What forward deployed engineering costs

This page is the buyer's cost model. It uses only figures we could trace to a source, and says plainly where no figure exists.

The two verified anchors

Palantir forward deployed engineer total compensation: median $278,000, range $185,000 to $631,000. Self-reported to Levels.fyi, United States, read on 8 September 2026. Total compensation includes base, stock and bonus. Self-reported aggregator data skews toward people who choose to report, so treat it as indicative rather than as a market average.

Median advertised salary $173,816. Across 1,000 postings carrying the title, analysed by Revealera and published in November 2025. This is base salary as advertised, across companies of all sizes, which is why it sits well below the Palantir total-compensation figure.

The two are measuring different things. Use the advertised median for a general market sense and the Levels.fyi figure for what a mature programme at a large company actually pays.

What we will not give you

A frontier-lab band. Levels.fyi does not have enough submissions to show a level breakdown for OpenAI's forward deployed engineer title. Confident ranges for OpenAI and Anthropic appear in many blog posts and we could not trace one to a primary source, so we do not publish them.

A day rate for vendors. Nobody publishes these credibly and they vary by an order of magnitude with geography and model.

An average engagement cost. It depends entirely on deployment length, which nobody has published a defensible distribution for.

Where a number is not knowable, saying so is more useful than guessing. The statistics page lists everything else in the same category.

From salary to actual cost

Salary is the input. The cost is bigger, and the gap is where hiring plans go wrong.

Benefits and employment costs. Payroll taxes, insurance, equipment, software. Your finance team has this multiplier.

Travel. A real line for this role, not a rounding error. 68 percent of postings require travel and 50 to 75 percent is typical where a figure is given. OpenAI states up to 50 percent on its own postings. Somebody at 50 percent travel generates flights, hotels and per diems most of the year.

Unbilled time. An engineer is not on customer work twelve months a year. Holiday, the gap between engagements, internal work, and onboarding to a new account. Two-thirds of the year on customer deployments is a starting planning figure and should be replaced by your own measurement.

Productization time. If you are running the loop, and you should be, that time is funded and not billable. It is the cost that buys the compounding.

Management and support. An account owner, and somebody senior reviewing deployments.

The practical output: cost per engineer-month of actual customer-facing deployment work, which is meaningfully higher than salary divided by twelve.

The comparison that decides it

Put three columns next to each other.

Column one: build it. Loaded cost per engineer-month, times the engineer-months a deployment takes, plus the hiring cost and the ramp. Add a hiring timeline in months, because the delay has a cost too.

Column two: buy it. A vendor's price for the same outcome. Higher per unit of time, and it starts sooner, carries the delivery risk, and has no hiring risk. Three routes to the capability compares the options.

Column three: do nothing. The cost of the deployment not happening. This is the column people forget and it usually decides the answer.

What goes in column three: the revenue from the contract that will not close without the deployment, the renewal at risk, the pilot that dies in place. MIT NANDA reported in July 2025 that 95 percent of generative AI pilots produced no measurable profit-and-loss impact, across 52 executive interviews, 153 leader surveys and 300 public deployments. If your pilot is heading for that outcome, its whole cost so far belongs in column three.

The floor to write down

Before any of the above, one number.

Take the loaded cost of one engineer for the length of a typical deployment. Compare it to the annual contract value of the accounts you would deploy on. If the deployment cost is a large share of the contract, the model does not work at your current pricing.

First Round Review reports Looker validating $25,000 or more in annual contract value before committing to the model, alongside a trajectory toward 2,000 customers and $100 million in recurring revenue. Do not adopt $25,000; do the same calculation and write your own floor down before the first engagement, because working it out afterwards is how a company discovers it has built a services arm by accident.

Why our price is different

Stated plainly because it is the substance of what we sell rather than a claim about quality.

We use AI to write the code instead of hiring more engineers, and every change has to pass an eval suite derived from the specification before it reaches your branch. The saving from that goes into the price rather than the margin. The parts of an embedded engagement that still take human time, discovery, integration and getting through your approval path, are unchanged, so the saving applies to a portion of the work rather than to all of it.

We do not publish a percentage, because no discount has been measured and a number we cannot show the working for would belong in the same bin as the salary bands above.

Next: how to price it, or the contract terms.