How FDE teams are actually structured: Ratios, pods, reporting lines and the first 90 days

Key takeaways

• No authoritative ratio for an FDE team exists in public. The closest published figure is 3 to 8 concurrent accounts per FDE, from one vendor's interviews with nine anonymised programs, and nobody publishes an FDE-to-core-engineer ratio at all.
• The forward deployed engineering team that companies describe is a pod, not a person: an owner who orchestrates and specialists who rotate in, including program managers, architects, data engineers and domain experts. Insight Partners' panel put it plainly, always a pod, never an individual.
• Reporting lines split. ServiceNow, Workato, Rocketlane and Baseten keep FDEs in product or engineering; nCino reports into customer success and Wonderful into a regional CTO. Among 1,000 FDE postings, 14% sat in go-to-market.
• Team size runs from 2 or 3 at the pilot stage to 20 to 100 or more once institutionalised. Ramp, the fintech, went from 2 to about 16, and OpenAI's organisation from 2 at its start in 2025 to about 140.
• The three published 90-day plans we found each end with something going back into the product. PostHog's handbook, Perspective AI's playbook and Vapi's job post agree on that and disagree on almost everything else.

Why this question matters

You're setting up an FDE team, or you're going from two or three engineers to a function, and you want the ratios: how many FDEs per customer, how many per core engineer, who they report to. Every guide you'll find gives you a confident number. The honest position is that the public record gives you a handful of company examples, one vendor-sourced range and some prescriptions, and the structure you pick should rest on which of those evidence types you trust.

The frame: Four decisions that set an FDE team's structure

The sources vary on almost everything, but they're answering the same four questions.

1. Who is in the pod? An FDE alone, an FDE with a deployment strategist or program manager, or a small cross-disciplinary unit.
2. Where does the team report? Product, engineering, customer success, a regional CTO, or go-to-market.
3. How many accounts does each engineer carry, and for how long? Load and engagement length together set headcount.
4. How does a new engineer ramp? What they own in the first month, and what must go back into the product by day 90.

The sections below take them in that order. For the build, rent or partner decision that comes before any of this, read when to hire forward deployed engineers. For what the role is, read what a forward deployed engineer is.

How many FDEs per customer, and per core engineer?

Nobody has published an authoritative answer to either. What exists is below, with the limits of each number stated, because the limit is the useful part.

Read that table as three kinds of evidence. Team sizes are observations about what companies did. The 3 to 8 range is one vendor's sample, and Tandem's own point is that the strain starts inside the range. The productization targets are what people recommend, and Baseten's is what one company says it aims for.

No source gives an FDE-to-core-engineer ratio. The nearest thing is the structural advice that FDE output should flow into the product, which is where the 70% figures come from, and a size story: Ramp's 16 and OpenAI's 140 are headcount without a denominator. If someone hands you a ratio such as one FDE per five core engineers, ask for the source. We couldn't find one.

zoom_in

What does a forward deployed engineering team's pod contain?

The consistent answer is that the unit is a pod with an owner and rotating specialists, and the specific mix varies by company and stage.

Insight Partners' panel of ServiceNow, Databricks, Workato and Wonderful gave the cleanest description: one FDE owns and orchestrates the engagement, and specialists (data scientists, engineers, domain experts) rotate in as the project needs them. Rocketlane's and ServiceNow's pods add program managers and architects to the engineers; nCino hires so that pod members have overlapping rather than identical expertise. Deloitte's offer describes small, self-contained, cross-disciplinary pods, and OpenAI's hiring has moved toward domain experts, including a former investment banker, former scientists and a chip verification engineer.

Palantir's original structure paired an engineer with a deployment strategist. Plank describes the Delta and Echo pairing as one person owning the technical last mile and one owning adoption and executive alignment, staffed separately because they're different problems. Paraform's account of the model adds the timing: early on one FDE wears both hats, and the function splits once deal sizes and customer complexity grow.

Tandem's interviews found four working models in practice: a lone wolf (one to three FDEs owning accounts end to end), an embedded pod, a professional-services-adjacent team, and a standalone function with its own leader. The sample is nine anonymised programs, and the labels describe how a team is run.

Where should an FDE team report?

There's a majority view and there are working counterexamples, and the counterexamples matter because they explain what the majority is protecting.

Most of the company rows are product or engineering. Perspective AI calls reporting into go-to-market the most common structural mistake, though it offers that as a view rather than data, and ServiceNow's panelist framed the choice as whether FDE belongs to the go-to-market motion or to product and R&D. The reason given is the feedback loop: field discoveries reach the roadmap when the team sits next to the people who own it.

The counterexample is nCino, which reports FDEs into customer success and describes the win as the customer building the next deployment itself. A team organised around adoption can sit in customer success and still work, as long as someone owns what goes back into the product. Where the team reports matters less than who pulls field learning into the roadmap, which is why Databricks rotates engineers between the FDE organisation and product, and why the Insight panel named permanent glue between customer and product as the way FDEs burn out.

How does an FDE team's load and engagement length set its size?

Headcount comes from how many accounts an engineer carries multiplied by how long each engagement runs, and the engagement length is the number companies publish.

Insight's day-in-the-life piece has windows of three to four weeks at Rocketlane, six to eight at nCino and 12 to 14 at ServiceNow. F-Prime's Rocio Wu recommends embedding be time-bound, usually three to six months, tied to a milestone such as a successful deployment. OpenAI's head of FDE has said the company should always be temporary at a customer, and OpenAI tells new joiners to expect about half their time on site, with figures near 80% in the UAE and Japan.

The load figure is where the evidence thins. The only per-engineer range is Tandem's 3 to 8 concurrent accounts, and its point is that the working day strains inside that range and context stops fitting in one head near 10 parallel projects. Two outliers show how wide the spread is: one delivery lead went from 4 accounts to 15 after building an internal system that held client context, and one solo FDE at a hardware company covers more than 50 customers and says that's manageable now and will compound as the team grows. Treat 3 to 8 as a starting estimate to test against your own engagement length, and expect the answer to change with ACV and integration depth. Rocketlane's three-to-four-week windows and ServiceNow's 12 to 14 imply very different loads at the same account count.

What does the first 90 days of an FDE look like?

Two of the three published plans keep the first weeks internal or paired, and all three point the first quarter at something returned to the product. They disagree on how soon the engineer is in front of a customer.

Two company practices sit alongside the plans. nCino's FDEs open an engagement by clarifying where FDE ownership of AI adoption ends and where implementation, product and customer success begin, and ServiceNow's start at the C-level to pick one business outcome and one use case. Both are about boundaries, which is what a new engineer most often gets wrong.

PostHog and Perspective AI keep the first weeks internal, and Perspective adds pairing with an existing FDE. Vapi, an early-stage company, puts the engineer on site in month one. All three point the engineer at building something the next engagement reuses. A plan that ends the quarter with only customer deliveries is a consulting plan, and that is Perspective's own warning.

How does an FDE team change as it grows?

The structure at two engineers and at fifty is different, and the transitions are where teams most often misread what they have.

At the pilot stage, one or two engineers cover the lifecycle and, per Paraform, one person wears the deployment-strategist hat as well. Ramp's team started in fall 2023 helping its largest customers, mostly fighting fires in the early days, and now covers customers from late-stage prospect through rollout to long-tail support. Around the point where deal sizes and customer complexity grow, the function splits: Paraform's account of the Palantir model describes adding a dedicated strategist, and Rocketlane and ServiceNow add program managers and architects.

Christian & Timbers describes the next transition as a change of kind: misclassifying one hire is a mistake, while building a 50-person function around the wrong profile becomes an operating-model problem. OpenAI's FDE organisation went from 2 to about 140 while shifting its hiring toward domain experts, and its custom-build share fell from about 90% to about 50% of projects as coding tools improved, per the Next Web's reporting. Retention becomes a structure question too. Databricks rotates engineers in both directions, and Workato builds career ladders to architect level, because of the burnout risk the panel named.

What does an FDE team cost?

Cost depends on stage and on which number you read, and a pod costs more than its FDEs. Perspective AI's table puts all-in year-one cost per FDE at $200K to $250K at seed, $275K to $350K at Series A, $350K to $450K at Series B and $450K to $550K or more at Series C. As arithmetic on that table, three FDEs at Series A is $825K to $1.05M in year one, before the program manager, architect or domain expert a pod may add, and before recruiting. Hiring a permanent senior FDE also takes six to nine months, per seat.

Those are in-house figures; posted salaries, total compensation and fully loaded cost are different numbers for the same seat, and when to hire forward deployed engineers works through the dependencies and the public ranges. Insight's panel found no clean benchmark exists yet.

Which seats in an FDE pod can be rented?

The seats that rotate by project need are the ones that can come from outside. That is a reading of the sources, not a figure from them: in the Insight panel's pod, the specialists who rotate in and out are defined by the engagement, while the owner and the person who carries field learning into the product are defined by the company.

An external senior engineer can fill an engineering seat in a pod, directed by your lead and wearing your badge, and the seat that has to stay yours is the one that decides what goes back into the product. If nobody on your side owns that, the Baseten and Perspective targets above have no one to hit them. The FDE readiness rubric covers how to judge whether an individual engineer can work that way, and embedding contractors alongside FTEs covers how the working arrangement goes.

What to do next

Write down your four decisions before you copy anyone's: the pod you'll run, the reporting line and who owns the loop back to product, your target accounts per engineer and engagement length, and what each new engineer must have shipped to the product by day 90. Check each against the source type: an observation, a vendor's sample, or a recommendation. Adjust the first two for your stage, since the public examples span 2-person pilots and 140-person organisations.

If the pod has a seat you can't fill in time, A.Team puts 1 to 3 named and priced senior engineers in front of you to interview in 3 to 5 business days, at about $25K a month for a senior builder, working under your badge and your team's direction.