How PTC Built an Enterprise-Grade Partnership AI Agent With the Crossbeam MCP

PTC is a publicly traded product lifecycle management company with more than 7,600 employees and roughly $3 billion in annual revenue, known for products like Creo, Windchill, Codebeamer, and Onshape. Arena, one of PTC's businesses and itself a multi-tenant cloud PLM and quality management platform, is one of several parts of PTC putting Crossbeam's Ecosystem Intelligence to work.

At a company that size, no single team can answer every partner question in real time, and no AI agent gets built without clearing the company's own governance bar first.

Heatherly Bucher, Director of Strategic Alliances at Arena, manages a partner network that spans referral partners, marketplace and technology partners, value-added services partners, and platform technology partnerships. Getting the right partner in front of the right customer, consistently, across a company built from multiple business units, meant her team needed something more durable than tribal knowledge and Slack threads.

To close that gap, Heatherly's team built a partnership AI agent in Microsoft Copilot Studio, connected to the Crossbeam MCP, Gong, Crayon, and a set of curated SharePoint folders. Nine months after launch, most of Arena's customer-facing teams use it as their default source for partner questions, and the partnerships team now runs early partnership evaluations through it as well.

"We're super excited about the Crossbeam MCP. We're super excited about Crossbeam." — Heatherly Bucher, Director of Strategic Alliances

The problem: partner knowledge at enterprise scale

Arena runs a large, varied partner network inside a much larger parent company, and Heatherly's team supports both the sellers and Customer Success Managers who need partner answers in the moment and the partnerships team that manages the relationships themselves. Before the agent, that meant reps asking in Slack, digging through SharePoint, or waiting on a reply from partnerships, the same bottleneck that shows up at any company once headcount and partner count both start to scale.

Every answer took time, and quality depended on who happened to respond.

The stakes were higher because of how many partnership types Arena runs at once, and because PTC, like most large public companies, treats AI adoption as a governance question, not just a productivity one. Heatherly's team had to work within PTC's existing AI committee, which sets rules for what tools are allowed and what data they can touch, before they could build anything.

"We do have corporate support and access to tools, and we have a very robust AI committee at PTC with what's allowed and what's not allowed and how we can use AI. As a publicly traded, large company, we do have to know those things, otherwise we would get in trouble with it, and our compliance." — Heatherly Bucher

That governance requirement shaped the agent from day one. Heatherly's team also knew that AI adoption inside Arena varied widely by person. Some teammates already used AI daily in their personal lives, while others have not knowingly embraced AI, a spread that's common at companies with thousands of employees across many functions.

"The critical components of AI are how good the data is, the quality of the data for the purposes of the agent, the constraints you build on the agent, and then how well users interact with your agent." — Heatherly Bucher

How they built it: starting narrow and expanding under governance

Arena's team started small on purpose, treating the agent's scope, and PTC's approval process, as the first design decision rather than an afterthought.

01 Get organizational buy-in before building

‍Before writing a single prompt, Heatherly's team confirmed what PTC's compliance and AI governance functions would allow. That step is easy to skip at a smaller company. At PTC's scale, it's the difference between an agent that survives an audit and one that doesn't.

02 Define a narrow job for the agent

‍The team resisted building an agent that tried to do everything. The first version had access to a handful of public web pages and one SharePoint folder, with no MCP connections at all. The goal was specific: be the authoritative source on Arena partnerships, not a general-purpose assistant for the business.

03 Restrict general web access

‍The agent cannot browse the open web. Heatherly's team made that choice deliberately, so the agent would never pull in blog posts, forum threads, or competitive chatter that could contradict Arena's own partner data, a control that matters more the more public-facing content a large company has.

04 Add the Crossbeam MCP once the foundation held

‍As the agent proved useful, the team added the Crossbeam MCP connection, along with Gong and Crayon. That gave the agent access to account mapping data alongside call context and competitive intelligence, all inside the same governed conversation.

"Our partnership AI agent does have access to Crossbeam through the MCP connection. It allows us to bring that Crossbeam data into a multi-data source analysis for these customer-facing teams and for our partnership team as well." — Heatherly Bucher

05 Use the agent for partnership evaluation, with a person still in the loop

‍More recently, the partnerships team began using the agent to evaluate a potential channel partner, an Enterprise Resource Planning (ERP) vendor adjacent to Arena's own space. The agent combined Crossbeam overlap counts from an early, NDA-only relationship with context on existing partnerships that could strengthen that overlap, plus Arena's own partnership criteria stored in SharePoint. It returned a recommendation, not just a data pull, and Heatherly's team kept the final decision with a person, consistent with how PTC expects AI to be used across the business.

What happened: a governed, trusted source at scale

Nine months in, most of Arena's customer-facing teams now default to the partnership agent instead of asking in Slack or searching SharePoint on their own. Heatherly credits that shift to consistency: the agent gives the same answer every time, drawn from the same set of vetted sources, rather than whoever happens to be online, a result that matters more as headcount grows and answers get harder to standardize.

"One of the things we really appreciated about Crossbeam as we've explored its use is that we've been using Crossbeam more and more in our vetting of new potential partnerships. One of the questions you often need to answer in looking at new partners is: what is our overlap in that ideal customer profile? What's our overlap in customers? What's our overlap in engaged opportunities and prospects, those greenfield opportunities?" — Heatherly Bucher

The partnerships team has started leaning on the agent earlier in its own process, using it to pull together an initial view of a potential partner before running the manual analysis that used to come first. Heatherly noted the agent's read on the ERP channel partner opportunity was faster than running a Crossbeam report and separately checking it against team strategy by hand, even though the final decision still came from her team, not the agent, the kind of human-in-the-loop pattern most enterprise AI governance functions require.

A final thought

If you're weighing whether to connect an MCP to your own partnership workflow at a large, multi-business organization, start by confirming what your own governance function will allow, then build narrow before you build wide. Heatherly's advice is to automate the task your team would be relieved to stop doing by hand, prove it inside one business unit, and let other parts of the company decide if it's worth adopting too.

Treat the rollout as ongoing rather than a single launch. Arena's team still checks in when the agent misses, and they review and maintain the tools themselves, the same discipline that governance-minded teams apply to any system of record.

Ready to see what a governed, enterprise-grade Ecosystem Intelligence deployment could look like for your organization? Book an enterprise strategy call.

Frequently asked questions

What is the Crossbeam MCP?

‍The Crossbeam MCP is a way to connect Crossbeam's partner overlap data directly into AI tools like Microsoft Copilot Studio, so an agent can answer questions using live account mapping data instead of a static export.

How does this fit into a larger organization's AI governance process?

‍At PTC, no team builds an AI agent without confirming what the company's AI governance committee and compliance function allow first. Arena's agent was scoped narrowly and restricted from open web access specifically to meet that bar.

Does Arena's agent make partnership decisions on its own?

‍No. Heatherly's team uses the agent to speed up research and surface a recommendation, but a person still makes the final call on any partnership decision, consistent with how PTC expects AI to be used.

Is this specific to Arena, or could other parts of PTC use it too?

‍Arena is one of several PTC businesses exploring how Crossbeam and AI agents fit into their own partnership workflows. The pattern Heatherly's team built, narrow scope, restricted sources, and a person in the loop, is designed to be repeatable across a multi-business organization.

‍