Here's what happens in most enterprises: three business units each hire an AI consultant. Each builds a proof of concept on different infrastructure, with different data pipelines, different governance approaches, and different technology choices. Six months later, the organisation has spent three times what it needed to and has three isolated capabilities that don't connect. The missing piece was a platform team.
What You Need to Know
- A dedicated AI platform team owns the shared infrastructure that all AI capabilities build on: data pipelines, model access, governance, and integration patterns.
- Without a platform team, each AI initiative reinvents the foundation. With one, each initiative builds on what came before. The difference is linear cost versus compound value.
- You don't need a platform team on day one. You need one after your first successful AI capability, before you build the second.
- The optimal size is small: 2-4 people initially. The team's leverage comes from what they enable across the organisation, not from their headcount.
- A platform team isn't an AI Centre of Excellence. Centres of Excellence advise. Platform teams build and operate shared infrastructure.
71%
of enterprises report redundant AI infrastructure across business units
Source: Gartner, Enterprise AI Infrastructure Survey 2024
What a Platform Team Actually Does
Owns the AI Foundation
The platform team builds and maintains the shared infrastructure layer: the data pipelines, model access layer, API gateway, governance tools, and integration framework that every AI capability in the organisation uses.
They don't build the AI capabilities themselves. Business unit teams build capabilities on top of the platform. The platform team provides the plumbing; domain teams provide the intelligence.
Prevents Redundant Work
Without a platform team, each project team solves the same problems independently. Data extraction from the CRM. Authentication with the ERP. Document processing pipeline. Vector storage for retrieval. Governance logging.
A platform team solves each problem once and makes the solution available to everyone. The second team that needs CRM data extraction uses the existing connector. The third team that needs document processing uses the existing pipeline with domain-specific configuration.
3.2×
average speed improvement from second to fifth AI capability when built on shared platform infrastructure
Source: RIVER Group, enterprise engagement data 2024
Sets Standards
The platform team defines how AI capabilities are built in the organisation: which models are approved, how data is accessed, what governance is required, how capabilities are deployed and monitored. These standards aren't bureaucracy. They're the conventions that allow different teams to build capabilities that work together.
Standards cover:
- Model access: Approved models, API endpoints, cost management
- Data access: How to connect to source systems, data quality requirements, PII handling
- Deployment: How AI capabilities are deployed, monitored, and maintained
- Governance: Logging requirements, audit trails, risk classification, escalation paths
- Integration: How AI output reaches downstream systems
Enables Self-Service
A mature platform team enables business units to build AI capabilities without deep AI engineering expertise. They provide templates, tooling, documentation, and guardrails that allow domain experts to build on the platform with minimal platform team involvement.
This is the leverage point. A 3-person platform team that enables 10 business units to build AI capabilities produces far more value than a 30-person central AI team that builds everything.
How to Structure It
Roles
| Role | Focus | Count |
|---|---|---|
| Platform Lead | Architecture, standards, stakeholder management | 1 |
| Data Engineer | Data pipelines, connectors, quality monitoring | 1 |
| AI Engineer | Model access, orchestration, deployment | 1 |
| Integration Engineer | API gateway, workflow integration, security | 1 (or part-time) |
Four people. That's the initial team. As the platform matures and the number of consuming teams grows, add capacity, but resist the urge to grow the team faster than the demand justifies.
Reporting Line
The platform team should report to the CTO, CIO, or equivalent, not to a single business unit. If the platform team sits inside one business unit, other units won't trust it to serve their needs impartially.
Operating Model
The platform team operates as an internal product team. The platform is the product. Business units are the customers. The team maintains a backlog, runs sprints, and prioritises based on which platform capabilities unblock the most downstream value.
Critical principle: The platform team should never become a bottleneck. If every AI initiative must wait for the platform team to build something, the model is broken. The team's primary job is building self-service capabilities so others can move independently.
When to Create One
Too early: Before your first AI capability is in production. You don't have enough information to design a platform. Build the first capability, learn what the shared infrastructure needs to look like, then formalise the platform team.
Right time: After your first AI capability succeeds and before you start the second. Take the infrastructure built for the first capability, generalise it, and establish the team that will maintain and extend it.
Too late: After three or more AI capabilities are built on separate infrastructure. You can still create a platform team, but the consolidation work is significant, and the teams using existing infrastructure will resist migration.
The best time to create a platform team is the moment you say "we should probably reuse some of this for the next project." That instinct is correct. Act on it before the next project starts from scratch.
John Li
Chief Technology Officer
The Compound Effect
Here's the maths that makes platform teams compelling:
Without a platform team:
- AI Capability 1: $120K, 16 weeks
- AI Capability 2: $110K, 14 weeks (slight learning, but separate infrastructure)
- AI Capability 3: $115K, 15 weeks (different team, starts over)
- Total: $345K, no shared infrastructure
With a platform team:
- AI Capability 1 + platform foundation: $160K, 20 weeks (higher initial investment)
- AI Capability 2 on platform: $60K, 8 weeks (reuses data pipelines, governance, integration)
- AI Capability 3 on platform: $45K, 6 weeks (platform is mature, mostly configuration)
- Total: $265K, shared infrastructure that accelerates everything after
The fourth, fifth, and sixth capabilities are where the compound effect becomes dramatic. Each new capability costs less and ships faster because the platform absorbs the common work.
Common Mistakes
Building Too Much Platform Before Delivering Value
The platform should emerge from real capabilities, not precede them. If the platform team spends 6 months building infrastructure before any business capability is live, they're over-engineering. Build the platform alongside the first 2-3 capabilities.
Making the Platform Team a Gatekeeper
If business units need to submit requests and wait for the platform team to build things, the team has become a bottleneck. The platform should provide self-service capabilities (templates, APIs, documentation) that let domain teams move independently.
Hiring Only AI Specialists
The platform team needs data engineering and integration skills more than AI model expertise. The integration layer is where most platform work lives. Hire engineers who can build reliable data pipelines and clean APIs. AI model expertise is needed but shouldn't dominate the team composition.
Confusing Platform Team with Centre of Excellence
A Centre of Excellence advises, trains, and produces guidelines. A platform team builds and operates infrastructure. You might need both, but they're different functions. The platform team writes code and runs services. The CoE writes documents and runs workshops.
- We're a mid-sized organisation (500-2000 people). Do we need a platform team?
- If you're planning more than one AI capability, yes, even if it's a team of 2 rather than 4. The alternative is each initiative building isolated infrastructure. At mid-sized scale, you can't afford the redundancy. A small platform team with a clear mandate pays for itself by the second AI capability.
- Can the platform team be external (a partner) rather than internal?
- Initially, yes. A delivery partner can build the platform alongside the first 1-2 capabilities, then transfer ownership to an internal team. This is often the fastest path for organisations that don't yet have AI engineering talent. The key is that the partner builds for transfer: clean documentation, standard tooling, no proprietary dependencies.
- How do we measure the platform team's value?
- Three metrics: (1) Time-to-deploy for new AI capabilities (this should decrease with each capability). (2) Infrastructure cost per capability (this should decrease as shared components are reused). (3) Number of teams building independently on the platform (this should increase over time). If all three trend in the right direction, the platform team is delivering.