Palantir’s US commercial revenue grew 149% in the second quarter of 2026, to $764 million. Plenty has been written about why. For a smaller business the useful part is not the software. It is the way Palantir sells it.
Its favoured way to win a customer is to run its AI platform on that customer’s own data, against a real problem, for anything from a few hours to five days. It calls these bootcamps. A buyer of any size can insist on the same thing.
What a bootcamp is
Bootcamps were a deliberate change of sales approach, introduced in 2023 alongside Palantir’s AI platform. In October that year the company set itself a goal of running 500 within twelve months. By February 2024 its chief revenue officer, Ryan Taylor, told investors it had completed “more than 560 bootcamps across 465 organizations”.
The format is short, and the length varies. On the same call, Palantir’s Shyam Sankar described “a recent two-day bootcamp” with a construction, engineering and architecture company. Taylor described cold outreach that “led to a five-day bootcamp, then an enterprise agreement”. Chief executive Alex Karp put the pitch more bluntly: “we’re going to run your data at a bootcamp for 10 hours.”
The customer brings real data and a real problem, and its own people build against it alongside Palantir’s engineers. At the end there is a working result on the customer’s own material, or there is not.
Why it works
It answers the one question a buyer has.
Nobody doubts that AI can read a document or summarise a spreadsheet. A demo on the vendor’s tidy sample data usually works. What the buyer cannot know is whether it works on their material, with its scanned forms and a spreadsheet that has three tabs called “final”. A few days on that material settles the question in a way no sales deck can.
It also puts the people who do the work in front of the result. They see where it gets things right and where it misses, and they are the ones who will have to trust it afterwards.
What the results look like
The clearest published example comes from the US Navy. In December 2025 the Navy announced a $448 million investment in a shipbuilding operating system built on Palantir’s software. It said a pilot at General Dynamics Electric Boat had cut submarine schedule planning from 160 manual hours to under 10 minutes. Another pilot, at Portsmouth Naval Shipyard, cut material review times from weeks to under an hour.
AIG, the insurer, is building AI agents for underwriting in partnership with Palantir and Anthropic. Its chief executive, Peter Zaffino, told investors in May 2026 that Palantir’s platform had been used to build “a digital map of our business”. He was careful to add that “human oversight is and will continue to be essential to our underwriting processes.”
The Navy’s figures are its own, and nobody has checked them independently. Treat them as a direction rather than a promise. The direction is what matters here. The work being sped up is people pulling information from several places and checking it against each other. Every business has some of that.
Why small firms usually buy the other way round
A small business buying AI usually gets a demo on someone else’s data, or a long discovery phase before anything runs. Sometimes it gets a login and nothing else. None of these tells you whether the thing will cope with your paperwork, which was the only thing you needed to know.
The result is the pattern we described in why AI pilots stall before production. The project proves the technology works, which was never in doubt, and leaves the real question open.
How to run your own version
You do not need Palantir’s budget to buy this way. You need a small, real test with the result agreed up front.
Pick one task. Choose something repetitive where a right answer can be checked, such as invoices keyed into the accounts package or orders copied out of emails. “Our operations” is not a task.
Bring a real sample, awkward ones included. A few dozen to a few hundred examples, taken as they come rather than picked because they are clean. The handwritten note and the supplier who formats everything differently are the test.
Agree what good looks like before anything runs. How many should it get right? What should it do with the ones it is unsure about? We have argued that the definition of done should be a number, and a short test is where that number gets set.
Have the person who does the job in the room. They will spot problems an outsider could take weeks to find.
Ask what it got wrong, and why. A supplier who only shows you the successes has not run a test. The misses tell you what the real build will involve.
What a few days will not prove
A short test shows whether this can work on your material. Whether it holds up every day, connected to your systems, after the person who championed it has gone on holiday, is a production question. Those take longer to answer.
Palantir can also afford to put its engineers into bootcamps to win contracts worth millions. A small supplier cannot do that for every enquiry, so expect a short test to be scoped and paid for. That still costs far less than finding out three months into a project that your documents were never going to read cleanly.
Frequently asked questions
What is a Palantir AIP bootcamp?
A short, hands-on session, from a few hours to about five days, in which a prospective customer’s team and Palantir’s engineers build a working use of its AI platform on the customer’s own data. Palantir introduced bootcamps in 2023 and described them as its go-to-market approach for that platform. It reported more than 560 of them across 465 organisations by February 2024.
How can a small business test an AI supplier before signing?
Ask them to run their approach on a real sample of your own documents or data, awkward examples included, against a result you agree in advance. Have the person who does the work today review the output, and ask to see what the system got wrong as well as what it got right.
How long should an AI proof of concept take?
For a single, well-defined task, days rather than months. A test that needs a long discovery phase before anything runs is usually testing the wrong thing. Getting from a working test to something used every day takes longer, because it involves your systems and your team.
Flux Dynamics is a fractional CTO who builds. We would rather show something working on a sample of your own paperwork than describe it, and we will tell you if the sample says it is not worth doing. Tell us about the task.