How to pick your first FDE project

Written by
Last updated on:
November 17, 2026

A forward-deployed engineer can only be as useful as the project you give them. Here's how to find AI use cases worth the effort and choose one that's likely to reach production.

Most teams have more AI ideas than time to build them. Maybe it’s a customer chatbot, or someone’s plan to automate the whole finance team. The tricky part is knowing which one to start with, and that’s where learning how to identify AI use cases pays off.

If you’re bringing in a forward-deployed engineer (FDE), the choice is worth some extra care. An FDE is a senior engineer who works inside your team and stays with the project from discovery until it’s running in production. A well-chosen first project usually ends with a working system and a clear case for the next one. With a vague or oversized project, even a strong engineer can spend weeks just working out what success means.

MIT’s NANDA initiative found that only about 5% of AI pilot programs achieve rapid revenue acceleration, as reported by Fortune. Picking the right project is one of the simplest ways to improve your odds. It’s also the first step on your AI roadmap, since a project that delivers tangible value gives you a concrete result to point to when you plan the AI initiatives that come after it.

New to the role? Read What is a forward-deployed engineer?

Two colleagues in business attire stand by a sunlit window, reviewing something on a laptop one of them holds. It reflects a team comparing and scoring a shortlist of AI use cases to choose a first project for a forward-deployed engineer.

How to identify AI use cases worth an FDE's time

Good use case identification starts with the work, not the technology. The best first projects rarely come from a brainstorm about what artificial intelligence could do. They come from looking closely at business processes where work is slow or expensive today, and at the pain points the people doing that work already know about. These are the places to start:

  • Repetitive tasks with clear inputs and outputs. Checking documents against rules or moving data from one system into another. If people follow a checklist to do it, an AI solution can often take on the routine tasks and free them up for higher-value work.
  • Back-office processes. The same MIT research found that more than half of generative AI budgets go to sales and marketing tools, but the biggest return came from back-office automation. Work like invoice processing or compliance checks can make a safer first project than a customer-facing one, since mistakes can be caught internally before they reach a customer.
  • Metrics with a gap to target. The US General Services Administration's AI guide recommends choosing problems tied to operational or strategic priorities, especially KPIs with significant gaps to their targets. If a team keeps missing a turnaround time or an accuracy goal, that's a strong lead.
  • Stalled AI pilots. A demo that worked but never reached real users is often a good FDE project. The idea has already been tested, and the hard part left is connecting it to real systems and the people who'll use it, which is exactly the work FDEs do.

It’s important to talk to the people doing the work, too. The GSA guide recommends interviewing a system's actual users, so you solve a real problem rather than an assumed one, and it advises keeping that user research going throughout the project. The person processing 200 invoices a week usually knows exactly which 20 cause the most trouble.

It also helps to look beyond one team. Ask other business units what slows them down, since the same problem may show up in several business functions. Collecting new ideas this way gives you a longer list to choose from, and it involves more teams early in your AI adoption.

Where AI use cases show up across the business

A useful way to sort potential AI use cases is into three broad categories. Grouping your ideas this way makes it easier to see what each one needs:

  • Automating routine work. AI agents and generative AI tools can handle repetitive tasks that follow clear rules, like routing requests or drafting first versions of documents, with human review before anything goes out.
  • Better data analysis. Machine learning models can find patterns in historical data and turn data points into forecasts that help identify trends early.
  • Working with language and documents. Generative AI can work with unstructured data, like contracts and internal documents, because it can read and summarize natural language.

Here's how those categories often look in practice, by team:

Example AI use cases by business unit, with the main data sources each one needs
Business unitExample use caseMain data sources
OperationsDemand forecasting, or flagging equipment that may need maintenanceHistorical data and sensor data
Customer serviceDrafting replies and summarizing cases to improve customer experience and customer satisfactionPast tickets and customer data
Sales teamsPrioritizing accounts based on intent signals and account historyCRM records and website activity
Marketing teamWriting first drafts of product descriptions, or summarizing market research and market trendsProduct data and research reports
Human resourcesAnswering policy questions and supporting knowledge sharingInternal documents and policies

Keep in mind that these are examples, not a checklist. The right first project depends on your business priorities and where your data is strongest. Some of the most useful projects capture specialized knowledge that only a few experienced people have and make it available to the rest of the team.

What makes a good first FDE project?

Once you have a few candidates, check each one against a short list of traits. The best first projects have most of them:

  • One business process and one owner. The project should tie to key business priorities, and someone on the business side should own the result, define what success looks like, and open doors when the FDE needs access or answers.
  • A result you can measure. Time saved per case or a lower error rate. Look for a measure tied to operational efficiency or cost reduction, or to risk reduction if the process is error-prone. If you can't measure it today, set up the measurement before the project starts.
  • Data the FDE can reach. Data readiness affects how much work a project takes. The GSA guide counts the effort to acquire, ingest, and wrangle data as part of a project's cost. Check data availability first: the data should exist and be accessible without a six-month approval process. Then check data quality, since AI models working from incomplete or inconsistent records can give unreliable answers. Waiting for access can stall a good project for weeks.
  • A scope measured in weeks. Aim for something that can reach production in one to three months. A smaller win that ships beats a larger one that's still in planning next spring.
  • A low cost of being wrong. For a first project, choose work where a person can review the AI's output before it reaches a customer or a ledger.
  • Users who want it. A team that's asking for help will give better user feedback and use the system once it's live.

A project that checks every box is rare. If a candidate is missing more than two of these, it's probably a better second or third project. Your AI maturity matters here, too. The GSA guide recommends starting small, with a single use case limited in scope, and growing from there.

How to score your shortlist

When two or three candidates look equally good, a simple scoring exercise helps narrow them down. The GSA guide groups the factors into impact, effort, and fit. Rate each candidate from one to five on each, then compare:

How to score AI use cases for a first FDE project: impact, effort, and fit, with questions to ask and what a high score looks like
FactorWhat to askA high score looks like
ImpactHow much would solving this move a business metric?Hours saved every week, or a KPI the business already tracks, like cost or business growth
EffortHow hard is the data and integration work?Clean, accessible data and a small number of systems to connect
FitCan your team support and own it after launch?A clear owner, a willing team, and budget to keep it running

When you rate effort, give easier projects higher scores. Then add up the three scores. The best first project is usually the one with the highest total, even if another candidate has a bigger potential impact.

High-value, low-effort projects are your quick wins and the strongest first choices. High-value opportunities that need a high-effort build, such as projects where the data isn’t accessible yet, can go on your AI roadmap for later, once the first project has shown real value.

Share the scores with the FDE early. Part of their job is discovery, and they may spot integration problems or quick wins that change the ranking. They can also check which AI tools and AI capabilities fit each candidate, and whether to build or buy. The GSA guide recommends studying the market before you decide.

Which projects should you avoid first?

Some AI use cases are worth doing eventually but make poor first projects:

  • Company-wide transformations. Rolling out AI across your business operations at once is a big goal. For a first project, pick one process and start there.
  • Fully automated, customer-facing decisions. Approving loans or answering customers with no review step adds risk before you've built trust in the system.
  • Projects without an owner. If nobody on the business side will own the result, the FDE will build something nobody is accountable for using.
  • Problems that aren't really AI problems. Sometimes a slow process needs a better form or a simple integration. Before you apply AI, ask what the AI adds that a simpler fix wouldn't. An FDE will tell you if the answer is "not much," but it's cheaper to notice before they start.
An engineer leans in with a notepad and pencil while a colleague points at code on a monitor lined with sticky notes. It reflects a team documenting decisions and planning the handover as they set up a first project with a forward-deployed engineer.

How to set up your first FDE project

Once you've chosen the project, a little preparation saves weeks. Before the FDE starts, confirm the business owner and target metric, and request system and data access. An FDE who spends the first two weeks waiting for credentials is a bit like a contractor who arrives to remodel your kitchen and finds the house locked.

Plan the handover from the start, too. Decide who will run and maintain the system once it's live, and ask the FDE to document decisions and pair with your engineers along the way. That way, the first project leaves your team with a working system and the knowledge to build the next one.

For a full checklist, download our Forward-Deployed Engineer Guide.

Starting your first project with FullStack

If you've found the right use case and need someone to build it, you can hire forward-deployed engineers through FullStack. Our FDEs are senior nearshore engineers who work in your time zone and start with discovery, so they can help confirm the project is the right one before they build. Each one is assessed on production code and enterprise integration, and you can review the recorded coding challenge and interviews yourself.

We send a shortlist within 48 hours of your first call, and the first two weeks are free. If it isn't working, we swap the engineer or stop, and you don't pay for those weeks.

Have a use case in mind? Talk to our team about your first FDE project.

Learn more

Frequently asked questions

The key principles are simple. Start with work that's slow or expensive today, especially back-office processes and metrics that keep missing their targets. Interview the people who do the work, then check each idea for a clear owner and a result you can measure with data you can reach.

A good first project covers one business process with one owner, has a result you can measure, and uses data you can already reach. It should be small enough to reach production in one to three months and low-risk enough that a person can review the output.

An FDE handles both discovery and engineering, so they can refine the use case and build the AI solution in the same engagement. That helps when the problem is clear but the requirements aren't yet written down.

Aim for something that can reach production in one to three months. If a project needs longer, look for a smaller first phase that delivers a measurable result on its own.

A stalled pilot can be a strong first FDE project. The idea has already been tested, and an FDE can focus on getting it into production: connecting the system to real data and the people who'll use it.