Palantir's sales to American companies passed its sales to the American government in the quarter that ended in September, a first for the company. The more useful fact is how those companies buy. The deals are quick to start, short in length, and able to grow a great deal afterward.
A short contract also looks small on paper. It adds less to a company's booked future sales even when the yearly amount is the same, which is why Palantir's booked contract value fell from one quarter to the next while its revenue rose 20%. Investors read that as weakness. A customer who is learning what the product does reads it as a safe way in.
Cloud computing made the same change in 2006
Before Amazon launched its rentable servers in August 2006, a company that wanted more computing power bought a machine and planned to use it for years. Amazon then offered a small server for 10 cents an hour, billed to the customer's account. A start-up, or one department inside a large firm, could try an idea for a few dollars and stop whenever it liked.
Nobody needed approval for a multi-year purchase, so people tried things, and the ones that worked kept growing. The sellers that won were the ones whose product earned its expansion. Fifteen years later the problem had turned around: companies routinely spent more on the cloud than they had planned. Easy entry means easy growth, and AI bills already rise with use the way electricity bills do.
Pilots are growing into much bigger deals
Palantir described four customers whose pilots, worth about $8 million together, became deals worth about $76 million a year. An industrial materials company went from a pilot of nearly $3 million to a $29 million yearly deal a quarter later. Palantir chose these examples, so they show the best case. What carries over is the pattern: a small start that the customer can judge on its own results.
The same ease works against the seller. Many of Palantir's contracts let the customer end them early for convenience. A customer who can say yes in weeks can also say no in weeks, so the seller has to keep proving itself after the first sale, not only before it.
What this changes for a buyer
The trial can be small enough for an ordinary business. A regional food distributor with twelve depots could try AI route planning at one depot for 90 days. If the planners there get ten hours a week back, that is about 500 hours a year at one depot, and about 6,000 hours if all twelve get the same result. The depot manager, who sees the work every day, can judge that number better than a vendor's slide deck, and the vendor only wins the other eleven depots by showing the saving in one quarter.
That shift in power only helps if you use it. Before the trial starts, write into the contract the price for the other eleven depots, how you can leave, and a monthly spending limit. A short trial makes it easy to say yes, and as the cloud showed, the bill that follows is the one most buyers did not plan for.