An AI-native transformation needs three roles, not one. A Center of Excellence sets the guardrails and approves what gets built. A vendor, implementer or forward deployed engineer builds it. And an AI project manager inside the business carries each system into the workflow and owns it after launch. In many companies, the first two are in place and the third is missing.
| Key facts
68% Share of AI value generated in business units, against 21% in enabling functions and 11% in technology and IT. Source: BCG Applied AI Index 2026. 3× AI workflows scaled by companies that keep risk controls central but let business and technology teams carry out the work, compared with companies that steer everything from the center. Source: BCG. ~50 + 100 A central team of about 50 and 100 local deployment teams: how one global outsourcer in BCG’s report organized its AI rollout across more than 5,000 frontline staff. Source: BCG. |
Why “the AI team” isn’t the whole answer
Ask who’s leading AI, and most companies point to the center: a Center of Excellence, an AI office, a chief AI officer, a CIO. That’s the right place to start. Strategy, standards and guardrails belong at the center.
But the value lands somewhere else. Executives in BCG’s Applied AI Index 2026, a survey of 1,330 senior leaders, place roughly two-thirds of AI’s value (68%) in business units. Enabling functions account for 21%, and technology and IT for just 11%. The work has to reach the places where that value sits.
BCG also found that companies which keep risk controls central but let business and technology teams carry out the work scale three times as many AI workflows as companies that steer everything from the center. The figures are self-reported, and the sample leans toward large companies. The pattern is still worth taking seriously: set the rules at the center, and carry the work out where the business is.
The three roles
The Center of Excellence approves
It sets the rules: which use cases get funded, what each risk tier requires, and what every AI system must have before it launches, such as an owner, evaluation criteria, monitoring and a way to stop it. It approves and audits. It shouldn’t become the team that has to build everything, or every request waits in one queue.
The vendor or forward deployed engineer builds
Vendors and implementers bring product depth and speed. Forward deployed engineers, the vendor-side specialists who work on site to integrate a product with a customer’s systems, are very good at getting a system working inside a real environment. That’s real value, and most companies will keep buying it. Our forward deployed engineer alternative explains how that role and an in-house owner complement each other.
The AI project manager carries it and owns it
This person sits inside the business, works with the people whose workflow is changing, and is accountable across every vendor and model type. They own the problem definition, the workflow redesign, the go-live and everything after it: evaluation, exceptions, changes and value. When the vendor’s team rolls off, this person is still there.
| Stage | Center of Excellence | Vendor or forward deployed engineer | AI project manager |
| Choose the use case | Approves it against the portfolio and risk appetite | Advises on what’s feasible | Frames the problem and success measure with the business owner |
| Set guardrails | Sets enterprise standards | Builds to them | Applies them to this workflow: what the system decides alone, where people sign off |
| Build and integrate | Reviews at stage gates | Builds the system | Runs the plan: data access, dependencies, change control, training |
| Launch | Signs off on readiness | Supports go-live | Owns go-live and the first weeks of exceptions |
| Run and improve | Audits | Available under contract | Owns evaluation, changes and value reporting |
| Change, pause or stop | Sets the policy | Makes changes on request | Holds the authority day to day |
A typical split, not a mandate. Adapt it to your own governance model.
What goes wrong when the third role is missing
The pattern is familiar. The pilot works. The vendor’s engineers know exactly how the system behaves, because they built it. The Center of Excellence signs off. Then the engagement ends, and the system starts to drift: a data source changes, a policy changes, users find workarounds, and nobody inside notices until a customer, an auditor or a quarterly number does.
Nobody did anything wrong. The role that owns the system after launch just didn’t exist. We cover that moment in who owns your AI after go-live, and what owning it looks like each week in what makes a company AI-native.
BCG describes a similar split in practice. A large global business process outsourcer paired a central team of roughly 50 people with 100 local deployment teams, and brought its new operating model to more than 5,000 frontline employees. Rather than deciding AI’s autonomy tool by tool, it decided it for each process. A small center, with deployment carried out close to the work.
| Name the three
For your most important AI initiative, write down three names: who approved it, who built it, and who will own it a year after launch. If the third line is a vendor or blank, that’s the gap. |
How companies staff the third role
- Hire an experienced AI project manager. Fastest if you can find one. The candidate needs to learn your business, and you’re competing for them with every other company deploying AI.
- Upskill a project manager you already trust. They know the business and the people. What they usually lack is practice owning an AI system in production, which is hard to get from a course alone.
- Grow one through a Registered Apprenticeship. Slower to start, but the person learns the role on your own deployments and stays on your payroll.
Master of Project Academy sponsors the nation’s first registered apprenticeship dedicated to AI Project Management, registered by the Maryland Apprenticeship and Training Council (MATC #2575, September 8, 2026) under the registered occupation Information Technology Project Manager (O*NET 15-1299.09). The registration was announced via Business Wire.
Apprentices are full-time employees of the participating employer from day one. Our employer guide explains how companies join.
Frequently asked questions
Who should lead an AI transformation?
Strategy and guardrails belong with senior leadership and a central AI team or Center of Excellence. Building often involves vendors or forward deployed engineers. Each AI system also needs an owner inside the business, such as an AI project manager, who carries it into the workflow and owns it after launch.
What does an AI Center of Excellence do?
It sets enterprise standards and guardrails, approves which AI use cases move forward and audits systems in production. In BCG’s 2026 survey, companies that kept risk controls central but let business and technology teams carry out the work scaled three times as many AI workflows as fully centralized ones.
What’s the difference between a forward deployed engineer and an AI project manager?
A forward deployed engineer usually works for the vendor and integrates that vendor’s product with the customer. An AI project manager works for the customer, is accountable across every vendor and model type, and stays to own the system after the engagement ends.
| Fill the third role from inside
The AI Project Management apprenticeship, registered under the Information Technology Project Manager occupation (O*NET 15-1299.09; MATC #2575), develops one of your own employees into the AI project manager who owns deployment after launch. |
About Master of Project Academy
Founded in 2012, Master of Project Academy has served 500,000+ enrollments across 180+ countries in project management training. It sponsors the AI Project Management apprenticeship, registered under the Information Technology Project Manager occupation (MATC #2575). Learn more at masterofproject.com/p/aipmapprenticeship.
Sources