Boring industries are the alpha

Palantir did $2.87 billion in revenue in 2024, up 29 percent, and roughly 55 percent of that came from the government segment, bureaucracies buying software to help them spend a defense budget. The company most famous for being a spooky defense contractor is now, by revenue, one that sells spreadsheets and dashboards to procurement offices. That is the whole business. Not lasers. Ontology and forms.

I keep coming back to that number because it embarrasses an instinct I used to have, which is that the interesting startups are the ones that sound interesting at a party. Consumer social. Frontier models. The pitch that makes a stranger lean in. I spent a couple of years as a forward-deployed engineer shipping AI into two of the least party-friendly industries on earth, and I have slowly, grudgingly come around to the opposite belief.

Here is the thesis, stated plainly so you can disagree with it early: the alpha for founders and early-stage angels is disproportionately in boring industries, because boring is a moat that scares off the exact people who would otherwise compete you into the ground. I am going to argue this from what I have actually seen, then spend real time on why it might be wrong, because I have been wrong about it before.

Two caveats before I earn any credibility. First, I have direct operator exposure to precisely two boring industries, commercial insurance and defense procurement, plus a side view into angel deal flow through Angel Squad, Hustle Fund's community of operator-investors. That is a narrow slice. Adjust everything downward for sample size. Second, "boring" is doing a lot of work in that sentence, and I will try to define it as I go rather than let it stay a vibe.

The numbers that made me a believer

US property and casualty insurers wrote $1.05 trillion in direct premiums in 2024, crossing a trillion dollars for the first time in the industry's history, up eight percent in a year. That is one industry, one country, one slice of insurance. The software that runs a large share of it is old enough to have a mortgage. Applied Systems, still one of the dominant agency management platforms, was founded in 1983, and the core policy administration vendors mostly trace to the late nineties. The reason a lot of independent agents still key the same client into several carrier portals by hand is not that nobody noticed. It is that the fix is unglamorous and the incumbents have no reason to hurry.

Now put the defense number next to it. The federal government obligated about $755 billion on contracts in fiscal 2024, and the Department of Defense accounted for $445.1 billion of that, roughly 59 percent of all federal contract dollars. The way a mid-size defense manufacturer wins a slice of that is by responding to requests for quote, and the way a lot of them respond is a human reading a PDF, pulling line items into a spreadsheet, and emailing it back. When I worked on bid-intake and quoting for manufacturers answering DoD solicitations, what struck me was not that the process was hard. It was that it was clerical, and it had stayed clerical for years, because who wants to spend their twenties learning the difference between an NSN and a CAGE code.

That is the shape of it. Enormous dollar volume, ancient tooling, manual seams everywhere, and a talent moat made entirely of tedium. The mispricing is not in the market cap. It is in the labor supply of founders.

The seams look the same in both places

The pattern rhymed across both industries in a way that surprised me, so let me be specific about it rather than tell war stories. The structure is the point.

The first seam is that the interface between systems is a document, usually a PDF, and a person is the integration layer. In insurance the ACORD forms are the lingua franca, standardized paper that everyone accepts and almost nobody exchanges as clean structured data. In defense procurement the solicitation arrives as a PDF on a portal like SAM.gov, and the line items get re-keyed downstream by hand. When the real API of an industry is "email me the PDF and I will re-type it," the automation surface is enormous and it has sat there in plain sight, because that work is exactly what a strong engineer assumes is beneath them.

The second seam is that the domain expert becomes a single point of failure, and everyone downstream has quietly organized around that person's calendar. The underwriter who knows which risks the carrier actually wants, the estimator who knows which line items are traps, the compliance lead who knows which checklist step is a real legal requirement versus a habit nobody has questioned since the Clinton administration. That knowledge is scarce, but a surprising amount of it is legible once you sit next to the person and write down what they do, and the step that felt immovable is frequently a convention rather than a rule.

The counterintuitive part, and the reason I think this generalizes past my two industries, is that the boringness compounds in your favor after you ship. A consumer app lives or dies on a fickle thing. A tool that cuts a regulated intake process from forty minutes to four gets embedded, and the switching cost is the retraining and the audit trail and the fear of breaking a compliance obligation, which is a stickier moat than most defensibility slides.

The angel view, kept vague on purpose

Through Angel Squad I see a steady trickle of early decks, and I have started noticing which ones I actually get excited about versus which ones I merely admire. I am going to keep this at the level of pattern rather than particulars, both because these are other people's confidential decks and because the pattern is the part that transfers.

The tell I trust is granularity about the work. A founder who has done the job describes the workflow they are killing in more detail than I could pull out of them, down to which field gets typed twice and which handoff loses information. They talk about a process. The founder I have learned to distrust talks about a market, a TAM slide lifted from an analyst report with a hand-wave where the workflow should be. That number is true and almost useless, because the hard part in these industries is never the size of the prize. It is distribution, which runs on trust, and a founder who has never carried a relationship in the space badly underprices how much of the moat is that rather than the software.

None of this is a scoring rubric. It is a bias I am admitting to, which is that I overweight scar tissue and underweight polish, and I could be wrong about that too.

Now the part where I might be wrong

I have flip-flopped on this thesis at least twice, so let me argue the other side as well as I can, because a thesis you cannot attack is just a mood.

Boring might be boring for a reason. The same tedium that keeps competitors out also makes the sales cycle brutal. You are selling to a risk-averse buyer with a long procurement process, a compliance review, and no personal upside for being the one who approved the new vendor that broke something. Plenty of correct products die in that pipeline because the customer took eighteen months to decide and the startup ran out of money at month fourteen. And the incumbents are not passive. A 1983-vintage vendor with the industry's data locked in its own format can buy you, bury you in an RFP requirement you cannot meet, or simply outlast you, because they have no burn and you do.

The moat also cuts both ways. The years I spend learning which fields matter and which relationships open doors are years I am not building anything a different market would pay for, and much of that knowledge does not transfer. If I am wrong about the specific wedge, the sunk cost is a domain education with a narrow resale value.

Then there is the obvious selection problem. I have watched exactly two industries, and they are the two where a version of this worked well enough that I am now writing an essay about it. That is close to the definition of survivorship bias. Maybe insurance and defense are unusually good and I pattern-matched a coincidence into a law. If you operate in healthcare billing or freight or municipal finance and none of this describes your world, I would rather hear that than be right on a whiteboard.

The counterargument that actually keeps me up is that AI may be quietly draining the moat I keep praising. Boring industries were protected because the domain knowledge was expensive to acquire and did not transfer, so few people bothered. If a language model absorbs the NSN-to-CAGE mapping and the ACORD field dictionary in an afternoon, that moat is partly a moat against humans that does not hold against a well-prompted model, and the boring premium shrinks precisely as the tools that exploit it improve. My weak rebuttal is that trust and distribution do not compress on the same curve as knowledge, and a regulator still wants a human name on the errors-and-omissions policy. But I hold that loosely, and if you think the window is closing faster than I do, you might be right.

The generalizing beat

The glamour discount is real, and it is paid by whoever is willing to be bored in a room where the money already is.