million users at the end of 2025.

What the factory does

Vista's launch article set out three parts:

1. Operational restructuring. Portfolio companies were reworking their workflows with AI tools, aiming for the efficient frontier of software development and customer operations.
2. Agent deployment at scale. Vista anticipated five to ten AI agents per user, potentially 4 billion to 8 billion autonomous agents across its portfolio.
3. Go-to-market enablement. Through the hyperscaler partnerships, companies could plug agents into systems of record and sell them through the cloud providers' app marketplaces.

The launch case study was Gainsight, a customer success software company in the portfolio. With Microsoft's Azure Foundry, it built agents that handle customer renewals on their own, "including confirming contract terms, managing license counts and flagging exceptions," and that answer product questions and spot upsell openings. Vista called the result "a shift from selling software to selling outcomes."

A second example came with Google. In April 2026 Google Cloud agreed to give Vista's companies access to its Gemini models and Gemini Enterprise, and to "allocate forward-deployed engineers (FDEs) to work side-by-side with Vista's portfolio companies and Vista's Value Creation Team." Duck Creek Technologies, an insurance software company Vista owns, used the factory and Gemini to build an agent that automates core parts of taking in a new insurance claim.

How Vista measures it

Vista's 2026 mid-year report on AI across its portfolio puts figures on both of those measures. On revenue, the clearest case is Nexthink, whose AI annual recurring revenue rose from $20 million in the first quarter of 2025 to $109 million a year later. Sonatype moved to pricing per scan: its licences were down 1% while scans were up 40%.

On margin, Vista reports averages from the portfolio companies that sent in data. Across 54 of them, research and development fell from 21.9% of revenue in 2023 to 19.2% in 2025. Across 17, annual recurring revenue per customer success manager rose from $5.4 million to $6.7 million. Sales and marketing fell furthest as a share of revenue:

2023

30.4%

2024

27.6%

2025

25.6%

Average across 55 Vista portfolio companies that reported to Vista, which encouraged but did not require a response. Vista analysis as of December 31, 2025.

Bain's 2025 report put a number on where Vista expected this to lead. Software investors have long used the Rule of 40, the convention that revenue growth plus profit margin should add up to at least 40. Vista expected AI to rewrite it: within three to five years, Bain wrote, Vista believed the new standard "will reach 50% or even 60%."

How lessons spread across the portfolio

Bain's report also described the habits the factory builds on: a GenAI CEO Council where the chief executives of Vista's companies were expected to share what they were learning, "organized so small companies can learn from large ones and vice versa," and annual hackathons in the US and India.

Vista's eighth hackathons, in September 2025, were built around agentic AI and drew more than 300 competitors on 68 teams, representing nearly half of its portfolio companies. The winner for business impact, Vena, built a planning agent that, in Vista's words, "cuts budgeting cycles by over 60%."

Worked example: what an AI plan does to one company's numbers

Take the portfolio averages above and apply them to a single company. The company is invented; the method is one an associate can use to size an AI line before it goes into the plan.

A software business has $200 million of revenue, grows 15% a year and earns a 25% EBITDA margin, so it scores exactly 40 on the Rule of 40. Suppose its cost ratios start where the 2023 averages did. Its AI plan has three parts: coding tools in engineering, AI tools in sales and support, and a paid AI add-on for customers.

On paper the plan lands inside the 50 to 60 range described above. The associate's work is deciding how much of the 12.5-point gain is real. Four checks do most of it:

• Is the cost fall really from AI? A hiring freeze, or a sales team cut for other reasons, lowers the same ratio. Credit AI only with savings tied to a specific tool or workflow.
• Did the capacity become savings? Engineers who write code faster can ship more features with the same headcount. That is valuable, but it shows up in growth later rather than in this year's margin.
• Is the AI revenue booked? A trial with free users is not $10 million of revenue. Count signed, recurring contracts, and note what customers are actually paying for.
• Is the timing right? Savings arrive in stages, as contracts lapse and roles go unfilled. Put each one in the quarter it can actually appear.

Suppose the checks leave half of each gain. The margin rises 3.75 points to 28.75%, and the add-on brings $5 million, lifting growth to 17.5%. The score is 46.25 and EBITDA is $57.5 million: still a better company, but a different plan, and a different story at exit.

Before the AI plan

40

Plan as written

52.5

After the four checks

46.25

Illustrative. An invented company, using portfolio averages as inputs and a 50% haircut after the checks.

How does Apollo use AI in its portfolio companies?

A central factory is one route. Apollo takes another: an operating team that works through each company's own management.

APPS and the Value Creation Offices

Apollo describes Apollo Portfolio Performance Solutions, known as APPS, as supporting its private equity portfolio "throughout every stage of ownership." In July 2025 it had 35 full-time professionals, a mix of functional specialists and generalist operators, working in domains that include digital transformation, AI integration, and procurement and supply chain. Its method is the Value Creation Office: "Through Value Creation Offices (VCOs), the team works with management to ensure rigorous execution and accountability."

That month Brian Chu joined as head of APPS, and Aaron Miller, who had led it since 2019, became chairman. Antoine Munfakh and Michele Raba, Apollo's heads of private equity for North America and Europe, gave the reason the team matters: "outperformance will be driven by improving businesses rather than expanding multiples."

The AI work sits inside this platform. Vikram Mahidhar, who had headed the AI practice at Genpact, joined in 2021 as an operating partner to lead data and digital transformation within APPS, and by 2025 headed Apollo's Data, Digital, and AI team.

In Bain's 2025 account, Apollo had also set up a centre of excellence for AI, staffed by two partners and an advisory board of outside AI experts, with a playbook covering an opportunity diagnostic, how to set up pilots and how to plan the rollout.

From value pools to a plan

MIT Sloan Management Review described the method in June 2025. As one of his first tasks at Apollo, Mahidhar identified a set of "value pools" where digital tools could lift performance: product engineering, cloud, sales and marketing, finance, AI, operations and customer care. AI "quickly became the most critical of these, since it serves as an enabler for most of the other value pools."

The work starts before Apollo owns the company, with an assessment of how AI is changing the target's whole industry; AI in private equity due diligence covers that half. What carries over is a list of specific uses and "a plan for implementing them once the business has been acquired."

After the deal, the goal is to make each company "as self-sufficient as possible," though Mahidhar's group can "supply recommended AI tools, vendors, potential senior hires, and consultants." Katia Walsh, previously chief digital and AI officer at several companies, was brought in to oversee how those AI plans are carried out in the portfolio.

What it has delivered

The same article listed results from Apollo-owned companies, as reported in mid-2025:

Every row is a cost, a rate or a return, which is the form an AI line takes once it is in the plan.

Which other private equity firms build AI into their portfolio companies?

The other large firms that describe their work in public differ mainly in who does the building: a team of their own, an incubator inside the companies, or a cloud provider's engineers.

Blackstone: three buckets and six areas

In that May 2026 interview, Zemmel sorted Blackstone's more than 270 portfolio companies into three groups: those where AI is central, those where it offers meaningful upside, and those where it is relatively immaterial, a group he said is shrinking fast.

His team's playbook covers six areas: software engineering, customer experience, combining data for new insights, content creation, supply chains, and automating corporate work with agents. The principles fit in one line: "CEO-level ownership, clear ROI, data as a competitive advantage, and designing for scale from day one."

The groundwork is older. In 2024 Blackstone's data science team said its 50-plus data scientists had worked directly with leadership at more than 70 of its companies, some of them on secondment inside the companies. At its CEO conference that March, almost 60% of more than 200 chief executives named access to talent and knowing where AI pays most as their main barriers.

Ode, set up with Anthropic and Hellman & Friedman, launched under its own name in July 2026 as a standalone AI services company. Zemmel's case for it echoes that 2024 finding: mid-sized companies "often lack the in-house resources to build and run frontier deployments themselves."

Hg: an incubator inside the portfolio

Hg, which invests in European and transatlantic software, services and data businesses, launched Catalyst in November 2025 as an in-house AI product incubator. At launch it had more than 80 AI engineers, product managers and designers in London and New York, working directly inside portfolio companies.

Hg counted more than 15 AI products shipped by then, more than ten businesses drawing over 10% of bookings from AI-enabled products, and more than $130 million of what it called EBITDA value-based uplift from AI initiatives across the portfolio. One product, GTreasury's GSmart AI, led to what Hg called its first AI-driven exit.

At the launch, Lloyd Hilton, head of Hg Catalyst, drew the distinction an AI plan needs: "There's a huge difference between AI that demos well and AI products that deliver deep customer value."

Who gets hired to do this work?

For a candidate, the programmes above sit on three kinds of seat:

• The deal team associate. The seat described above, now with AI lines to track and size. It stays a finance job: the tools change what gets measured, not who measures it, and it is the usual graduate route into this work.
• The operating specialist. Hg says its value creation team holds more than 100 AI experts, 20 of them in-house specialists, and Apollo's Mahidhar and Walsh came from running AI at Genpact and at several companies. Seats like these tend to go to people with years of operating, consulting or technology leadership behind them.
• The builder. When it launched Catalyst, Hg said it was hiring AI engineers, software engineers and technical product managers to work across its 55+ portfolio companies. Ode says many of its staff are former technical founders, and most hold advanced degrees with a decade or more of engineering and AI experience.

Cloud providers send in engineers of their own as well: under the partnerships above, Google agreed to provide forward deployed engineers to portfolio companies, a seat mapped in forward deployed engineer in finance. Whether AI shrinks the junior finance seat itself is a separate question, taken up for bank analysts in Will AI Replace Investment Banking Analysts.

What might a private equity interview ask about AI value creation?

EQT describes its associate process as "around five stages, combining competency-based interviews with case or analytical exercises," and a value creation case is where AI fits most naturally. None of the firms above says it asks candidates about its own AI programme. The questions below follow from what they say they measure.

How to prepare

The preparation is practical, and most of it uses public documents:

1. Build the ratios for a real company. Take a listed software company's last three annual reports and work out growth, sales and marketing spend and research and development spend as shares of revenue, and its Rule of 40 score for each year.
2. Write its AI plan, then discount it. Size two AI initiatives for that company on one page and run the four checks from the worked example against them. Then take one AI figure from the company's own investor materials and write down what would have to be true for it to reach the accounts.
3. Read two firms in their own words. Blackstone's interview with its technology chief and operating head, and Hg's Catalyst announcement, show how operators describe AI value creation and which numbers they choose to publish.
4. For engineers, ship something measurable. Build a small agent on a real workflow, such as sorting support tickets from a public dataset, and report its resolution rate and cost per ticket, including where it failed.

Rehearse the case questions out loud with the interview readiness tool, and use the guides to learn the AI tools a junior meets on the desk.

The bottom line

AI in private equity portfolio companies is no longer an experiment run beside the plan. It is a line inside it: set as a target with a number, built by the owner's own engineers or its partners', and judged, as the firms that publish results put it, in revenue and margin.

The associate's job has moved with it, from tracking the plan to sizing its AI line and checking that the number is real. A candidate who can do that arithmetic, and can describe how the firms leading this work talk about it, is ready for the ownership half of the job.