On June 16, we’re hosting our summer release event where we’ll show exactly how we’re helping teams close this gap—RSVP here.
Every company has a version of this meeting. Someone (likely in finance) pulls up the slide with the AI tools budget and asks the question nobody has a clean answer to: what are we actually getting for this?
It’s not a hostile question, but it can feel like one. Enterprise teams have spent the last two years deploying AI tools, automation platforms, and agent workflows at a pace that would have seemed impossible a handful of years ago. And yet when it’s time to report on outcomes, most leaders are working from inference rather than evidence.
Return on AI—AI ROI, ROAI, however your team is referring to it—are the terms of the moment for this problem. But the concept outpaces the practice, because most teams can tell you what they spent but only a fraction can tell you what changed because of it.
The measurement gap is structural, not a reporting failure
When companies struggle to show ROAI, the instinct is to point at the analytics stack, get better dashboards, or run more reports. But the problem usually isn’t the reporting layer. The problem is that the underlying data doesn’t capture what actually happened to produce the reports.
AI tools don’t operate in a vacuum. They’re embedded inside human workflows: a support agent who uses an AI-suggested response, an automation that routes a ticket before a person ever touches it, a copilot that surfaces a knowledge article mid-conversation. The AI acts, but the action lives inside a process that spans multiple systems, multiple people, and often no clean audit trail.
The result is a process you can’t see, measuring an outcome you can’t trace.
What good measurement actually needs
To understand whether an investment in AI is working, you need a clear view of the full process around it — what happens before the AI acts, what happens after, and how outcomes shift depending on whether and how it was used.
For support and CX teams, that maps to questions like:
Did average handle time actually drop after you deployed that agent-assist tool, or did it drop for some ticket types while quietly climbing for others?
Are your top-performing support agents using the AI differently than the rest of the team, and if so, what does that tell you about how you should be training people?
When a ticket escalates despite AI involvement, where in the process did things go sideways?
None of those questions can be answered with just AI output logs. You need a clear view of the workflow the AI is part of, the full path a ticket takes from open to close, which systems it moves through, and where time is actually being lost or saved.
A framework for thinking about ROAI
ROAI measurement works when it connects to outcomes that were already meaningful before AI entered the picture. The mistake most teams make is building new AI-specific metrics that don’t map back to the business numbers leadership is already managing.
A more grounded approach starts with a metric that predates the AI deployment: average handle time, time to resolution, escalation rate, first-contact resolution. Something already on the scorecard, already owned by someone accountable for it. From there, you need a ground-truth view of the process that metric lives in. Not a survey of how agents say they work, and not a sample of tickets. A complete picture of how work actually moves across your systems. Then you measure the delta, what changed when AI was introduced, where the gains are real, and where friction just moved somewhere less visible.
The hard part is the middle step. Ground-truth process visibility is what most teams are missing, and it’s why ROAI conversations so often stall at “we think it’s working.”
Why this is harder than it sounds for support teams
Support and CX operations are some of the most process-dense environments in any enterprise. A single ticket can move through a CRM, a knowledge base, an AI assist tool, a quality review system, and an internal messaging platform, sometimes within a single interaction. Support agents routinely switch between six or more applications in the course of resolving one issue.
Traditional process mining tools weren’t designed for this. They need instrumentation on every application you want to track, they have no view into what happens in the browser, and they can’t follow a workflow that crosses systems the way a modern support agent’s day actually does. So process maps look clean on paper while the actual work looks nothing like them, and without better data, there’s no reliable way to square the two.
Getting to an answer your CFO will believe
A credible AI ROI story doesn’t need perfect data, but traceable data, a clear line from the AI investment to a process change to an outcome metric, with enough visibility into the middle that the story holds up when someone starts asking follow-up questions.
That’s what finance and leadership are really asking for when the budget slide goes up. Not a guarantee, not a model built on layered assumptions. A before-and-after view of how work gets done, with evidence that the investment moved something that was already worth measuring.
The teams getting there aren’t necessarily the ones with the most sophisticated AI deployments. They tend to be the ones that solved for process visibility upfront rather than as an afterthought.
If you want to see what we’re building to help teams get there, we’d love to show you at our June 16 summer release event. RSVP here.
