There are three common ways to get an AI system built: hire forward deployed engineers, bring in a consultancy, or engage a systems integrator. They overlap more every year, and the labels are less useful than the questions behind them. This guide sets out the differences and the questions worth asking.
The three models
| Consultancy | Systems integrator | Forward deployed engineers | |
|---|---|---|---|
| Main deliverable | Strategy and a roadmap | A platform implementation | Running software for a specific problem |
| Who writes the code | Usually someone else, later | Large delivery teams | Engineers embedded in your team |
| Where it runs | Not applicable | Often the vendor's platform | Your environment and your account |
| Typical length | Weeks to months | Months to years | Days to months, depending on the model |
| Best when | You do not yet know what to build | You are implementing a known platform at scale | You know the problem, the data exists, and production is the hard part |
Published forward deployed models range from five-day bootcamps[1] to 45-day sprints[2] to about three months with a client[3]. The consultancy line is also blurring: AWS describes enterprise AI as having outgrown the advisory model.[4]
How it is priced
Public pricing is rare across all three. The structures you will meet are time and materials, fixed-length engagements, placement fees for permanent hires, and, so far mostly in AI software rather than services, fees tied to outcomes.[5] Whatever the structure, agree how the outcome will be measured before any work starts.
Seven questions to ask any provider
- Whose account will it run in? If the answer is the provider's platform, ask what happens when the contract ends.
- Who owns the code and the models? Get it in the contract, including how any pre-existing tools are treated.
- Who writes the code, and can we interview them? You should meet the engineers before they start.
- How will the result be measured? A measure agreed before the build, not a success story written after it.
- Who starts the security review, and when? If the answer is “later”, the timeline is already wrong.
- What will the users see? If they cannot question a recommendation, they will not use it.
- What happens when the engineers leave? Ask how the knowledge stays with your team.
Which one fits
If you do not yet know what to build, start with advice. If you are rolling out a known platform across the enterprise, an integrator is built for that scale. If you know the problem you want to solve, the data exists, and the hard part is getting a trusted system into production inside your own environment, forward deployed engineers are built for that. See how hiring FDEs through FastFDE works.
Sources
- Palantir, AIP Bootcamp, accessed 18 Sep 2026. www.palantir.com
- CIO Dive, AWS creates forward deployed engineering hub, 30 Jun 2026. www.ciodive.com
- Salesforce, “Today’s Hottest Role: Forward Deployed Engineer”, 19 Nov 2025. www.salesforce.com
- AWS Partner Network Blog, “Introducing Forward Deployed Engineering for Partners”, 30 Jun 2026. aws.amazon.com
- Sierra, “Outcome-based pricing for AI agents”, 10 Dec 2024. sierra.ai