Building an AI Champions Network in an AEC Firm

What An AI Champions Network Actually Is

An AI Champions Network is a structured, peer-led group of trusted employees who help colleagues adopt AI tools through demonstration, support, and feedback— paired with light central governance, not centralized control.

It is not a training program. It is not a center of excellence. It is not shadow IT with a friendlier name. The network runs on three operating elements:

• Peer-led: Adoption travels through trust, not training portals.
• Trusted: Champions are picked for social capital, not technical fluency.
• Governed: A central group sets the approved-tool list and the data boundaries, then gets out of the way.

A Champions Network distributes expertise; a center of excellence centralizes it. The difference matters because adoption happens at the edges, not at the center.

Citi has built a network of more than 4,000 AI Accelerators across 182,000 employees in 84 countries, reaching over 70% adoption of firm-approved AI tools³. GitHub's playbook describes the champion role as 30 to 60 minutes a week, embedded in normal work⁴. Microsoft's Responsible AI Champions evolved from volunteer groups into official teams operating under their Responsible AI Council⁵. The structure repeats. The context changes.

Why Top-Down Mandates Fail (And Pure Bottom-Up Fails Differently)

Top-down AI mandates fail because adoption requires trust, not compliance. Pure bottom-up experimentation fails because it fragments— creating shadow AI, security exposure, and learning that never compounds across teams. The Champions Network is the structural answer to both.

The numbers behind that claim are blunt. AI4SP research finds that grassroots AI adoption succeeds about 80% of the time, while top-down mandates fail roughly 80% of the time⁶. Most firms keep choosing the losing strategy.

Bottom-up AI adoption succeeds about eighty percent of the time. Top-down mandates fail about eighty percent of the time. Most firms keep choosing the losing strategy.

Why the gap? Because the actual work of AI adoption is human. BCG's research finds that only ten percent of AI's value lives in algorithms— twenty percent in technology and data, and seventy percent in people, processes, and change management⁷. McKinsey's data tells the same story from the other end: 88% of organizations use AI somewhere, but only 6% qualify as high performers capturing disproportionate value². Eighty-two percent of organizations are spending money on AI without getting much back. That is not a model problem. It is an adoption problem.

The bottom-up failure mode looks different but lands in the same place. IBM's 2025 research finds that more than 80% of workers already use unapproved AI tools, and one in five organizations has experienced a breach linked to unsanctioned AI⁸. When a breach happens, shadow AI adds an average of $670,000 to the cost⁸. When everyone improvises, the cheapest mistake gets expensive fast.

The case for an AI governance strategy that prevents shadow AI is not a separate conversation from the case for adoption. They are the same conversation.

The Architecture: Three Tiers That Make It Work

An AI Champions Network has three tiers: a small steering committee that sets policy and approves tools, a working group that operates the program, and a champion network that drives peer adoption inside the business. Each tier solves a different problem, and skipping any of them is what causes most programs to stall. The ratios in the table below are coverage-oriented guidelines drawn from large-firm case studies; Section 6 right-sizes them for AEC scale.

The steering committee owns governance. The working group owns operations. The champions own adoption. When any tier collapses into another, the program fails.

Champions themselves do three things, mostly. They run live demos for their teams— ten minutes at the start of a project meeting, showing one workflow that worked. They answer Slack questions when peers get stuck on a prompt or a tool. They surface friction back to the working group so the next champion does not hit the same wall. GitHub's playbook describes the role as a "choose your own adventure"— it fits inside work, not on top of it⁴. That sentence is doing real work. The moment the role becomes "a second job," champions burn out, and the program collapses.

Citi runs the same loop at scale. Their AI Accelerators are local points of contact, not formal trainers, embedded in the teams they support³. No pay or promotion is attached. Internal badges create visibility. The work is voluntary, structured, and measured against use-case validation rather than hours-billed.

PwC Netherlands offers the proof at scale. They scaled from 300 AI enthusiasts to all 6,000 employees over roughly a year, using organizational network analysis to identify the most naturally influential people in the business⁹. The scaffolding was not magic. The selection process was rigorous. The program had governance from day one.

AI4SP's research finds that grassroots success rises from about 80% to roughly 90% when intervention is added— what they call "guided grassroots"⁶. Governance is the guide. The network is the ground game.

Who Belongs In The Network— Not Who You'd Think

Your best AI champions are not your most technically fluent employees. They are your most trusted peers— the people whose project-management opinion already carries weight in a meeting. Everett Rogers identified this pattern in 1962, and it still holds.

Within an organization, certain individuals are termed "champions" who stand behind an innovation and break through opposition.¹⁰

That definition is older than Microsoft. It is still right. The technical employee teaches the tool. The trusted peer changes the room.

There are two valid ways to pick champions:

• Volunteer-based. Open a call. The people who raise their hand are usually the ones whose teams will follow them anyway. GitHub uses this approach⁴.
• Network-analysis-based. Use a tool to map who actually influences whom inside the firm, then recruit from that map. PwC Netherlands used this⁹.

Either works. What does not work is naming "the IT person" or "the AI nerd" and assuming peer adoption follows. It does not. The technical depth lives in your working group or external partner. The champion's job is trust and translation.

Picking criteria that actually work:

• People other team members already ask for opinions
• Cross-discipline credibility (an architect respected by engineers, a PM respected by everyone)
• Existing curiosity about AI— not expertise
• Time to invest 30–60 minutes a week without crushing their billable hours
• Comfort showing unfinished work to colleagues

Picking criteria that quietly destroy programs:

• Technical fluency as the primary filter
• Title seniority as the proxy for influence
• IT or operations only

Building an AI culture that lasts starts in the same place: the people, picked correctly.

Right-Sizing The Network For An AEC Firm ($20M–$100M)

At a 100-person AEC firm, you do not need a steering committee, a working group, and a champion network. You need a principal who owns it, two or three trusted champions across disciplines, and a monthly review. The principles are the same. The scaffolding is much smaller.

Citi's structure does not transplant to a fifty-person engineering firm. The principles do. The scaffolding does not.

The reality your firm is operating in: PE-credentialed engineers who are skeptical of corporate change-management language. Billable-hour pressure that makes any "extra time" cost real. A regulated environment where mistakes have client consequences. Deep pride in craft. The framework has to respect all of that or it dies on contact.

The AEC adoption gap is real. Fifty-three percent of AEC firms now use AI tools, but only 27% use AI for anything that actually changes the business¹¹. Most firms have started. Most firms have not finished starting.

What stays: peer-led adoption, governance pairing, time commitment respected. What goes: separate steering committee, formal working group, organizational network analysis as a process. This is the right-sized version we recommend at AEC firms in the $20M–$100M range. It is editorial framework, not a third-party finding, and we name it that way so you know what you are looking at.

The economic case is concrete enough to take to your partners. Bluebeam's 2025 report found that 68% of AEC AI adopters have saved at least $50,000, and 46% have saved 500–1,000 hours¹². The work pays. The scaffolding is cheap.

A pattern we see at AEC firms in the $20M–$100M range: a principal who already trusts AI personally is the one who unlocks the program for the rest of the firm. Skepticism about AI is fine. Indifference is what kills the program before it starts.

The founder's AI decision framework is built around exactly this kind of right-sizing problem.

Your First 30 Days: A Practical Sequence

The first thirty days of an AI Champions Network are not about deploying tools. They are about identifying the people, agreeing on the boundaries, and running one or two visible quick wins so the rest of the firm has something concrete to copy.

Don't start with the tool. Start with the person and the use case. The tool is the easiest decision in the room.

1. Days 1–7: Name the program. Pick the people. Principal-sponsor announces internally. Identify two or three champions across disciplines (architects, engineers, project managers, marketing— not IT). Approve the initial tool set: likely Microsoft Copilot, ChatGPT Team, Claude— whatever is already paid for, with admin-tier configuration verified so paid-tier data-handling actually applies. Write a one-page data-handling rule: no client data into public ChatGPT, paid tiers only.

1. Days 8–14: One workflow per champion. Each champion picks one repeatable workflow. Examples: proposal narrative drafting, RFI response support, project meeting summaries. They run it for two weeks. They document what worked. They document what failed.

1. Days 15–21: First peer demo. Champions show their workflow to one team meeting each. Friction feedback gets collected from the room— what would block someone else from using this tomorrow? The working group (informal, in your case) captures the friction list.

1. Days 22–30: Decide what scales. Publish an approved-tool list and an approved-use-case list. Schedule the monthly review. Pick the next two workflows to attempt in days 31–60.

This is the sequence governance lives inside. The reason most champions networks devolve into shadow AI⁸ is not because champions are reckless. It is because no one wrote down the boundaries early. Day 7 is the cheapest insurance you can buy.

Where Champions Networks Fail

Champions networks fail in four predictable ways. Naming them up front is operational hygiene, not pessimism.

Most champion programs don't die from a bad strategy. They die from skeptical department heads, burned-out volunteers, and silent governance gaps.

1. No executive air cover. A skeptical department head can quietly kill a champion's time. "We need her on this proposal— can the AI thing wait?" The principal-sponsor must protect that 30–60 min/week explicitly. Without that protection, every champion eventually loses the time war.

1. Burnout from over-reliance. When two champions are doing the work of ten, they quit. Rotation matters. Recognition matters. Clear scope matters. The role is 30–60 minutes a week. When it becomes five hours a week, the program is broken— not the champion.

1. The governance gap. Without an approved-tool list and a clear data boundary, a single mistake— client data into a free LLM— becomes the story that ends the program. IBM's data is direct: shadow AI breaches add an average of $670,000 to incident cost⁸. One incident pays for years of governance work.

1. Champion-as-IT-tech anti-pattern. When the program defaults to "let's pick the technical people," peer adoption breaks. Trust beats fluency. Every time.

These failures are not exotic. They are the four modes that show up in almost every program that stalls. Naming them in your first principal-sponsor meeting is cheap. Discovering them at month nine is not.

FAQ: Common Questions From AEC Firm Leaders

These are the questions AEC principals ask in our first conversation about a champions network.

Should our champions be technical employees? No. The most effective champions are trusted peers across disciplines— project managers, senior engineers with social capital, operations leads. The technical depth comes from your working group or an external partner. Rogers' opinion-leader theory and GitHub's playbook both converge on this¹⁰⁴.

How long until we see results? Two to six weeks for the first validated use cases at a champion level. Three to nine months for broader adoption inflection across the firm. PwC Netherlands took roughly a year to scale from 300 to 6,000 employees, and they had a much larger organization to drive through⁹. Measuring AI success at a smaller firm should be tracked at the champion-and-workflow level, not the enterprise level. One champion validating one workflow in week three is more meaningful than a firm-wide adoption percentage at month nine.

How is this different from a center of excellence? A center of excellence centralizes expertise and gates access. A champions network distributes expertise and accelerates access. CoEs work for compliance-heavy domains where access control is the point. Champions networks work for adoption-heavy ones where peer trust is the point. Most AEC firms in the $20M–$100M range need adoption first. A CoE arrives later, if at all.

What does it cost? Mostly time— 30–60 minutes per week per champion, embedded in normal work⁴. Tooling cost depends on existing AI subscriptions; most firms already pay for Copilot or ChatGPT Team. The hidden cost is principal-sponsor attention— roughly two hours a month for the first quarter, less after. That attention is what protects champion time from skeptical department heads who would prefer it spent elsewhere.

What if we have only 50 people— is this overkill? Yes for the three-tier structure. No for the principles. At 50 people you need a principal-sponsor, two champions, and a monthly review. Same logic, smaller scaffolding. This is editorial recommendation grounded in the framework's principles, not a finding from a public case study. The case studies are all at firms with thousands of employees because that is who has been willing to publish.

Build The Network You Already Have

Your firm already has an AI network. People are sharing prompts in DMs, copy-pasting client data into the wrong tools, and quietly succeeding without anyone noticing. The question is not whether you have one. It is whether you are going to give it a structure.

Both top-down mandates and shadow-AI experimentation are losing strategies. The Champions Network is the both/and answer— peer trust on the inside, governance on the outside, real adoption in the middle. The framework is older than Microsoft and the data is from 2025. Both inform what works.

Build the network. Pair it with governance. Right-size it for your firm. The principal who already trusts AI personally is the one who unlocks it for the rest. The trusted peer is the one who carries it across the project meeting and into Monday's RFI response.

Both Are True. All Of It Matters.

If structuring an internal AI champions network for your firm feels worth a conversation, designing an AI implementation approach is the kind of work we do with AEC firm leaders every week. Most of the work is not technical. It is in picking the right two or three people, naming the boundaries, and then getting out of their way.

References

1. IBM, "What is Systems Network Architecture (SNA)?" — https://www.ibm.com/docs/en/zos-basic-skills?topic=implementation-what-is-systems-network-architecture-sna
2. McKinsey & Company, "The State of AI: How organizations are rewiring to capture value" (March 2025) — https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
3. AI News, "The quiet work behind Citi's 4000-person internal AI rollout" (2025) — https://www.artificialintelligence-news.com/news/the-quiet-work-behind-citi-4000-person-internal-ai-rollout/
4. GitHub, "Playbook series: Activating your internal AI champions" (2025) — https://resources.github.com/enterprise/activating-internal-ai-champions/
5. Microsoft Inside Track Blog, "Responsible AI: Why it matters and how we're infusing it into our internal AI projects at Microsoft" (2024) — https://www.microsoft.com/insidetrack/blog/responsible-ai-why-it-matters-and-how-were-infusing-it-into-our-internal-ai-projects-at-microsoft/
6. AI4SP, "Why Fortune 500 AI Strategies Fail While ChatGPT Soars" (2025) — https://ai4sp.org/fortune-500-ai-strategies-fail-while-chatgpt-soars/
7. Boston Consulting Group, "From Potential to Profit: Closing the AI Impact Gap" (2025) — https://www.bcg.com/publications/2025/closing-the-ai-impact-gap
8. IBM, "What Is Shadow AI?" (2025) — https://www.ibm.com/think/topics/shadow-ai
9. Lead with AI, "AI Champion Programs: Why, Who, How" (2025) — https://www.leadwithai.co/guides/ai-champion-programs
10. Wikipedia, "Diffusion of Innovations" (citing Everett Rogers, Diffusion of Innovations, 5th ed., 2003) — https://en.wikipedia.org/wiki/Diffusion_of_innovations
11. Building Design + Construction (citing Deltek Clarity 2025), "AI in AEC: Where firms should start and how to scale adoption" (2025) — https://www.bdcnetwork.com/aec-tech/article/55359703/ai-in-aec-where-firms-should-start-and-how-to-scale-adoption
12. Bluebeam, "New Bluebeam Report Shows Early AI Adopters in AEC Seeing Significant ROI Despite Uneven Adoption" (October 2025) — https://press.bluebeam.com/2025/10/new-bluebeam-report-shows-early-ai-adopters-in-aec-seeing-significant-roi-despite-uneven-adoption/