Forward Deployed Engineer Interview: Customer Call Practice
By Saurav Sharma · November 18, 2026
A forward deployed engineer (FDE) works directly with customers to understand what they need and build software they can use. Customers rarely hand over a complete specification. A customer conversation round checks whether you can find the real requirement before you start building. This guide gives you three role-play scenarios, a scoring rubric, and a 30-minute format that two people can run without any special setup.
What a forward deployed engineer does
An FDE is a software engineer who ships working systems at a customer's site or inside the customer's cloud environment. The job mixes code, integration, and talking with the people who own the problem. Typical work is wiring a model into the customer's data, deciding what to build first, and explaining trade-offs to a non-engineer.
The customer side of the job shows up in the hiring data. From aijobprep's study of 3,647 AI-engineering job postings at 198 companies, collected August 2026, the topic "customer-facing / forward-deployed" appears in 1,212 postings across 136 companies. Only five of the 39 topics in the study appear at more companies. Separately, we counted postings whose title contains "forward deployed" (hyphenated or not). There are 340 of them, at 71 companies.
Many employers also hire cloud solutions architects and AI engineers into customer-facing seats. If a posting for any of these roles mentions working directly with customers, use this drill to practice that part of the job. This is a practice format built from the skills the postings ask for. It does not copy any named company's interview.
Check this: open three postings for the role you want. Underline every sentence that mentions customers, stakeholders, or requirements. That list is your practice agenda.
Why a customer conversation round exists
Technical rounds test whether you can build. A customer round tests whether you build the right thing. In customer work, the first request often hides the customer's actual problem. A team asks for a chatbot when they need shorter call handling time. A team asks for a model when the data cannot leave their network.
An interviewer can see three things in a short role play. They see whether you ask questions before you propose. They see whether you can change a plan when a constraint appears. They see whether you speak in plain words to someone who is not an engineer.
Check this: write down the last time a requirement changed after you started building. Ask which question, if asked on day one, would have caught it.
How to run the drill
You need two people, a timer, and the briefs below. The candidate reads only the candidate brief. The customer reads the customer brief, the sample questions, and the rubric, and keeps them hidden until the call ends. Here is the 30-minute plan.
- Minutes 0 to 3: the candidate reads only the candidate brief. The customer reads only the customer brief, including the hidden constraint.
- Minutes 3 to 18: discovery call. The candidate asks questions. The customer answers honestly, but only what is asked. The customer never volunteers the hidden constraint.
- Minutes 18 to 24: the candidate states a proposal in five sentences or fewer and names the first thing they would build.
- Minutes 24 to 30: score the call together. To swap roles, start a new 30-minute round with a scenario the new candidate has not read.
The customer player should answer in the voice of a busy manager. Short answers. Some business words. No technical jargon unless the candidate uses it first. If the candidate asks a direct question about the constraint, such as "Where does the data live?", the customer answers it fully. If the candidate asks a vague question, such as "Tell me about your setup", the customer answers vaguely.
Check this: record the call on your phone, audio only, with permission from your partner. Listen back for how long you talked before your first question.
Scenario 1: Support ticket assistant
Candidate brief
You are an FDE at an AI vendor. A mid-size software company wants an assistant that reads support tickets and suggests replies. You have a 15-minute discovery call with their head of support operations. Your goal is to leave with a clear project scope and a first deliverable you can start next week.
Customer brief (hidden from the candidate)
You run support operations. You have 40 agents and about 6,000 tickets a week. Your visible request is "a model that drafts replies." Hidden constraint: ticket data contains customer account details and cannot leave your cloud account. Your security team will reject any design that sends tickets to an outside service. You will say this only if asked where the data can go, or what security review applies. If asked, you also say the review takes about four weeks for any new external vendor.
Questions that uncover the constraint
- Where do tickets live today, and who owns that system?
- Is there any data in tickets that cannot leave your environment?
- What security or compliance review would a new tool go through, and how long does it take?
- Who signs off on sending data to a model provider?
How the proposal should change
Before the candidate learns the constraint, a reasonable proposal is a hosted drafting tool. Once it surfaces, the proposal must keep ticket data inside the customer's cloud account. The candidate should say so plainly: ask the customer's security team to confirm where the model processes and logs ticket data, then pick a setup that keeps both inside the customer's account, such as a model service offered in their own cloud or a model they host themselves. They should also scope the first build to a small ticket category, so the security review covers one narrow flow. The candidate should name what they will need from the customer, such as read access to a ticket export with account fields masked.
Rubric for scenario 1
- Asked about data location before proposing (2 points).
- Changed the architecture after the constraint, with a reason (2 points).
- Named a narrow first build that fits the review window (1 point).
- Named one thing they need from the customer next (1 point).
Scenario 2: A chatbot request for a call center
Candidate brief
You are an FDE. A customer's operations director has asked for a chatbot for their call center. You have a 15-minute call to decide what to build. You do not yet know how success is measured.
Customer brief (hidden from the candidate)
You run a call center with 200 agents. Your visible request is "a chatbot." Hidden goal: your real target is to cut average handle time (the minutes an agent spends per call) from 11 minutes to 8. Most of the time is lost while agents search three internal systems during the call. A customer-facing chatbot would not help, and your callers dislike bots. You will share the handle-time target only if asked how you measure success, or what you would look at on a good day versus a bad day. You do not volunteer the three-systems detail unless asked what agents do during a call.
Questions that uncover the constraint
- How do you measure success for this project, and what is the number today?
- Walk me through what an agent does during a typical call.
- Where do agents lose time?
- If the chatbot worked perfectly, what would change in your report next quarter?
How the proposal should change
One strong option is an agent-assist tool instead of a customer-facing chatbot. It would search the three systems during the call and show the answer on the agent's screen. The first build could cover only the top five call reasons. The candidate should define a before and after measurement of handle time on those call types, so the customer can see the result in a month. They should also say what they would not build yet.
Rubric for scenario 2
- Asked how success is measured (2 points).
- Asked the customer to describe a typical call and where agents lose time (1 point).
- Changed the solution when the real goal appeared (2 points).
- Proposed a measurable first build with a number to track (1 point).
Scenario 3: Model output in front of end users
Candidate brief
You are an FDE at an AI vendor. A health insurance company wants a feature that explains benefit rules to members in plain language. You have 15 minutes with their product lead. Aim to leave with a scope and a plan for who must agree to it.
Customer brief (hidden from the candidate)
You are the product lead. Your visible request is "an assistant that explains benefits to members." Hidden constraint: legal must approve the wording members see. Legal only reviews templates whose wording is fixed in advance. If the model writes a new reply for each member, it cannot ship. You will mention legal only if asked who must approve the feature, or what happens before a member-facing release. If asked, you say legal review takes six weeks, and the legal team has not heard about this project yet.
Questions that uncover the constraint
- Who has to approve this before members see it?
- Has that group seen the project yet?
- Are there rules about what wording members can be shown?
- What is the worst outcome if the assistant says something wrong?
How the proposal should change
A strong candidate shifts from free-form answers to a design legal can review. Examples: the model selects from a set of pre-approved explanation templates, or it fills approved templates with the member's own plan values. The candidate should suggest bringing legal into the next meeting, because legal approval takes the longest. They should also ask legal what staff can use during the six-week review, for example an internal tool whose output staff read but do not send to members.
Rubric for scenario 3
- Asked who approves the output before launch (2 points).
- Changed the design to fit review, with a specific option (2 points).
- Proposed how to involve the approver early (1 point).
- Offered a lower-risk first release (1 point).
What your practice partner should listen for
These listening points come from the skills the postings ask for. They are our reading of the data. No employer published them as a scoring sheet.
- Questions before answers. Before suggesting a solution, the candidate asks what the customer wants to change and how the work gets done today.
- Questions about the work and the goal. Good candidates ask about the person's day, the number they report, and who else has a say.
- A change of plan with a reason. When a constraint appears, the candidate says what changes and why.
- Plain words. Terms such as "handle time" are used as the customer uses them. Technical terms are defined in a sentence.
- A small first step. The candidate names one small thing the team can build first and which approvals it needs before release.
- A named next action. The call ends with who does what by when.
Check this: after each practice round, mark each listening point as seen or missed. Pick one point you missed and practice it in the next round.
Common mistakes
- Pitching too early. Before you propose anything, summarize what you heard and ask the customer if you got it right.
- Asking only technical questions. Data formats and APIs matter, but the constraint in these scenarios sits in security, goals, and approvals. Add at least one question about each.
- Accepting the first request as the goal. Ask what changes if the request succeeds.
- Hiding behind jargon. If your customer would need to look up a word, replace it.
- Keeping the plan fixed. Explain out loud how the customer's answers shaped your proposal. If your first idea still fits, say why.
- Ending without a next step. Close with one action for each side.
Score yourself and repeat
Each scenario scores out of six points. Compare scores across rounds and note which question uncovered the customer's main concern. For a second attempt, ask your partner to change the hidden detail, so you practice discovery instead of recall. Then write your own scenario. Pick a hidden constraint from a real project you know, such as a rate limit, a data residency rule (data must stay in one country), or a budget owner who is not on the call.
Pair this drill with the technical questions for the role. For AI engineers, see AI engineer interview questions. For cloud solutions architects, see cloud solutions architect interview questions.
Practice the technical side of a customer-facing interview process: work through the forward deployed engineer interview questions, then run the three scenarios above with a partner this week.