Sponsors build AI engineering benches while the proof of returns lags.
Blackstone and hellman & friedman put Anthropic engineers inside portfolio companies.
Blackstone and Hellman & Friedman have formed a roughly 160-person team of AI specialists with Anthropic to deploy at businesses, starting with their own portfolio companies ([1]). The team sits inside a $1.5 billion joint venture called Ode, also backed by Apollo, General Atlantic and Goldman Sachs, with Blackstone, Hellman & Friedman and Anthropic each committing around $300 million ([2]). The plan is to sell Ode's services well beyond private equity-owned businesses ([3]).
The deployment detail matters more than the headline. Blackstone intends to introduce Ode across 25 of its more than 270 portfolio companies, and engineers are already working at six of them, according to Private Equity Wire's summary of the Journal's reporting ([4]). At Chamberlain Group, the LiftMaster garage-door maker Blackstone bought in 2021, Ode specialists sit inside an 18-person engineering team building connected-home features. Chamberlain now expects its digital-access business to reach about $500m of annual revenue by 2030, against an earlier estimate of around $160m and $40m last year ([5]). Hellman & Friedman has started at Baker Tilly, working with audit executives to map the daily work of accounting and tax staff and find where AI agents fit ([6]).
Three points for value-creation teams. First, the case is framed on revenue, not headcount: executives involved said revenue generation is a central objective, rather than using AI primarily to cut headcount or costs ([7]). Revenue growth carries a multiple at exit, while one-off savings are easier for a buyer to discount. Second, portfolio companies pay Ode through commercial consulting or service agreements and participation is voluntary, so each business still has to justify the spend in its own P&L ([8]). Third, the market for embedded AI engineering is filling up: OpenAI has set up a competing AI services venture with backing led by TPG ([9]). Sponsors without a shared engineering bench now need a clear position on whether to build one, rent one or rely on portfolio CTOs, and on how they will evidence the result to a buyer.
Nationwide shows the gap between AI use and measured value.
Nationwide gave an unusually candid account of where corporate AI stands. The building society is rolling out workplace AI to around 16,000 staff, with about 1,000 people in the combined Nationwide and Virgin Money finance team on premium Microsoft 365 Copilot licences and some 2,500 engineers using GitHub Copilot ([1]). Yet its technology transformation director, Paul Ballard, said: "The thing that we've not nailed yet… is what's the value that we're receiving from it?" ([2]).
Where Nationwide can measure, the numbers are concrete. Using GPT-4 through Azure OpenAI cut the average time to produce a customer letter from around 45 minutes to between 10 and 15 minutes, with staff reviewing the output before it is sent ([3]). The lender also consolidated 19 separate data stores onto Azure and Azure Databricks and set up an AI council spanning technology, security, legal and procurement ([4]). It is now looking at AI agents for anti-money laundering, customer due diligence and mortgage operations, but Ballard said it is "nowhere near autonomous agents" ([5]).
The wider evidence is just as thin. One Atlanta Federal Reserve study found that about 90% of executives believe AI has not yet boosted productivity at their companies ([6]).
For portfolio CFOs, this is the gap a buyer will probe. Time saved per task is not EBITDA until it changes staffing plans, throughput or price. Value creation plans that count licences issued or hours saved will be marked down in diligence. The Nationwide pattern is the more defensible one: choose processes with a countable unit (a letter, a ticket, an invoice), baseline the cost per unit before rollout and report the change monthly through the hold. Its multi-model stance, looking at "different models for different use cases" to keep "optionality" as costs and regulation change ([7]), also limits dependence on a single vendor at a time when model pricing moves quarter by quarter.
Headcount-led AI cases are losing the argument.
Research republished by Fortune this week challenges the default AI line in many value creation plans: fewer people. Mark Ma of the University of Pittsburgh and colleagues studied AI investment and layoff announcements by US public companies over five years and found that as AI investment announcements rise, so do announcements of job cuts attributed to AI ([1]). The stock market reaction to those layoff announcements averaged close to zero, and was negative or close to zero for more than half of the events ([2]).
The mechanism is the useful part. Using millions of Glassdoor reviews, the researchers found a strong association between employee sentiment towards AI and firm productivity, and a sharp fall in that sentiment when companies announced AI-driven layoffs ([3]). Management optimism about AI on roughly 10,000 earnings calls bore no significant relationship to productivity outcomes ([4]). Employees cited job security, thin training and poor AI leadership as their main complaints ([5]).
There is a talent cost as well, sharpest in professional services. Chris Churchman, the Goldman Sachs partner who leads its Marquee platform, warned of "a huge danger here that in the era of AI, we outsource our reasoning to these models" ([6]). He said junior traders learn by fielding client pricing requests under supervision, and questioned whether automating that work still produces senior traders who fully understand it ([7]). He added that the hardest technical problem is making AI answers fully factual and auditable ([8]).
For sponsors holding accounting, legal, consulting or financial services platforms, both points bear on exit value. A buyer of a people business is buying its pipeline of future partners and senior staff. Cutting junior hiring to show margin in years two and three can leave a thinner bench at exit, and a diligence team will find it. The better-evidenced route is to redeploy capacity into revenue, such as more clients per head or faster turnaround, and to track AI sentiment and attrition alongside margin. That is also how Ode's sponsors are framing their own programme.
Automated decisions and cyber now carry nine-figure price tags.
The Dutch Data Protection Authority has decided to fine Uber €825 million ($966 million) for deactivating driver accounts through automated systems without adequately informing drivers, according to a 17 August decision reviewed by Reuters ([1]). It would be the second-largest fine ever issued under GDPR, behind the €1.2 billion imposed on Meta in 2023 ([2]). GDPR bans decisions made solely by algorithm when they have a significant impact on people's lives, and requires meaningful human review and a way to challenge the decision ([3]). Uber disputes the ruling and says only 126 drivers in Europe were deactivated for low customer ratings in 2021 ([4]).
The size of the fine relative to the number of people affected is the signal. Portfolio companies are putting AI into exactly these decisions: credit and collections, fraud flags, hiring screens, claims triage and workforce scheduling. Each needs a documented human review step and a clear notice to the people affected. Diligence teams should ask for an inventory of automated decisions with significant effects on customers, staff or contractors, not just a list of AI tools. Where a target cannot produce one, that is a pricing point.
Cyber exposure reached a sponsor directly. Apollo disclosed that hackers had unauthorised access to certain cloud platforms between 6 and 10 July and that the information potentially affected included names, dates of birth, home addresses and social security numbers ([5]). Reuters had reported that hackers built websites aimed at stealing passwords from employees of private equity firms and financial companies, and experts said low-tech phone-call tactics still rank among the most effective, despite AI-driven threats ([6]).
A sponsor that rolls out AI tools and shared platforms across its portfolio also widens the shared attack surface. Identity controls, help-desk verification and access reviews belong inside the AI programme budget, not in a separate line that gets cut first.
Stalled exits push AI payback into the hold period.
European direct lending hit a record €63.2bn in the first half of the year, up from around €40bn a year earlier, according to Debtwire, driven partly by sponsors refinancing portfolio company debt as exits slowed ([1]). "The lack of exit prospects, especially for private equity-backed companies, […] is a big reason lenders and sponsors look to push maturities out with a [refinancing]," said Debtwire's Patrick Costello, who expects many more refinancings into next year unless M&A picks up ([2]). Second-quarter lending fell 25% year on year to €28.4bn, with the UK and Ireland the busiest market at 186 deals ([3]).
Where sponsors are buying, public-to-private remains a favoured route. Australia's Steadfast agreed an A$7.7 billion buyout by a KKR-backed consortium, with Amwins taking the underwriting agency business and Dragoneer the broking operations, at a premium of nearly 52% to its undisturbed price ([4]). Emanuel Ajay Datt of Datt Capital called the deal "symptomatic of the persistent valuation arbitrage between Australian public and global private markets" ([5]).
Longer holds change the AI calculation. When an asset is refinanced instead of sold, the AI programme has to pay back inside the extended hold and support the lender's view of cash generation, as well as the eventual equity story. That favours use cases with near-term cash effects, such as working capital, pricing and cost to serve, over long-dated platform bets. It also argues for the revenue framing Blackstone is using at Chamberlain, where a buyer can see the line growing in the accounts.
Labour-heavy distribution businesses such as insurance broking, where sponsors keep paying premiums, are also where buyers will ask hardest whether AI is a margin lever or a threat to commission income. Value creation plans for these assets should model both cases and show which one the evidence supports.