How Treasure AI Built an AI Center of Excellence

Two years ago, Treasure AI set out to be deliberate about AI from the start. We formed a cross-functional AI committee, drafted our first AI policy, and mapped departmental use cases so adoption fit each function rather than being imposed on it. We built an AI champions program to carry that thinking into every team, not just IT.

That foundation was real, and every tool in use had cleared review on its own merits. What we didn't have was a shared view across them. Each ran its own console with its own usage data. Departments tracked their AI independently, so no one held a complete picture across the company. Reporting was manual, and there was no single place to see what had been approved or point employees to what they could already use.

We had the intent right. We had governance running tool by tool. What we didn't have was a system that connected the two.

What We Built

The answer wasn't another tool or a thicker policy. It was one place: an AI Center of Excellence (CoE), built for two audiences. The employees who use AI to augment their day to day work, and the leaders who steer and fund it.

The AI CoE makes everything findable. There's a full, searchable catalog of production AI agents and Skills for internal use-cases, each with a golden path so someone can go from "is there a Skill for this?" to running it in a few minutes. Discovery stopped being word of mouth.

The Skills inventory is the piece I'd point to first. Every internal Skill the company has built, more than a hundred now, lives in a single internal GitHub repo, and the inventory syncs from it in near-real-time. Skills are sorted into more than twenty categories by the team and use case that owns each one: Engineering, Customer Support, Product, Security, company-wide, and more. That sounds mundane until you sit with what it unlocks. It's one place to see what's already been built, what hasn't, and where the gaps are, so teams stop quietly rebuilding each other's work and the org can see where it's worth investing next. Work that used to live in scattered repos and people's heads is now a shared, visible asset.

Additionally, the AI CoE puts governance where the work is. AI policies, review processes, department guides, and our build-vs-buy frameworks sit right alongside the tools they apply to, instead of being buried three systems away from the people who need them. When the guidance is one click from the work, following it stops being a chore.

Finally, the AI CoE turns usage into intelligence. A live command center shows adoption and utilization across every AI platform on a single screen, so leadership can weigh a renewal, a redundancy, or an expansion with real usage data behind it instead of a gut call.

Find it, govern it, measure it, all in one place.

The centralized hub isn't only a dashboard, either. It's paired with a companion Enterprise AI Confluence space that carries the depth: a growing library of real use cases, practical governance guidance (ie, Model Context Protocol [MCP] security best practices), and the full detail behind every policy and review process. And because tools alone don't drive adoption, there's a human layer on top. A company-wide AI community in Slack, and weekly office hours run by our ITS team, so anyone can ask a question or learn what's working from someone who's already tried it.

We're Our Own Customer

One aspect of our AI Center of Excellence is built on our own product: Treasure AI iCDP. The iCDP is used to create a reliable, cross-platform view of how AI is being adopted and used across the company.

Every AI platform reports usage differently, and none of them share a common schema. Each day, automation pipelines sync usage data from source AI systems, including Claude Code, Treasure Work, and other AI platforms, into Treasure AI iCDP, where it is normalized into a unified model. This gives our Enterprise AI function a consistent foundation for measuring adoption, utilization, and trends across tools, teams, and time.

Treasure Work adds value on top of that foundation by making the data usable. We use it to analyze the iCDP data, surface insights, and give leaders a practical way to understand what is happening across the AI portfolio. Because we ingest Treasure AI employee usage telemetry into the iCDP every day, we are not looking at a one-time snapshot. We are building a time series. That is the difference between asking, “How many people used AI this week?” and asking, “How is adoption trending across the year?” The first is a status check. The second gives our Enterprise AI function the longitudinal view it needs to guide AI strategy, evaluate investment, identify underused or duplicative tools, and determine where additional enablement is needed.

This is a concrete example of how our own products work together: Treasure AI iCDP provides the governed data foundation, while Treasure Work helps turn that data into analysis, visibility, and action for the people responsible for driving AI adoption across the enterprise.

What Changed, For Both Audiences

For the people using AI day to day, the shift is about friction. What used to be discovered by word of mouth is now a searchable catalog with golden paths, and getting unstuck went from asking around to self-service in minutes within our AI CoE. Because the policies and guidance sit right next to the tools, the governed path and the easy path are finally the same one.

For the people steering and funding it, the shift is about visibility. Usage that used to be scattered across separate applications, reported manually and quarterly if at all, now sits on one screen, refreshed daily and trending over time. Governance that had no home has one. And renewals, redundancies, and expansions became decisions backed by evidence instead of instinct.

Same platform, two very different payoffs. That was the point all along.

This Is Version One

Everything here is version one, and deliberately so. The point was to prove that a single, governed home for AI could work, and it does. From this foundation, our Enterprise AI function takes it further, with deeper cost tracking down to the tool and team, formal ROI measurement that ties the value AI delivers back to what it costs, and a next wave of intelligence that answers not just "how much are we using this?" but "what is it worth?" The system is in place. The next iterations make it sharper.

The Real Transformation

Two years in, we didn't need more AI. We needed one place for it. Somewhere the employee could find the right tool and the executive could see the whole board. We built that. A scatter became a system, and that made all the difference.