Forward-deployed engineer, consultant or integrator: who does what in your AI project?

A forward-deployed engineer (FDE) builds the AI solution directly in your systems, with your data and your teams, then stays until you can run it yourselves. A consultant analyses and recommends, and the implementation is left to you. An integrator installs and configures software you have already chosen. All three are legitimate lines of work. What they deliver differs widely, and so does the price. This guide from Djtal tells them apart so that you know what you are buying.

Where does the forward-deployed engineer come from?

Palantir invented the role and coined its name more than a decade ago. Instead of delivering software and a training course, the company sent engineers to work on the client’s premises and build the solution there, in the live systems. The model has since become central to enterprise AI. At the end of June 2026, AWS announced a USD 1 billion investment in its own forward-deployed engineering organisation (CNBC, 30 June 2026). The reason is that the bottleneck in AI projects has moved. The models are good. Companies now lack engineers who can build them into their actual processes.

The comparison in one table

	Consultant	Integrator	Forward-deployed engineer
Deliverable	Analysis, recommendations, roadmap	Software installed and configured, users trained	A solution running in your systems, with teams able to run it alone
Where do they work?	At their own desk, from your documents	On your platform, within the scope of the software	Inside your systems, with your data, under your governance
Accountable for	The quality of the advice	An installation that meets the specification	A measured operational result
Typical duration	A few weeks	Varies with the project	8–16 weeks, then handover
After they leave	The implementation is still to be done	The software runs; changes depend on the integrator	Your teams run the solution, backed by runbooks
When to choose them	A strategic decision needs clarifying	The software is chosen and the scope is stable	AI has to deliver a result in your processes

Each role has its place. A company torn between three ERPs needs advice. A company that has chosen its CRM needs an integrator. A company that wants an AI agent to handle its orders, tickets or accounting needs someone who answers for the result. That is the FDE’s territory.

The fourth-role trap: the solutions engineer

A fourth role slips in among these three, and it rarely announces itself: the pre-sales engineer, or solutions engineer. Before you sign, they build a convincing demonstration on anonymised data. The demo works, because that is their job. One question protects your budget: who stays once the contract is signed, and which data do they work on? In 2025, MIT found that 95% of organisations are getting zero return on their generative AI investment (MIT NANDA, The GenAI Divide: State of AI in Business 2025). The hardest step lies between the demonstration and production, and the only role that takes that step with you is the one that works in your real systems.

Four questions to ask before you sign

1. “Which data will you work on?” Ours, in production, or a demonstration dataset?
2. “Who owns the operational result?” A recommendation delivered, or a metric measured before and after?
3. “What happens when you leave?” Training on the tool, or a full handover with runbooks, documentation and a team trained on the process?
4. “How do you test what you build?” The answer should contain the word evaluations (or evals): test sets rerun after every change and versioned like code.

A serious supplier, whatever their role, answers these four questions without hesitation. Their answers tell you what you are buying.

What about French-speaking Switzerland?

The FDE model is still concentrated among the giants, with budgets sized for large enterprises. Djtal runs it on a scale that fits companies in French-speaking Switzerland. Our engineers embed with the client’s team, build on the existing ERP or CRM (Zoho, Odoo, Dynamics) and hand over autonomy at the end of the engagement. The full method is on our forward-deployed engineering page, and the key terms are defined in our business AI glossary.

To see why so many projects stop at the demonstration, read why AI projects fail.