when to create an AI centre of excellence explains how an organization turns AI ambition into owned decisions, reliable delivery, responsible operation, and measurable business results.
This guide is written for Canadian organizations. The recommended structure should be adapted to company size, sector, portfolio, skills, sourcing, existing controls, and the potential impact of each use.
Table of contents
- Portfolio demand is sustained
- Teams repeat the same work
- Scarce skills cause bottlenecks
- Platforms need enterprise ownership
- Control inconsistency creates risk
- The mandate is clear
- Review after 12 to 18 months
- Practical checklist
- Frequently asked questions
- References
“Create a centre of excellence to solve demonstrated enterprise problems, not to signal ambition. If its purpose, customers and measures cannot be stated clearly, strengthen ownership and working practices first.”
Mehrzad Verdizadegan,
CEO, Praevion Consulting Inc.
Portfolio demand is sustained
Create a centre when multiple business units have a continuing AI portfolio, not one or two exploratory ideas.
Temporary demand may be served by a virtual working group or external specialists.
Teams repeat the same work
Duplicated vendor reviews, architecture, data access, evaluation, and governance suggest shared capability could reduce cost and delay.
Confirm the duplication with evidence before adding a new function.
Scarce skills cause bottlenecks
A centre can pool platform, evaluation, data, product, or responsible-AI skill that individual units cannot support.
Do not centralize common business ownership or domain knowledge away from the work.
Platforms need enterprise ownership
Shared models, identity, integration, monitoring, and cost controls require an owner when use grows.
The service should have users, funding, standards, and performance measures.
Control inconsistency creates risk
Repeated policy exceptions, weak evidence, or uneven vendor controls may justify central standards and higher-risk review.
Independent control functions must keep their mandates.
The mandate is clear
Define customers, services, decision rights, funding, staffing, sourcing, transfer rules, and success measures before launch.
Decide whether the centre advises, builds, operates, governs, or combines these roles.
Review after 12 to 18 months
Test reuse, delivery time, quality, capability growth, risk outcomes, and whether business ownership improved.
Change or close the centre if it becomes a queue, duplicates control functions, or creates dependence without value.
Questions for the next operating-model review
Ask whether decisions sit with people who have authority, delivery teams can access the capabilities they need, governance matches risk, and business owners can show realized value. Check delays, rework, duplicated tools, control gaps, weak adoption, and systems that remain in operation without a clear owner.
when to create an AI centre of excellence checklist
- Define the outcomes, portfolio scope, and accountable executive.
- Assign business, product, technical, data, and control ownership.
- Map the lifecycle from idea and procurement through operation and retirement.
- Set risk-based decision rights, evidence, approval, and escalation.
- Provide shared data, technology, learning, vendor, and governance services.
- Measure decision time, adoption, outcomes, cost, incidents, and realized value.
Related Praevion guidance
- Read the related Praevion operating-model guide
- Explore the next related article
- Explore Praevion Consulting Inc. digital transformation services
Frequently asked questions
Is a centre needed at the start?
Usually not. Begin with named ownership, a virtual team, approved tools, and a few controlled use cases.
How large should it be?
Start with the smallest team that solves proven enterprise problems and scale only when demand is sustained.
Where should it report?
Place it where it can serve business units, influence platforms, and reach executive decisions without taking away outcome ownership.
Executive takeaway
When Should an Organization Create an AI Centre of Excellence? The practical answer is to design around decisions and workflows, not titles alone. Keep business value close to operating leaders, share capabilities that benefit from scale, and make accountability visible from discovery through retirement.
To discuss your needs, contact Praevion Consulting Inc..
References
- Statistics Canada, Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026
- OECD, BCG and INSEAD, The Adoption of Artificial Intelligence in Firms, 2025
- OECD, Skills in the AI Age, 2026
- NIST, Artificial Intelligence Risk Management Framework
- ISO/IEC 42001:2023, AI management systems
- Government of Canada, Guide on Departmental AI Responsibilities
