A partner at a mid-sized Sand Hill Road firm once told me something that reframed how I think about venture capital entirely. He said: ‘We don’t bet on technology. We bet on friction.’ He wasn’t being poetic. He was describing a literal investment thesis that his firm had quietly used to call four major automation waves before the press caught on. The industries they targeted, radiology software, legal document review, freight brokerage, and mortgage underwriting, all looked completely different on the surface. But underneath, they shared the same fingerprints.

This is how the smart money actually works. Not by reading trend reports or listening to keynotes, but by running a repeatable diagnostic on industries that still haven’t figured out what decade they’re operating in. And if you understand the pattern, you can see what’s coming next before the term sheets start flying. Venture capitalists often make these calls faster than you’d expect, and the speed is only possible because they’re matching what they see to a template they’ve already internalized.

The Friction Fingerprint

Every industry that gets automated shares a specific set of traits before the disruption hits. VCs I’ve spoken with, and founders who’ve successfully raised in automation-heavy spaces, describe the same checklist almost word for word.

First: the industry runs on credentialed labor doing uncredentialed work. When you see lawyers spending 60% of their hours on document review that requires no legal judgment, or radiologists manually flagging images using criteria that can be reduced to a decision tree, that’s the signal. The credential is being used to provide liability cover, not cognitive value. Automation doesn’t eliminate the credential. It just handles the part that never required it.

Second: the industry has pricing power that’s completely disconnected from the underlying cost to serve. Freight brokerages historically took 15 to 20 percent margins on transactions that were essentially information arbitrage, knowing which carriers had capacity and which shippers needed it. That’s not a service business. That’s a data bottleneck wearing a service business costume. Any time you see margins that fat in a coordination role, automation is already circling.

Third: the customer is captive and unhappy. Not switching because alternatives don’t exist yet, not because they’re satisfied. This is a crucial distinction. Captive-and-happy industries are hard to disrupt. Captive-and-miserable industries are waiting for permission to leave.

Why VCs Look at the Workforce Before the Technology

Here’s the counterintuitive part. The best automation investors don’t start by asking ‘what can AI do now?’ They start by asking ‘which workforces are doing things that feel like thinking but aren’t?’

This is a meaningful distinction. True cognitive work, the kind that requires judgment under genuine ambiguity, is still hard to automate profitably. But a massive percentage of white-collar work isn’t actually that. It’s pattern matching, data retrieval, form completion, and rule application. Work that looks like thinking because it happens inside a human brain, but that breaks down into deterministic steps once you examine it closely.

Mortgage underwriting is the canonical example. Before the automation wave hit, underwriters were considered skilled professionals. And some of them were. But the core of the job, checking debt-to-income ratios, verifying employment history, flagging property appraisal anomalies against comps, was a rules engine. The humans were executing an algorithm. The VCs who saw this early weren’t smarter than everyone else. They just looked at the actual workflow instead of the job title.

This is also why sectors with high unionization or strong guild dynamics (medicine, law, finance) get disrupted more slowly but more completely. The institutional resistance delays the entry point, which paradoxically gives early investors more time to build before incumbents respond. Successful startups often deliberately choose markets that don’t exist yet for exactly this reason. Regulatory moats work both ways.

The Sequence Is More Predictable Than It Looks

Automation doesn’t hit industries randomly. It follows a sequence based on data availability and error tolerance. Understanding the sequence is where pattern recognition becomes genuinely predictive.

It starts with industries where errors are cheap and data is abundant. Customer service routing. Basic image classification. Fraud flagging on low-value transactions. These are the training grounds where the models get good.

Then it moves to industries where errors are expensive but containable. Radiology is a perfect example. A missed diagnosis is catastrophic, but the error is discoverable, the domain is bounded, and the training data (millions of labeled scans) is unusually rich. High stakes but high structure.

The frontier right now is industries where errors are expensive and the domain is messy and contextual. Legal strategy. Complex medical diagnosis. Financial planning. These are taking longer not because AI can’t do parts of them, but because the cost of an error is diffuse and the liability structure hasn’t caught up yet.

Savvy investors aren’t just looking at where AI is good today. They’re watching which industries are quietly building the data infrastructure that will make automation viable in five years. When a sector starts digitizing records aggressively, mandating structured data inputs, or standardizing workflows in ways that look like operational improvements, that’s a pre-automation signal. The industry is, often unknowingly, building its own replacement’s training set.

What Gets Overlooked Because It Looks Too Boring

The industries that are obviously ripe for disruption have already been picked over. The interesting bets right now are in sectors that look unglamorous, which is exactly why they’re rich with opportunity.

Construction project management. Home inspection. Municipal permitting. Agricultural equipment servicing. These aren’t sectors where you’ll find a lot of startup mythology, and that’s precisely the point. Digital minimalists ignore most tech trends on purpose and routinely outperform the hype-followers, and the same logic applies to investors who resist the gravitational pull of whatever category is already oversubscribed.

The VCs calling these shots are applying the same friction fingerprint to industries that haven’t been on anyone’s pitch deck. They’re asking: where is credentialed labor doing uncredentialed work? Where are margins fat because of information asymmetry? Where are customers captive and miserable?

The pattern doesn’t care how boring the industry looks from the outside.

The Pattern Is a Tool, Not a Crystal Ball

I want to be honest about the limits here, because the VC world has a bad habit of retrofitting clean narratives onto what were actually probabilistic bets made with incomplete information. Most software patents are never used, and the strategy behind them is more calculated than it appears, and the same strategic opacity applies to investment theses. Funds rarely publish their actual playbooks because the edge disappears the moment everyone is running the same screen.

Pattern recognition in venture is a real and useful tool. But it’s a tool for narrowing probability distributions, not predicting certainties. The investors who’ve called automation waves correctly have also made bets in the same framework that went nowhere. The sector had all the friction fingerprints and still didn’t tip, usually because the regulatory environment, the liability structure, or the customer’s willingness to trust a machine didn’t move on the timeline the thesis required.

What separates the funds that get this right consistently is that they treat the pattern as a starting hypothesis, not a conclusion. They build positions early, stay close to the actual workflow changes happening inside industries, and they update. The pattern is the door. What’s behind it still requires judgment.