Forward Deployed Engineers: How AI Companies Actually Charge for Them

Almost nobody bills forward deployed engineers by the hour. Companies with a “forward deployed” motion charge for it in one of four ways: they bundle it into the enterprise contract, charge a paid pilot or implementation fee, recover the cost through usage or outcome-based pricing, or sell a fixed-scope pod. In a GoodFirms survey of 133 companies running an FDE function, only 3.6% used time-and-materials billing. (GoodFirms)

That’s the short answer. The more useful question is which of those structures fits your product—and how to bill it without turning your software company into a consultancy.

What “forward deployed” actually means

A forward deployed engineer is your engineer, working inside the customer’s world. They sit with the customer’s team, wire your product into their data, identity provider, CRM and approval flows, and stay until it’s running in production.

Palantir made the role famous. AI companies made it mainstream.

The reason is simple: an AI pilot that works in a demo often dies in integration. The model performs on the customer’s data, then six months disappear into permissions, data pipelines and edge cases. FDEs exist to close that gap.

The title has also escaped engineering. You’ll now see:

• Forward deployed product managers. Scale AI hires FDPMs to “own product outcomes” across a portfolio of enterprise accounts, working alongside its FDEs. (Scale AI)
• Agent product managers. Decagon employs a team of technical PMs who set up support agents with each customer. (a16z)
• Forward deployed designers, sales and solutions roles. Earlier-stage, but the pattern is the same: put the expert next to the customer, not behind a ticket queue.

And the money behind it has become serious. In May 2026, OpenAI launched the OpenAI Deployment Company with more than $4 billion of initial investment and a base of 150 FDEs from its acquisition of Tomoro. (Constellation Research) In October, Anthropic committed $100 million to train 10,000 “Frontier Deployed Engineers” at customers and partners by the end of 2027.

Why AI companies are choosing services on purpose

For a decade, SaaS founders were told services were a margin killer. Andreessen Horowitz’s Joe Schmidt made the opposite case in Trading Margin for Moat: AI apps that try to replace core workflows with thin integrations break, and the companies that do the hard implementation work end up owning the workflow and the data.

He points to the incumbents. ServiceNow went public at roughly 63% gross margin and Workday at roughly 54%. By 2024 they were at 79% and 75%. Heavy services early didn’t stop them becoming great software businesses—it’s how they got embedded.

The a16z advice on pricing is blunt: sell services at cost, and optimize for total gross profit rather than an 80% margin target in year one.

So the strategic question isn’t whether to do forward deployed work. It’s who pays for it, and when.

The four ways companies charge for forward deployed work

1. Bundled into the contract (“free”)

The FDE cost is folded into a large platform price, or simply treated as customer acquisition cost. The customer never sees a services line.

Palantir is the model here. Its AIP bootcamps let prospects build on their own data with Palantir engineers before buying. Palantir told investors more than 915 organizations went through bootcamps in Q1 2024, and described one customer signing a paid engagement after a single day, then a seven-figure deal three weeks later. (Nasdaq transcript) Alex Karp framed it as spending on letting partners “use our software in order to decide what works” instead of on marketing.

In the GoodFirms survey, 25% of companies with an FDE function fold the cost into enterprise contract pricing rather than billing it separately.

Works when: contracts are large, conversion is high and the platform expands across departments. Breaks when: you’re selling $50K deals and spending $150K of engineering time to land them.

2. Paid pilot or implementation fee

The customer pays a fixed, upfront fee for a defined deployment—often a 6–12 week pilot that ends in production, not a slide deck.

Sierra is a reported example: third-party pricing analyses put its implementation fees at around $50,000 per account, on top of the ongoing contract. (Quiq) Consultancies selling FDE-style work describe the same shape: a short assessment, then a fixed-scope pilot.

Works when: you need buyer commitment and a budget line before your engineers show up. Breaks when: the fee becomes the product. If customers start negotiating scope like a consulting SOW, you’ve drifted into professional services.

3. Recovered through usage or outcome pricing

No separate services line. The vendor deploys hard, then earns it back through a per-conversation, per-resolution or per-task price.

AI customer support is where this is most visible. Third-party reports describe Sierra’s outcome contracts at roughly $1.50 per resolution with annual starts near $150,000, and Decagon pricing as a platform fee plus per-conversation or per-resolution usage. (Quiq, Voiceflow) In the GoodFirms survey, 21.4% of companies price their FDE work purely on outcomes.

This is the most aligned model: the vendor only gets paid well if the deployment actually works. It’s also the riskiest, because the deployment cost is fixed and the revenue depends on volume and success rate. (We walked through that math in outcome-based vs consumption-based AI pricing.)

Works when: the result is measurable and inside your product’s control. Breaks when: low volume or low success rates leave you never recovering the upfront engineering.

4. Pods and retainers

A fixed monthly fee for an embedded engineer or small team, sometimes tied to agreed OKRs. FDE-as-a-service providers quote roughly $8,000–$30,000 per engineer per month. (ilmtec) In the GoodFirms data, OKR-linked retainers are 10.7% and fixed-scope pricing 7.1%.

This is the natural model for services firms and systems integrators. For a product company, it’s a warning sign if it becomes permanent.

And the most common answer: a hybrid

The single largest group in the GoodFirms survey—28.6%—uses a hybrid pricing structure: some combination of deployment fee, platform subscription and usage or outcomes.

Why hourly billing lost

Hourly billing rewards the wrong thing. The FDE’s job is to get the customer to production and then become unnecessary—every hour they don’t need is a win. Billing by the hour makes efficiency a revenue loss, invites line-by-line timesheet audits and positions your best engineers as contractors.

The whole point of forward deployment is that the services drive software revenue. Hourly billing makes services the revenue.

The structure we’d use: paid, credited, capped

If you’re building a forward deployed motion, here’s what we recommend for most B2B AI companies:

1. Charge a fixed deployment fee. A real number, invoiced upfront. It qualifies the buyer and gets the project a budget owner.
2. Define exit criteria, not hours. “In production for the support team, handling X% of tickets” beats “240 hours of engineering.”
3. Credit the fee against year one. If the customer signs the subscription, the deployment fee becomes a credit. The customer feels it was free; you’re covered if they walk.
4. Cap the embedded period. Six to twelve weeks, then the FDE rolls off. Ongoing optimization becomes a priced add-on or part of a higher plan—not open-ended free labor.
5. Ramp the subscription. Start lower while the deployment is maturing and step up as usage arrives, using a ramp deal or a minimum commitment that grows over the term.

Internally, track FDE cost per account and payback period. It’s the only way to know whether you’re “trading margin for moat” or just running an expensive consultancy with a software logo.

Make sure your billing can express the deal

Forward deployed deals are messy on paper: a one-time deployment fee, a credit that only applies if the customer converts, a ramped platform fee and usage or outcome charges—often on a single order form.

If your billing system can only handle a flat subscription, someone ends up tracking credits and ramps in a spreadsheet, and finance discovers the problem at the first invoice.

Bunny handles one-time, recurring and usage-based charges on the same subscription, along with ramp deals and discounts on the quote, so the deal sales signed is the deal that gets invoiced. If you’re pricing a forward deployed motion, talk to us about how to model it.