Originally published on LinkedIn, John’s Corner is where Fortimize CEO John Hamon shares the perspectives, challenges, and hard truths he sees from the front lines. View original publication.
In October, Dana signed the contract. A 3-year Agentforce 1 FSC deal. Her CEO approved it on the promise of a 30% reduction in member servicing costs by year-end.
Dana is the CIO at a $4B credit union.
By April, servicing costs were flat. Her agent was executing 5,000 actions a week against a member data model last cleaned up before COVID.
The board hadn’t asked about the 30% yet. They will.
What She Thought She Was Buying
Dana thought she was buying a layer. An agent that would sit on top of her existing Salesforce instance, work the data and processes already there, and convert routine member interactions into automated ones.
2,000 cases a week handled without human touch. Service team capacity freed for higher-value conversations. Roughly 5 FTE of run-rate savings by next year.
The instance underneath was not perfect. She knew that. It was the same instance her board had been told was “good enough” at the 2019 launch and “a strong foundation” in every annual roadmap review since.
The math she signed against assumed “good enough” was good enough for an agent to stand on. Most boards in financial services right now are signing against the same assumption.
What Actually Changed
Every layer Salesforce sold for the previous decade sat on top of the substrate. Lightning components rendered against bad data. Reports rolled up bad data. Dashboards visualized bad data. None of them acted on it.
The dysfunction was static. Humans were still the agents executing on member relationships, and humans noticed when the data was wrong. They worked around it. They emailed each other. They picked up the phone.
The instance was a passive cost. It slowed people down, but people stayed in the loop.
Agents do not sit on the substrate. They act on it.
They route cases. They send messages. They update records. They create tasks. They escalate to humans. At machine speed. At machine confidence. With a complete audit trail.
Whatever the substrate tells them, they execute. Whatever the substrate is wrong about, they execute — at the rate of 5,000 actions per week.
The Math Has Flipped
In the old frame, Dana’s instance was carrying technical debt. Carrying debt had a cost. Things moved slower than they should, certain reports could not be trusted, the team had internal workarounds.
The debt was real, but it sat on the balance sheet as a someday item. Static.
In the new frame, the instance carries debt and has an agent acting on it. Each dollar spent on Agentforce purchases more action against the bad substrate. The agent is not paying down the debt. The agent is leveraging it.
The board approves Agentforce to drive ROI on the platform. The agent drives the platform faster. The platform is wrong. The faster it goes, the more wrong it produces.
This is not low ROI. Low ROI is money wasted. This is negative ROI: money spent to make the situation worse than it was before the money was spent.
Most Agentforce deployments in financial services this year are running at negative ROI right now. The dashboards do not show it because the dashboards report on agent activity, not on the downstream consequences of agent activity.
What To Do Right Now
If you are signing an Agentforce deal in the next 6 months, the question to ask is not whether the agent works.
The vendor’s demos work. The agent works.
The question is whether the substrate underneath your agent has been touched, audited, or rebuilt within the last 12 months by someone whose job depended on the answer being honest.
If the answer is no, the order of operations is wrong. The substrate cleanup is not a parallel workstream. It is not a phase 2.
It is the prerequisite.
The agent deployed against an unaudited substrate will run beautifully against the wrong configuration, at machine speed, until the consequences arrive at the board level 6 months out and someone has to explain them.
Dana will explain hers in a few weeks. She is now spending the rest of this year doing the substrate work she should have done before the agent went live. The agent is paused. The 5,000 weekly actions have stopped. The audit trail has gone quiet. The 30% is a long way away.
The good news, if there is any, is that she now knows what she was actually buying.
