The Half-Million Gap

Loaning engineers to customers sounds expensive — a cost center wearing a sales badge. So the motion has to answer a blunt question: does it land contracts that nothing else can reach? The numbers say yes, by an order of magnitude.

Read the ladder

Take the public SaaS companies and measure them by ACV — average contract value: for any given customer, how much is that customer spending with the vendor?

• Palantir: ~$4M per customer
• ServiceNow: ~$1.2M
• Workday: ~$600K
• After that — not a single public SaaS company cracks half a million.

One of these companies runs a forward deployed motion as its go-to-market. It isn't a coincidence that it sits 3× above the next name on the list, at a striking valuation, with only a few thousand employees.

Why the gap exists

The half-million line is roughly the ceiling on configure-it-yourself value: what a customer will pay for software plus the promise that their own people will make it useful. Above the line lives delivered value — contracts priced against the outcome, not the license.

The pitch that lands a global Fortune 500 isn't a feature matrix. It's: "here's the setup — we'll loan you really good engineers that you don't have to hire, recruit, manage, or retain." Engineers trained on the platform, working the customer's problem directly. For a buyer with no engineering depth, that removes the only reason not to sign — the fear that the platform will sit on the shelf.

In the wild

The gap is also why "forward deployed" job titles have spread from Palantir to the AI labs and the agent startups: OpenAI, Anthropic, and a wave of applied-AI companies now field FDE teams, because a frontier model sold to a non-technical enterprise has exactly the same shape as Foundry sold to oil and gas — enormous latent value, no in-house capacity to realize it.

Two readings of the same ladder

The figure above puts Palantir at roughly $4M, which is Kevin Bai's number. Apoorv Agrawal — an Altimeter partner who began his own career as a Palantir FDE — puts it higher when he sets up the same comparison for Colin Jarvis: "the median public software company earns maybe hundreds of thousands of dollars per customer. Palantir is probably the highest end of that, earns about five, six million dollars median ACV per customer."

Both are stated as approximations by practitioners, neither publishes a methodology, and ACV moves with the year, the customer mix, and whether you count government. The site's rule is to leave that alone rather than average it into a false precision: the two sources agree on the structure and differ on the magnitude, and the structure is what the sheet is about. A number that survives an order-of-magnitude claim from two directions is doing its job even when the second digit disagrees.

The question the number can't settle

Then Jarvis declines the frame entirely, and the refusal is more useful than the figure. Asked what the median size of the prize has to be for his team to engage, he does not answer in contract value:

We definitely don't see the FDE function being a function that is going to be like a cast of thousands. Like we definitely want to be very focused, focusing on problems that we think are likely to be generalizable and turn into platform in future, or are going to push our research in a new direction. So we aim at problems that are fairly high value — problems that are going to be saving customers or generating customers to the tune of, like, tens of millions to sometimes the low billions in terms of value.

That is a different denominator. Palantir's ACV measures what the customer pays. Jarvis is measuring what the customer stands to gain, and then selecting on a second criterion that has nothing to do with either: whether solving it teaches the platform something. His team's capacity is split along exactly those two levers — engagements taken because there is a product hypothesis to prove, and engagements taken because an industry has an interesting problem worth pushing research at.

The way the P&L actually goes wrong

This sheet's stated risk is that a dev shop can claim the same contracts and get eaten by maintenance. Jarvis names a second failure that arrives earlier and looks like success while it happens:

I always saw a failure mode in those consulting firms where they had a vision of, like, we're eventually going to be a product company, but unfortunately the short-term lure of services revenue just starts to drag everyone in that direction and then suddenly you kind of lose the strategic view.

Nothing in the ladder distinguishes the two. A services business and a platform business with an FDE motion book identical revenue at identical contract values for as long as it takes for the platform not to get built. His only proposed defence is a decision made in advance about which one you are — "be very clear as to what the purpose of this FDE team is" — and the willingness to enforce it against money: "be prepared to say no at some very difficult times, like when somebody's going to offer you a lot of money to do something that's not strategic."

Which is the sheet's own limitation, restated by someone inside it. The contracts prove reach. They do not prove leverage. What separates the two is the next sheet.