Your organization has probably rolled out a dozen AI tools in the last 18 months. Maybe more. But here's the honest question: do you actually know which ones are working?
That's the problem Pendo tackled head-on in a recent webinar — and the answer involves more than a spreadsheet.
The Visibility Problem Is Worse Than You Think
Jeremy Smith, Pendo's Director of IT and Applied AI Engineering, put it plainly. Before using their own tool internally, his team was "flying blind" — too many apps, no real visibility into usage, and a spend he couldn't defend to the CFO.
The wake-up call? Command Center revealed 59 apps his team didn't even know existed. One of them had 207 users, no contract, and $112,000 in exposure. Nobody approved it. Nobody was tracking it.
That's not unusual. That's standard.
What "Measurable ROI" Actually Means
Kobi Stok, SVP of Product, framed the core challenge simply: organizations are at an inflection point where they need to decide which AI tools to keep and which to cut. There's no universal answer — some teams are moving off Claude for OpenAI, others are doing the opposite. What matters is having the data to make that call.
Pendo's Command Center is built around three questions:
- What are your employees actually using?
- What is it costing you?
- What action do you take with that information?
The tool deploys via a browser extension pushed through MDM. Once it's running, it maps employee usage against a catalog of roughly 5,000 business applications — including AI agents — and surfaces utilization, cost estimates, and engagement patterns. According to the team, one customer went from zero visibility to 420 discovered apps within six hours of deployment.
Adoption Metrics That Actually Matter
One of the more useful moments in the session came when the question of bad AI adoption metrics came up. Micky Sapir, Senior Director of Product Management, pointed out that raw engagement scores tell you almost nothing on their own. What matters more: are users getting stuck? Are they repeating the same prompts in different ways — a signal of frustration? Are they abandoning the tool mid-task?
Pendo's agent analytics layer captures those patterns — frustration clicks, repeated queries, drop-offs — and connects them back to cost. If 100 people have access to a tool but only 50 are using it as intended, the ROI calculation changes significantly.
Stok added a useful reframe: there's no inherently bad AI tool. The question is whether a given tool fits the workflow for which it was deployed. Adoption is a proxy metric — not perfect, but a reasonable signal for whether the investment is creating value.
What IT Teams Need to Do Right Now
Smith's advice for teams getting started was practical: don't rush the configuration. Specifically, set up the visitor ID before touching anything else. Bad metadata early means months of unreliable data. And communicate ahead of the rollout — employees will see that their usage is being tracked, and getting ahead of that message matters.
Command Center is currently free through the end of 2025. Pendo is building toward native integrations with contract management platforms such as Workday and Zip, which will significantly improve cost accuracy. Upcoming features include the ability to block unapproved shadow IT apps and push Slack or Teams notifications to employees whose licenses are going unused.
The bigger picture: AI spend decisions are getting harder to defer. The organizations that have instrumented their AI stack — who's using what, how much it costs, and whether it's working — are going to make better calls heading into 2026.
The ones that haven't are still flying blind.