The Model Doesn't Matter

I round up the most relevant AI-in-finance news - the deals being done, who’s rolling out what, and what’s actually working on the front lines

…a million-token context window. Native computer use. Financial plugins for Excel and Sheets. LinkedIn and X do what they do best. "Best model ever." "This changes everything." Stop me if you’ve heard this one before.

Meanwhile, Reuters reported that Thoma Bravo and Hellman & Friedman have been eyeing FactSet after AI fears drove a 39% drop in its shares. The paradox is striking: the same disruption fears creating buying opportunities are also making firms too nervous to pull the trigger.

But first, my take on why the model wars are a distraction.

In This Week’s Issue:

From The Trenches

The Model Doesn't Matter

GPT 5.4 dropped on March 5. Within hours, the familiar chorus began. "This is the one." "Finally, a model that can actually do finance." "We're switching from Claude." I've heard versions of this on calls for the past year. Every few months, a new release resets the conversation.

And I get it. For the past few months, everyone I've spoken to has been telling me they're moving to Claude. The document output is excellent. The Excel and PowerPoint generation is ahead of anything else. I use it daily and prefer it for most of what I do. But now GPT 5.4 lands with many of the same capabilities, and suddenly the conversation flips again. That's kind of the point.

Agonising over Claude vs. ChatGPT is like debating which protein powder to buy. They're roughly the same ingredients. They'll both get the job done. The value isn't in the powder. It's in the workout. And right now, most firms are spending all their time comparing labels and none of their time in the gym.

What I keep hearing on calls is firms treating a model subscription as if it's going to unlock AI for them. It won't. The models are converging. They leapfrog each other every quarter and settle back into rough parity within weeks. Have access to both. See what you prefer. But that's a preference, not a strategy.

The strategy is what sits underneath.

I think about AI maturity in levels. Level 1 is basic visibility: you can see your data and draw insights from it. Drop a CIM into ChatGPT, get a summary. Level 5 is autonomous decision-making across a connected data network, where agents are plugged directly into your operational infrastructure and acting on your behalf.

Most firms I speak to are at Level 1. They're getting summaries and thinking they're at the frontier. They're not. They're at the starting line.

The jump to Level 2, where you start making operational decisions from your data, has nothing to do with which model you're using. It has everything to do with whether your data is structured, connected, and accessible in a way that any model can work with.

Anyone who's been reading this newsletter since Issue 1 will recognise the pattern. I've been circling around this theme for months: systems of record, data architecture, the brittle stack problem. It all comes back to the same thing.

I've had a version of the same conversation on at least five calls in the past two weeks. A firm has thousands of deals in their CRM, hundreds of thousands of pages in Box or Dropbox, years of notes scattered across email and Word documents. They open up Claude or ChatGPT, paste in a document, and get a decent summary.

But then what?

That summary doesn't connect to anything. It doesn't know about the last three deals you looked at in the same sector. It doesn't remember how you adjust for stock-based compensation. It doesn't know your IC's preferences. Every time you start a new chat, you're starting from zero. You're re-explaining your frameworks, re-uploading your documents, and getting slightly different outputs each time.

You're essentially hiring a brilliant analyst who forgets everything at the end of each conversation and has never met anyone else at your firm.

The real promise of AI for deal-making isn't a better chatbot. It's connected context. Everything that's ever touched your firm, structured and accessible in one place. Every set of financials linked to the company. Every customer summary, every budget, every forecast, all connected.

That's what we've been building at DealSage. Every data point connected to every other data point through what we call an ontology, a structured map of how all your deal context relates. When you ask a question, the answer draws on everything. Compare this CIM to the last three HVAC deals you evaluated. How does this company's customer concentration compare to the one you acquired last year? What did your IC flag in businesses with similar revenue profiles?

I spoke to a PE professional this week who put it perfectly. He said his firm's data isn't in a bad place. It's in no place. They have everything they need. But it's scattered across Dropbox, email, a CRM nobody trusts, and Word documents that nobody can search. No model change fixes that.

And that's the irony of the GPT 5.4 excitement. You could give every firm access to the best model on earth tomorrow. Most of them wouldn't be able to use it for anything beyond what they're already doing. Because the constraint was never the model. It was the infrastructure.

The model is the engine. But right now, most firms are debating engine specs while sitting in a car with no wheels. Its time to start thinking bigger.

News Digest

AI Anxiety Tests PE's Record Bet on Financial Services

PitchBook's weekend analysis dropped a timely question on March 8: what happens when the industry most exposed to AI disruption is also PE's biggest portfolio bet? Private equity firms have poured record capital into financial services over the past decade. Insurance, wealth management, payments, lending. Those same sectors are now directly in the firing line.

The report follows weeks of mounting evidence. Altruist's AI tax tool wiped $20 billion from wealth management stocks last month. AIG processed 370,000 submissions without adding headcount. Munich Re cut 1,000 roles to AI. And the Block layoffs, where 40% of staff were let go explicitly because of AI, showed that tech-forward financial companies are already acting on the thesis.

The details:

Why it matters: Software was the first shoe to drop. Financial services may be the second. And PE's exposure here is arguably deeper and harder to unwind.

My take: The pattern is the same one we saw with software, just slower. The firms that moved early to embed AI into operations, like AIG, are pulling ahead. The ones that treated it as a future problem are discovering it's a present one. What worries me most is the intersection. PE firms that loaded up on both software and financial services are now facing disruption risk on two fronts simultaneously. And unlike software, financial services assets can't easily be repositioned as "AI-native." The firms doing the acquiring need to be asking a very specific question during diligence: not just "is this business profitable?" but "can this business survive the next wave of automation?"

Read more

AI Fears Create a Paradox for PE Firms Eyeing Data Companies

Reuters reported this week on a fascinating tension. FactSet's shares have dropped 39% in six months on AI disruption fears. Morningstar is down 27.6%. Gartner is off 29.5%. Those markdowns have caught the eye of Thoma Bravo and Hellman & Friedman, both of which have been running the numbers on a potential FactSet acquisition.

But the same AI fears that are creating the buying opportunity are also making the buyers hesitate. The selloff deepened after Anthropic released its latest Claude Cowork upgrade last month, and it's hitting indiscriminately. Microsoft included. The question PE firms can't answer: are these companies cheap, or are they cheap for a reason?

The details:

Why it matters: This is the first time we've seen AI disruption fears directly shape the deal pipeline for major financial data companies. The assets PE would normally be racing to acquire at these prices are the ones they're most afraid to touch.

My take: There's a real irony here, and it connects to this week's FTT. The firms hesitating on FactSet are worried about AI disruption. But FactSet's core asset is structured financial data. That's exactly the kind of asset that becomes more valuable in an AI-driven world, not less. The question isn't whether the data matters. It's whether FactSet's delivery model survives. And that's a question about architecture and connectivity, not about the data itself. The buyer that figures out how to acquire these data assets and rebuild the delivery layer for an AI-native world is going to look very smart in three years.

Read more

Other Interesting Things I’ve Read of Seen This Week:

AI to transform how credit market works, JPMorgan's Jhamna says (Mar 2) - Sanjay Jhamna told Bloomberg TV that generative AI will reshape credit trading by handling unstructured data at scale. When JPMorgan's credit desk starts talking about AI as infrastructure rather than experimentation, the timeline just got shorter.

BlackRock GIP and EQT make $33bn bet on electricity demand (Mar 3) - The AES acquisition is being framed as an AI play. Data centres need power. Power needs generation. PE owns the generation. Sometimes the best AI trade has nothing to do with software.

Bezos returns as CEO of $30B AI lab to buy companies and rebuild them with AI (Mar 5) - Project Prometheus is raising tens of billions to acquire industrial companies (jet engines, semiconductors, automobiles) and reconstruct them using AI. 120+ researchers from OpenAI, DeepMind, and Meta. Abu Dhabi Investment Authority and Jamie Dimon reportedly in funding talks. "It's the difference between selling pickaxes and owning the mine."

AI needs management consultants after all (Mar 8) - OpenAI and Anthropic are striking deals with McKinsey, BCG, Accenture, and Deloitte to push AI deeper into enterprises. Two-thirds of organisations haven't started scaling AI. More than half of CEOs say they've seen no financial benefit. The consultants AI was supposed to replace are now being hired to explain it. You couldn't script it better.

China says it can keep jobs stable for 5 years despite AI (Mar 7) - China's human resources minister told parliament that AI won't destabilise employment over the next five years. The confidence is notable. Whether it ages well is another question entirely.

KKR eyes multibillion-dollar sale of data centre cooling company (Mar 8) - CoolIT Systems could fetch over $3 billion. AI servers generate enormous heat that traditional cooling can't handle. KKR bought the picks-and-shovels play and is now selling at a premium. The infrastructure layer of AI continues to print money while the software layer burns.

Acquisition Intelligence is a weekly newsletter on AI in M&A for finance professionals, private equity investors, investment bankers, corp dev teams, and deal-makers.

For questions, feedback, or to share what you're seeing in the market, reply to this email.

P.S. I'm Harry, co-founder of DealSage. We're building an AI-native deal intelligence platform to help professionals turn their institutional knowledge into better decisions. If you're curious what we're up to, check out dealsage.io or just reply here