How AI Cut Loan Approvals by 94%—and the Logic Behind a $770M Acquisition

Source: No Priors | Published: 2026-09-24T10:00:05Z

Six months after Sequence Holdings invested in BankSouth, Q2 loan volume doubled while the underwriting team actually shrank.

These are the numbers Michael Lee, co-founder and CEO of Sequence Holdings, shared on the No Priors podcast. Shortly before that episode dropped, Sequence had just announced the largest AI-related take-private deal to date: a $7.7 billion acquisition of insurance broker Baldwin, in partnership with the Dell family office.

Startups Will Win — But Not in Every Industry

Michael Lee has been studying AI since 2017 — the era when AlphaGo first appeared and the original Transformer paper had just been published. He led AI coverage at Lone Pine and lived through the full arc from convolutional neural networks and LSTMs to large language models.

After ChatGPT, he started thinking seriously about one thing: AI's impact on the economy would be uneven.

Some industries AI will barely touch — restaurants, golf courses. Others, startups will inevitably win — coding tools, for instance. His exact words: "Having Sequence go buy an outsourced programming services company sounds like a terrible idea. That money should go to Cognition or Anthropic."

But there's a third category of industries where incumbents have always held every advantage: brand, scale, network effects, regulatory moats. In these industries, when the AI wave arrives, the winner isn't the disruptor — it's the incumbent that deploys AI best. Sequence Holdings was built to capture exactly these opportunities.

Software Companies Always Sell Yesterday's Workflow

Lee distills the paths large enterprises take toward AI transformation into three options, then explains why each falls short.

The first: build it in-house. The obstacle is obvious — you can't recruit top engineers. Every company has an "archetypal celebrated persona." Blackstone celebrates investors, so it attracts the world's best investors. In a world where alpha comes from engineering and AI, you need a culture that celebrates engineers. Traditional enterprises can't pull that off.

The second: hire consultants — Accenture, McKinsey, Palantir. The problem is incentive misalignment. Consultants want to get into your wallet, stay in your wallet, and expand their share of it. That structurally pushes toward incremental improvement, not fundamental reorganization. More critically, consultants can't touch what actually needs to change — who's doing the work, how it's organized, what the incentives are.

The third: buy software. This one has a subtle, hidden problem. Every software company designs its product by identifying a broadly applicable workflow, then optimizing for fast deployment and deep stickiness. That means software always sells today's workflow — not the workflow you should have after a genuine reorganization. Because that workflow doesn't exist yet. It can't be packaged and sold to you.

This is the logic behind Sequence's existence: use ownership to align incentives, use an internal engineering team to execute, and use a long enough time horizon to complete real transformation.

"We Have Two Problems: an Engineering Problem and a People Engineering Problem"

Lee says the people problem is far harder than the technology problem.

Every time Sequence enters a new company, the biggest source of anxiety is the employees: what does this mean for me?

Sequence's approach: position AI tools as "letting you do what you're best at," not "replacing you." In the BankSouth case, the lending team shrank — but not through layoffs. One person retired; another was reassigned to frontline sales. Loan officers stopped writing credit memos and started seeing more clients. Credit analysts stopped handling obvious cases — someone with cash in the bank applying for a boat loan — and focused on genuinely complex deals with real exceptions.

This narrative works on employees, but only when management believes it first and actively champions it.

Why Baldwin Is Worth $7.7 Billion

The insurance brokerage industry has a structural quirk outsiders often miss: your client isn't the one paying you. The insurer is. That means competition in this industry isn't about price — it's about relationships. Client retention rates have historically hovered around 90%. Startups can't break in through subsidies or product differentiation. This is a textbook "incumbent holds every advantage" industry.

Over $2 trillion in premiums flow through brokers in the U.S. annually. Brokers take their cut — and the industry has historically spawned dozens of large companies.

Baldwin hit every item on Sequence's checklist: large enough scale, highly centralized operations, solid technical foundation. CEO Trevor Baldwin had been pushing Anthropic across the entire organization early on, and the whole company runs on a single AMS system. Lee says the management trait they care about isn't "talking about AI" — it's "already doing it": moving to the cloud, consolidating data infrastructure, actually deploying model-based tools. Baldwin had done all of it.

Traditional PE Can't Pull This Off

The most interesting segment of the episode is Lee explaining why large traditional buyout funds can't replicate Sequence's model.

He breaks it down along three dimensions.

Engineer recruitment. The culture at top PE firms celebrates investors. A 25-year-old engineer there has little shot at being in the room where investment decisions get made. Sequence's engineers want exactly that.

Time horizon. PE fund structures inherently demand liquidity within a few years — "how do we package this thing for sale in three years" is a question that has to be answered. That's fundamentally incompatible with the timeframe AI transformation actually requires.

Selection criteria. Sequence does one deal a year. There's no pressure to deploy capital, no need to explain to LPs why they sat on their hands. "If we don't do a deal this year, great," he said. That scarcity is intentional, not a resource constraint.

The Logic of Permanent Ownership, No Exit

Sequence is structured as a permanent holding company, not a fund. That choice connects directly to their core thesis: the time horizon of the capital has to match the time horizon of the transformation.

If you're selling a company in three years, your engineers will have lower tolerances for system complexity — they'll worry whether anyone can maintain what they built after handoff. Remove the exit pressure, and that constraint disappears. You can go deeper and more complex.

Portfolio company employees are also more willing to cooperate when they know these people will be around for a long time, not gone once the project wraps. The sense of ownership is completely different.

The Atlas platform — Sequence's internal technology foundation, comprising a data ontology layer, an agent-building layer, a workflow orchestration layer (Lattice), and an application layer (Artifacts) — only makes sense because of this long-term structure. Infrastructure built at BankSouth can be reused directly at Baldwin, then at the next company. That kind of compounding is nearly impossible under a fund model.

Back to First Principles: Good People, Big Markets

Lee's career arc — Goldman, Apollo, Lone Pine — traces a path from debt structures and capital markets to fundamental research on the world's best businesses. He met Jensen Huang at Lone Pine in 2017. He notes that Nvidia then bears no resemblance to Nvidia now, but Jensen's clarity of thought and ability to execute have always been there.

Toward the end of the interview, the host asked him about early-stage venture investing. He's not a VC by background, but he gave a very VC answer: find smart people, find big markets, bet on them. The idea itself isn't worth much. Execution is brutally hard. Truly exceptional people always find a way to make things work — whether by finding someone who fills the gaps, or by grinding through until everyone who thought they were crazy comes around.

He said the same holds true in public markets.