The System of Record Problem

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

In This Week’s Issue:

From The Trenches

The System of Record Problem No One's Talking About

Most AI-for-private-markets pitches go something like this: "Connect our tool to your Box (or Salesforce, Affinity, SharePoint, etc.) and you can query all your deal documents." Sounds great. The problem is that Box isn't a system of record. It's a version graveyard.

I heard this surface on a call Friday. An associate at a PE fund had evaluated several AI vendors. One connected to Box. Her team liked the idea. But they kept coming back to the same question:

This is the hidden flaw in the "just connect AI to your file storage" pitch. If your AI layer is querying a graveyard, it's guessing. Confidently.

Jamin Ball wrote about this in his newsletter this week. Ask four departments for "ARR" and you get four different answers. Now tell an AI agent to "calculate ARR for the board deck." Which one does it use? The more you automate, the more important it becomes that someone has done the unglamorous work of deciding what the correct answer is and where it lives.

This matters more in finance than almost anywhere else. Deal work lives and dies on version control. Which model has the right assumptions? Which memo reflects the latest call with management? Which CIM is the final version vs. the one with the typo on page 47? These aren't edge cases. They're the entire job. And if your AI can't distinguish between "Revenue Model v12 FINAL" and "Revenue Model v12 FINAL (2) updated," it's not saving you time. It's creating liability.

The winning pattern won't be "AI on top of Box." Yes, that might mean restructuring how you work. It might mean migrating to a new system. But AI is not a bolt-on to existing systems. It's an opportunity to rearchitect your data and your workflows. Just patch-working something on top of messy file storage isn't going to give you the answers you're looking for.

The firms that get this right will move to deal-native systems of record that ingest materials, normalize them into structured deal objects, and then layer intelligence on top. The agent shouldn't be guessing which file to pick. It should already know.

News Digest

BNY Mellon's Agent Architecture Is Actually Driving Results

BNY announced this week it's integrating Google's Gemini 3 into Eliza, its enterprise AI platform. The bank now has over 100 "digital employees" working alongside staff. But the story isn't the model. It's the process.

Every agent at BNY passes an internal model-risk review before going live. Once deployed, the team monitors performance daily and incorporates results into a continuous feedback loop. As their chief data and AI officer told Business Insider: "Eliza, through the agent tech workflow, is able to make the process much more simpler, efficient, and orchestrated."

The use cases are bounded and specific. Payment validations. Code repairs. Market research synthesis. Data automation. Each agent is governed by tight access controls that determine what information it can use. Google's head of financial services emphasized the approach: "You need to make sure that whatever these agents are doing is grounded in the business context. That requires these models to understand, and then adhere to, certain policies and rules."

The bank has 117 AI solutions in production, up 75% from the prior quarter. Nearly the entire firm has completed generative AI and responsible AI training.

Why it matters: This is the opposite of "roll out Copilot and hope for the best." BNY is treating agents like employees: onboarded, governed, monitored, and constrained to specific roles.

My take: The firms getting value from AI aren't deploying it broadly. They're deploying it narrowly, on tasks they understand well, with structure around it. Model-risk review. Daily monitoring. Access controls. Feedback loops. That's not exciting. But it's what actually works.

Read more

Tide Rock's Contrarian AI Playbook: Growth, Not Cost-Cutting

At buyout firm Tide Rock, there's a mandate: don't use AI resources to cut costs. CEO Ryan Peddycord told Business Insider this week that the firm's AI engineers are aimed entirely at growing businesses, not cutting.

This goes against the prevailing narrative. Vista Equity is cutting a third of its workforce. Everyone's talking about efficiency. Tide Rock is betting the opposite direction. Their first AI investment was finding companies to buy. The data on PitchBook and Crunchbase is "very, very incomplete" at the sub-$10 million EBITDA level. So they built tools to find non-public information about potential targets. That same capability now helps portfolio companies find new customers.

Here's what caught my attention: Tide Rock is happy to use third-party tools for cost-cutting, but won't build them internally.

"I have a belief that everybody's so focused on cost-cutting that third parties are going to pick off all the low-hanging fruit there," Peddycord said. "So us trying to invest our dollars to go create things that other people are creating and probably investing more dollars to do isn't the right place to spend our money."

Why it matters: Most firms are doing the opposite. They're building internally for efficiency and buying externally for growth. Tide Rock's thesis: let specialists solve cost-cutting.

My take: Think about what your alpha is as a firm and develop specific, unique processes for that. If it's sourcing, go all-in on that internally. If it's value creation through digital marketing, spend your time there. But there are huge efficiency gains to be made elsewhere. Buy existing solutions for those. It's about how you focus your time internally. Double down on your niche. Let specialists handle the rest.

Read more

Enterprise AI Spend Tripled to $37B: Menlo Ventures Report

Menlo Ventures released its annual State of Generative AI report this week. Enterprise AI investment hit $37 billion in 2025, triple the $11.5 billion from 2024. More than half went to applications rather than infrastructure.

The market share shift is notable. Anthropic now holds 40% of enterprise LLM spend, up from 12% in 2023. OpenAI fell to 27%, down from 50% two years ago. As Menlo partner Deedy Das put it: "The era of automatic OpenAI wins is over."

But the most interesting finding is about build vs. buy. In 2024, 47% of AI solutions were built internally, 53% purchased. Today, 76% are purchased rather than built. Enterprises tried internal AI development. It was hard. Now they're buying.

Why it matters: The shift from building to buying is accelerating. Green light for vertical AI companies with purpose-built solutions.

My take: The build vs. buy shift is the story. Last year, nearly half of AI solutions were built internally. This year, barely a quarter. Massive tailwind for AI-native vendors with domain expertise.

Read more

Other M&A News This Week:

IBM acquires Confluent for $11B (Dec 8) - IBM's largest acquisition since Red Hat. Confluent provides real-time data streaming infrastructure. 40% of Fortune 500 are customers. CEO Arvind Krishna: "Nobody can live with month-old data." The AI plumbing story continues.

Intel signs term sheet to acquire SambaNova (Dec 10) - Non-binding agreement for AI chip startup once valued at $5B. Intel CEO Lip-Bu Tan is also SambaNova's chairman. Intel Capital is already an investor. Governance questions abound. Closing could take months.

Other Cool Stuff I’ve Read of Seen This Week:

Disney invests $1B in OpenAI, licenses 200+ characters to Sora (Dec 11) - Mickey, Darth Vader, Iron Man in AI video generation. Same week Disney sent Google a cease-and-desist for copyright infringement. If you can't beat 'em, license to 'em.

Accenture and Anthropic form dedicated business group (Dec 9) - 30,000 Accenture professionals trained on Claude. Claude Code deployed to tens of thousands of developers. Anthropic now has 40% enterprise market share. The enterprise AI race has a leader.

Trump signs AI executive order preempting state regulations (Dec 11) - Single federal AI framework, DOJ task force to sue states with "onerous" AI laws. California's SB-1047 in the crosshairs.

Time names "Architects of AI" Person of the Year (Dec 11) - Huang, Altman, Zuckerberg, Amodei, Su, Musk, Hassabis, Fei-Fei Li. Five billionaires with $870B combined fortune. The robots have officially arrived.

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