The second AI use case in an FMCG enterprise almost always costs less and moves faster than the first, and the third moves faster still — but only if someone captured what the first project learned instead of letting it live entirely inside one plant's project files. Without a center of excellence, every plant and every business unit relearns the same lessons about data quality, change management, and model validation independently, at full cost, every single time. A well-designed AI CoE isn't a large centralized team that slows every project down with governance overhead — it's a small, focused hub that handles what genuinely benefits from central expertise, paired with a spoke network of local champions who understand their own plant's specific context. If your third AI pilot is taking as long as your first one did, book a demo to see how a right-sized CoE changes that trajectory.
Your Fifth AI Project Shouldn't Cost the Same as Your First
A right-sized AI Center of Excellence captures what each project learns, standardizes intake and validation, and lets every subsequent use case move faster and cheaper — without becoming a bottleneck that slows the whole organization down.
Why Ad Hoc AI Projects Don't Scale on Their Own
A single successful AI pilot, run by one motivated plant team, is a genuinely good outcome — and a poor predictor of what happens when a second plant tries to replicate it independently. Without a structure to capture and transfer what worked, each new use case starts from zero, repeating avoidable mistakes at full cost.
Each plant independently solves the same data extraction and cleaning problems from similar underlying systems, with no shared library of solutions to draw from.
Without a shared framework, one plant's "validated" model may have gone through far less rigor than another's, creating inconsistent risk exposure across the enterprise.
A hard-won lesson about model adoption or workflow design at one plant never reaches the next plant attempting a similar use case months later.
Separate plants independently negotiating with the same or overlapping vendors miss the leverage and cost efficiency a centralized procurement view would provide.
The Hub-and-Spoke Pattern That Scales Without Becoming a Bottleneck
The most common failure mode in CoE design is over-centralizing — building a large team that reviews and approves every decision, which slows every project down to the speed of the central team's bandwidth. The hub-and-spoke pattern avoids this by centralizing only what genuinely benefits from shared expertise.
The Hub
A small central team owning shared infrastructure, model validation standards, vendor relationships, and the cross-plant knowledge base — deliberately kept lean rather than growing into a bureaucratic approval layer.
The Spokes
Local champions at each plant or business unit who understand their own specific context, drive adoption on the ground, and feed local learnings back into the hub's shared knowledge base.
The Connection
A structured intake and knowledge-sharing process that lets spokes draw on the hub's accumulated expertise without needing central approval for every local decision.
Design a CoE Structure That Fits Your Organization's Scale
See how a hub-and-spoke model maps to your specific number of plants, business units, and current AI maturity level.
Core Roles in a Right-Sized FMCG AI CoE
A CoE does not need a large headcount to be effective — most FMCG organizations can start with a small core team and expand only as demonstrated demand justifies it.
| Role | Primary Responsibility | Typical Staffing Level |
|---|---|---|
| CoE Lead | Overall strategy, use-case prioritization, executive reporting | 1 dedicated or 0.5 FTE at smaller scale |
| Data Engineer | Shared data pipeline infrastructure and integration standards | 1–2 depending on plant count |
| Model Validation Lead | Consistent validation and measurement standards across use cases | 1, often shared with quality function |
| Change Management Lead | Adoption playbooks, super-user network coordination | 1, often shared with operations excellence |
| Plant Champions (Spokes) | Local adoption, context-specific implementation, feedback to hub | 1 per active plant, part-time role |
Funding Models: Central Budget vs. Chargeback vs. Hybrid
How a CoE is funded shapes its incentives and its relationship with the plants it serves. Each model has trade-offs worth considering against your organization's culture and structure.
The CoE operates from a dedicated corporate budget line, removing cost friction for plants adopting new use cases but requiring strong executive sponsorship to sustain funding over time.
Plants pay the CoE for services used, creating natural demand discipline and clear cost attribution, but potentially discouraging early experimentation if plant budgets are tight.
Core infrastructure and shared capability funded centrally, with plant-specific customization or expanded scope charged back — balancing accessibility with cost discipline.
A Standard Use-Case Intake Process
Without a structured intake process, use-case prioritization tends to follow whoever asks loudest or has the most executive access, rather than genuine business value. A standard process makes prioritization defensible and consistent.
Submit
A plant or business unit submits a use case against a standard template capturing the problem, expected outcome metric, and estimated business value.
Screen
The hub screens for data availability, technical feasibility, and overlap with existing or planned use cases elsewhere in the organization.
Prioritize
Screened use cases are ranked against a consistent value and feasibility framework, with the ranking and reasoning made visible to requesting plants.
Deploy and Document
Approved use cases are built using shared infrastructure where possible, with the resulting lessons and reusable components documented back into the CoE's knowledge base.
From 14-Month First Pilot to 3-Month Third Deployment
A multi-plant confectionery manufacturer's first AI predictive maintenance pilot took fourteen months from initial scoping to validated results, largely due to data integration challenges that nobody in the organization had solved before and change management lessons learned entirely through trial and error on the floor. The project succeeded, but the timeline made leadership hesitant to fund a second attempt at a different plant without a clearer path to faster results.
Establishing a small three-person CoE hub — a lead, a data engineer, and a shared change management resource — to document the first pilot's data integration approach and adoption playbook meant the second plant's deployment, using a similar underlying use case, was completed in five months. By the third plant, with a mature intake process, reusable data connectors, and a proven change management playbook, deployment took just three months. None of the individual use cases were technically different in difficulty — the difference was entirely in what got captured and reused from one project to the next.
Frequently Asked Questions
How many active AI use cases justify establishing a formal CoE rather than continuing with ad hoc projects?
Most organizations find the case for a formal CoE becomes clear once they have two or three active or planned AI initiatives across different plants or business units, since this is the point where the cost of independently repeating data integration and validation work starts to clearly outweigh the overhead of establishing shared infrastructure. A single isolated pilot doesn't necessarily need a formal CoE structure, but organizations that already know they intend to scale to multiple use cases benefit from establishing at least a lightweight hub before the second project begins, rather than retrofitting one after duplicated effort has already occurred. Our team can help assess whether your current or planned use-case pipeline justifies a CoE investment — book a demo to discuss your specific situation.
How do we prevent the CoE from becoming a bottleneck that slows down plant-level innovation?
The key design principle is centralizing only what genuinely benefits from shared expertise — data infrastructure standards, validation rigor, cross-plant knowledge — while explicitly avoiding a model where every local decision requires central approval. A CoE that reviews and signs off on every plant-level configuration choice inevitably becomes a queue, while one that provides shared tools, standards, and knowledge that plants can draw on independently accelerates rather than gates local work. Regularly measuring average time-to-deployment for new use cases is a useful check on whether the CoE is helping or hindering. For guidance on right-sizing central involvement, contact our support team.
Should plant champions be full-time roles or part-time responsibilities added to an existing job?
Most FMCG organizations start with plant champions as a part-time responsibility layered onto an existing operations or quality role, since the time commitment for most plants doesn't yet justify a dedicated headcount, particularly in the earlier stages of CoE maturity. As the number of active use cases at a given plant grows, some organizations transition their most active champions to a more formal, partially or fully dedicated role. The specific threshold varies by organization size and AI program maturity. For a staffing recommendation based on your specific plant count and use-case pipeline, schedule a session with our team.
How does the CoE's role change as the organization's AI maturity increases over time?
Early on, the CoE typically focuses heavily on foundational work — establishing data standards, running the first few use cases directly, and building the initial knowledge base — while later, as maturity increases, its role shifts toward governance, cross-plant knowledge curation, and strategic use-case prioritization, with more of the actual implementation work happening at the plant level using shared tools and playbooks the hub has already built. This evolution should be planned for explicitly rather than assumed, since a CoE that never evolves past its early hands-on role risks becoming an implementation bottleneck as demand grows beyond its direct capacity. For a maturity roadmap tailored to your organization's current stage, reach out to support.
What's a realistic timeline to stand up a functioning CoE from scratch?
A lightweight CoE hub — core roles identified, a basic intake process defined, and a shared knowledge repository established — can typically be stood up within six to eight weeks, since the initial version doesn't need to be fully mature to start providing value on the next use case. The knowledge base and shared infrastructure genuinely mature over the following six to twelve months as more use cases pass through the intake process and contribute learnings back into the shared repository. Treating the CoE as a capability that improves iteratively, rather than something that needs to launch fully formed, tends to produce a more sustainable outcome. For a specific stand-up plan for your organization, book a demo to map it out together.
Stop Paying Full Price for Every AI Project
See how a right-sized hub-and-spoke CoE structure lets your organization's third, fourth, and fifth AI use case move faster and cost less than your first.
