Forward deployed engineer or implementation consultant: which does your AI plan need?

If the work is setting up a product you have already chosen, use an implementation consultant. If the work is making an AI plan land inside your teams, so that finance, operations and service actually work differently, you need a forward deployed engineer, full-time or fractional.

The two roles answer different questions. One asks how the product should be configured. The other asks whether the plan works in your business, and changes the plan when it does not. Most AI programmes need both, at different moments, and the expensive mistake is hiring one to do the other's job.

A forward deployed engineer works inside the business, not beside it. The model was pioneered by Palantir in the early 2010s, according to Andreessen Horowitz, and AI companies have copied it because so many AI projects stall before they reach production.

The job descriptions vary, but the work comes down to four things:

• Embed with the teams. Sit with finance, operations, sales or service until each workflow is understood from the inside, including the exceptions nobody wrote down.
• Turn the strategy into scoped use cases. Translate a plan written at the top into concrete work with owners, data needs and a measure of success.
• Connect the AI to real data. The definitions, integrations and hand-offs that decide whether the model sees the business as it is. This is usually where the plan meets the most resistance, which is why getting the data ready comes first.
• Stay until it runs. Not until the demo. Until the team that owns the process uses it without being asked.

A forward deployed engineer is measured on whether the outcome lands, not on how much of a product gets configured. One thing to check: many FDEs are employed by a software vendor, and their job is to make that vendor's product succeed in your business. That is useful, but it is not the same as someone whose only interest is your plan.

An implementation consultant sets up a product to a specification. They usually come from the vendor's side of the table, often as a certified partner, and they know the product deeply: its configuration options, its limits and the fastest way to get it live.

That is real expertise, and on a defined scope it is the right expertise. A good implementation consultant will tell you when a requirement fights the product, deliver to a plan and hand over a working system.

Their limits are structural rather than personal. They work to the statement of work, so anything outside it becomes a change request. They are paid for effort or for the build, so their commercial interest is in the scope being delivered, not in whether the scope was the right one. And they leave at go-live, which is usually when the business starts to find out whether the plan works.

“We have chosen the CRM and need it live by April.” Use an implementation consultant. The decision is made, the product is known and the work is configuration against a deadline. A forward deployed engineer would cost more and add little.

“The board approved an AI plan six months ago. Two pilots worked, and nothing is in daily use.” You need a forward deployed engineer. The gap is not configuration. It is that nobody owns making the plan land in the teams that run the processes, and nobody is checking whether the plan survives contact with them.

“We are replacing the ERP, and we want AI in finance after that.” You need both, in order. An implementation consultant configures the ERP. Someone on your side, in the forward deployed role, makes sure the move fixes the data your AI plans depend on and holds the consultant to the business case. Leave the second role out and the new ERP inherits the old data, and the AI plans wait another year.

Once you know you need the forward deployed role, the next question is how to fill it.

• Hire full-time when AI is a permanent part of how the business runs, there are several use cases a year, and you can find and keep someone who knows both the technology and your functions. That last part is the hard one. Andreessen Horowitz reported in January 2026 that job postings for the role had risen by hundreds of percent over 2025, so the market is thin.
• Fractional, on a retainer when the plan is live but the work comes in waves: a set number of days each month alongside the CTO, rising and falling with the programme.
• Per project when there is one use case or one function to land, from plan to working, with a clear end.

Hiring is the right call more often than people selling the alternatives tend to admit. If the role will be needed every week for years, build it in. Fractional and project work suit everything before that point, including working out whether the role is needed at all.

I work in the fractional and project models myself, so I have an interest here. The test above is the one I would apply to my own business.