The “Get Those Agents ROI-ing” Conversations.
by Raphael Steinman
Every leadership team is having some version of this conversation right now. CAC is climbing. NRR is under pressure.
The old GTM playbook; more reps, more tools, more spend; is breaking down. AI agents are the obvious answer.
So the mandate lands: deploy agents, cut cost, show ROI.
And then something uncomfortable happens.
The agent needs to act on operational data. It needs to trace a customer interaction through the process chain to an outcome. It needs to reconcile what marketing spent, what sales touched, what product delivered, and what the customer actually experienced; across systems that were never designed to talk to each other.
The data is fragmented. Unharmonized. Stripped of the process context that makes ROI attributable.
The agent is not the bottleneck. What the agent operates on is.
This is the pattern playing out across every company trying to move from AI demos to AI deployment. The models are capable.
The agents are ready. But the foundation underneath; the operational data those agents need to reason about; was built for humans reading dashboards, not for machines making decisions.
A brilliant agent pointed at unresolved data does not produce ROI. It produces confident answers that no one can verify.
The companies getting real returns are not the ones with the best agents. They are the ones that resolved the data first. They built a foundation where every activity knows where it sits in a process, every entity is harmonized across systems, and every metric traces back to its source. The agents work because the material they reason about was made trustworthy before the agents were deployed.
The uncomfortable truth in the “get those agents ROI-ing” conversation is that the ROI problem is not an agent problem. It is a data architecture problem. And until that layer is resolved, every agent deployment is measuring shadows.
The bottleneck was never intelligence. It was producing material worth reasoning with in the first place.