Why Palantir Sends Engineers, Not Salespeople, to Sell AI

Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle

Enterprise software sales conjures the same picture for most people: build a sales team, cultivate client relationships, pass POC, sign contracts. Palantir is the exception. This company, once approaching a $500 billion market cap, has built almost no traditional sales army. The front line is staffed by engineers. In 2025, its new contract value exceeded $10 billion (company disclosure).

In an enterprise software industry where customer acquisition costs rise year after year, this playbook deserves dissection. What it sells is no longer software, but a sales machine built around live demonstrations.

Enterprise AI Sales Are Stuck on Trust

Customers buying enterprise AI have heard about large models and agents—information is not scarce. What’s actually missing is trust, for two reasons.

First, AI product effectiveness is hard to imagine before use. What can a large model do for a factory’s supply chain? Explain it a hundred times and the customer still has no visceral sense. You have to run it on their data for them to believe. Second, enterprise software purchase decisions traditionally relied on two forms of trust: relationships cultivated by salespeople and peer case study validation. Both are slow-moving variables—relationships require nurturing, case studies require accumulation. In the AI era, product forms change every six months. Yesterday’s case study may not explain today’s product.

Palantir’s approach replaces these two slow variables with one fast variable: dispatch engineers directly to the customer site, use the customer’s real data, and solve a problem the customer is clearly struggling with within days.

Internally, this approach has a straightforward name: bootcamp. Customers bring real data, engineers help them build a working prototype on the platform in a few days, and decision-makers see results on the spot. Public data provides a benchmark: in 2022, Palantir ran 92 pilots for the year; after scaling this approach in 2023, they ran over 500 sessions in one year, covering more than 400 organizations. The six-month sales cycle common in enterprise software was compressed to around three months.

The mechanism is not complex. Persuasion builds trust through language, demonstration through results. The former takes time, the latter engineering capability. In an era when AI capability itself is rapidly appreciating, the latter is getting cheaper.

A Three-Stage Machine: Enter Cheap, Expand Deep, Exit Clean

Replacing persuasion with demonstration is just a sales technique. What’s truly worth dissecting is the complete growth model behind it. I break it down into three stages: entry, expansion, exit.

The entry stage targets problems that, left unsolved, would cause disasters for the customer: supply chain disruptions, fraud risk, pandemic resource allocation. Then enter at extremely low prices or even free. The most typical example is the UK NHS. In early 2020, as the pandemic hit, the British healthcare system faced challenges in resource allocation and vaccine rollout planning. Palantir signed on for a symbolic one pound, with engineers integrating thousands of independent medical databases within days to build tracking dashboards. The intent of that one pound was direct: buy the fastest ticket into a real, high-value scenario.

The expansion stage, after solving one pain point, spreads outward along the business chain. It relies on a mechanism called the ontology: translating business concepts like parts, suppliers, aircraft, and routes into a unified data structure that AI and systems operate on. The Airbus case is典型. Initially just helping the A350 production line improve efficiency—boosting A350 delivery speed by 33%—it then expanded to supply chain, quality control, and flight data. Eventually, thousands of Airbus suppliers and hundreds of airlines worldwide connected to the same platform. In 2024, both parties signed a ten-year contract worth approximately $1 billion.

Why does this expansion create strong lock-in? Because the essence of the ontology is translating the customer’s business language into your system’s language. The deeper the translation, the higher the migration cost. Switching vendors means re-translating the entire business operation.

The exit stage happens when the platform matures: the customer’s own employees can use it, engineers withdraw from the front line, marginal costs plummet, and gross margins explode. BP has used it for over a decade, with tens of thousands of employees making decisions on the platform. Palantir’s manpower investment in them has decreased while revenue climbs year over year.

The financial results of this machine are quite visible. After 19 consecutive years of losses, it turned its first annual profit in 2023. Net profit more than doubled in 2024 and climbed another level in 2025, with net retention reaching 139%. The Rule of 40 (growth rate plus profit margin), standard in the industry, scored over 100. All figures are from company disclosures or public reports.

But This Machine Has Three Prerequisites

Cold water must be poured. This model looks elegant but is not universal. It stands on three prerequisites.

First, engineering capability must have surplus. Free entry means you can actually solve the problem with engineers in a few days. Without sufficient capability, free demonstrations become paying to do outsourcing for customers—worse, publicly proving your incompetence. Palantir spent over a decade grinding in high-difficulty scenarios like intelligence and defense to build this surplus. Most startups lack it.

Second, results must be verifiable in the customer’s own business language. Demonstrations work because results are concrete: delivery speed up 33%, thousands of databases integrated, prototypes that can be taken directly to the CEO. If your product value can only claim good atmosphere or smooth experience, no matter how lively the demo, it won’t produce persuasive results. The first stage of the machine won’t turn.

Third, there must be an exit mechanism. This is the dividing line between service companies and platform companies. If every customer requires resident engineers, no matter how large the business grows, it’s just high-end consulting. Labor costs are fixed, and gross margins can’t rise. Whether delivery experience can be crystallized into a platform that customers can use themselves determines whether you can reach the third stage.

Beyond the three prerequisites, two facts must be acknowledged. First, Palantir’s success includes an unteachable piece: decades of accumulated security clearances, government relationships, geopolitical positioning, and brand narrative built around that positioning. This anti-Silicon Valley stance is medicine for Palantir but likely poison for ordinary companies. Second, some circulating figures have questionable methodology—for instance, claims of bootcamp conversion rates reaching 92%. That’s the best case under specific conditions. Safer to treat it as an upper bound rather than an average.

Implications for Action

For companies building enterprise AI, three self-check questions suffice.

First, in your product, is there a problem that will cause major trouble for customers if left unsolved? If you can’t find one, go back to finding scenarios. Don’t rush to imitate free demonstrations.

Second, given five days and the customer’s real data, can you produce a result the customer’s CEO can understand? If yes, move demonstrations to the very front of the sales process and compress the cycle. If no, first solidify the product. Don’t learn to give things away.

Third, does each delivery crystallize something reusable: templates, ontologies, industry components? If yes, you’re moving toward a platform. If no, you’re moving toward consulting. Price accordingly.

For buyers, the judgment criterion is simpler. When selecting an AI vendor, see if they dare to demonstrate live with your real data. Don’t look at how polished the PowerPoint is. If they dare, they at least have confidence in delivery. If they don’t, they likely don’t have the goods.

Persuasion has not expired. It’s just that in the enterprise AI market, it’s yielding to demonstration. This shift actually raises the bar for companies: salespeople can be trained, but demonstrations require real capability.

Key points: The bottleneck in enterprise AI sales is trust; demonstration replaces persuasion; a three-stage machine where entry is cheap, expansion deep, exit clean; three prerequisites—surplus capability, verifiable results, exit mechanism—missing any one makes imitation suicide; Palantir’s clearances and positioning are not replicable, some marketing figures have questionable methodology.