Companies are racing to implement AI. It’s the gold rush of our time — the chance to strike it rich in productivity, efficiency, and growth. And AI spending across industries matches the market’s enthusiasm. Of the 800 organizations Coastal surveyed for our 2026 AI Operations Report, 84% said AI is making them more competitive, and 74% are putting more money into it this year.
However, the results are lagging. Only 20% strongly agree that AI has delivered measurable business value, and 46% say it has not met expectations.
Leaders believe in AI, but the value is still slipping away somewhere after deployment. The technology usually works as intended; it’s the operating model around it that gets left unmanaged. As Salesforce Agentforce brings autonomous agents into the enterprise mainstream, solving this operational gap is becoming the next major hurdle.
Running an AI Agent Is Ongoing Work
Despite the frenzy around AI, it’s still early, and everyone’s focused on getting agents live. But what happens post-launch, after everyone moves on to new projects? When costs spike, accuracy drifts, and nobody is monitoring or measuring impact?
Every AI agent’s scale, value, and ROI hinge on these questions: “Who’s responsible for keeping it running while optimizing its performance and delivering ongoing value?”
An agent works in environments and conditions that keep evolving. The closest analogy is a new hire. An Agentforce agent runs against the instructions, data grounding, and boundaries it was given. When it underperforms, someone has to notice, diagnose why, and refine what it was working from.
Companies that treat that refinement as critical and routine work end up with agents that get better over time. Companies that treat it as an exception end up with agents that degrade until someone turns them off.
Four Things That Break After Launching AI
Four patterns recur across the 800 organizations in Coastal’s AI survey. Each one shows up after the agent is live, as the demands of running it outpace the teams responsible.
1. Data Quality Degrades After Launch
70% of organizations name data access or quality as a stage where AI programs stall or fail. After deployment, 73% report that data accuracy or availability problems limited what their AI could deliver.
For Agentforce users, this highlights the role of Data 360. An agent is only as intelligent as the data grounding it. The dependency grows as use cases scale, because the CRM data, custom objects, and external data streams feeding the agent keep changing underneath it.
“Getting the data into a state that we could actually use took a lot more work than we thought it would. This is because we had to do a lot of cleanup and sorting.”
– CFO, Higher Education, Coastal AI Survey
2. AI Adoption Stalls Without Trust and Fit
77% of organizations say their employees are eager to work with AI, yet 73% still hit adoption problems once a capability is live. 50% say users didn’t trust the outputs enough to act on them, and 46% say the AI didn’t fit how people work.
If you deploy an agent but your teams ignore its recommendations inside Salesforce Agentforce Service or Sales, your ROI is zero.
Big Brothers Big Sisters of America got ahead of this by building the AI into the tool their match specialists already used. One button in a familiar workflow now returns a ranked list of mentor matches with plain-language reasons for each. Specialist review meetings that ran an hour dropped to thirty minutes or less.
“If you just drop a solution on people, you lose trust.”
– Travis Gibson, CTO, Big Brothers Big Sisters of America
3. Unclear Business Problems Undercut Agentforce ROI
Just 26% of organizations begin an AI initiative from a clearly defined business problem. 43% start with the technology and then look for a use case to fit it.
Because Agentforce provides a framework for rapid build and deployment, it’s tempting to build simply because the capability exists. The cost stays invisible at launch and surfaces later: 75% of business problem-first organizations are confident they can scale what they built, compared with 63% of everyone else.
“Involving business stakeholders sooner could have supported better coordination and more effective planning.”
– CTO, High Tech, Coastal AI Survey
4. No Clear Owner, No Accountability
Only one in six organizations puts a dedicated AI team in charge of its AI work. The rest fold AI into IT or business leadership, on top of the jobs those teams already have. Who manages Agentforce daily? Is it the Salesforce Admin? RevOps? IT?
Where a dedicated team owns it, measurable business value is far more common: 87%, compared with 67% for IT-led programs.
“The worst mistake was not assigning clear responsibility. Everyone thought IT was in charge, while IT thought operations was in charge, which caused things to stagnate.”
– CTO, Pharmaceuticals, Coastal AI Survey
What Running Agentforce Well Looks Like
All four are operational problems that surface after the build and get worse the longer they go unmanaged. They’re what happens when an organization stands up agents faster than it builds the capacity to run them.
The organizations getting Agentforce ROI build that capacity on purpose, in sequence. They define the business outcome before configuring Agentforce actions, so there’s something concrete to measure against. Governance comes before the next pilot. They analyze the data architecture and map the workflow while the solution is still being designed, so quality and fit are settled early. And they assign the work to an owner to keep managing it after launch.
Each step sets up the next, and together they close off the four ways value leaks out: an undefined problem, weak data, poor fit, and no one accountable for results.
Next Steps to Achieve Agentforce ROI
Operating agents is a standing function, and Coastal built Waves for Agentic Ops, our managed service, to do just that: launch high-value agents and stay on to run them. We monitor performance, manage the data, maintain governance, and keep each agent tied to a business outcome as it scales.
The companies turning AI spend into the productivity and growth everyone is chasing treat running their agents as a job. The agents still delivering a year from now will be the ones that had an owner from day one.
Knowing which AI use cases are worth backing (and whether your Salesforce foundation can support them) is critical. Coastal’s AI Pathfinder gives you a clear read on your IT architecture, data quality, and technical debt — and a prioritized AI roadmap of which use cases to build first, how to sequence the work, and where the biggest payoff is.
