Enterprise AI budgets are being approved faster than companies can measure their returns. Too often, successful pilots fail to translate into measurable business results. Boards approve the spend, vendors deliver the capability, pilots impress everyone in the room, and then the quarterly numbers look exactly the same as they did before. Tom Gersic, founder of YouEx.ai, has spent 26 years in enterprise software, including 12 leading product adoption at Salesforce. He puts the failure in blunt commercial terms: “AI only creates value when it changes how work gets done. If you deploy the technology and people keep working the same way, you’ve just added cost.” That sentence should unsettle any executive currently reporting AI progress in terms of deployment milestones. Added cost is not a neutral outcome. Those costs can accumulate across the business when the technology question gets solved and the work question never gets asked.
The Work Around The AI Is The Problem
The common diagnosis is integration. Companies conclude the AI has not been connected to enough of their data, so they connect more of it, and the needle still refuses to move. Gersic argues the plumbing is necessary but not where the failure sits. “The breakdown happens when people still have to do all the work around the AI,” he says. “They’re moving information between tools, checking for updates, and remembering to trigger the next step. The AI can do more, but their day hasn’t gotten much easier.” That last clause is the whole test. An employee whose day has not changed is an employee who will revert, and reversion is invisible on a dashboard that only counts licenses issued.
What makes this expensive is that the surrounding work tends to be the low-value work. In sales, Gersic points to researching the prospect, preparing outreach, updating the customer relationship management system, and coordinating follow-up, tasks that should move as one connected workflow – rather than as a sequence of manual handoffs a human has to orchestrate. Done properly, the seller is responsible for understanding the customer, exercising judgment, and building the relationship. Building trust. His own measure of success is practical: “How much work have we taken off someone’s plate? And what does that allow them to do better?” Two questions, both answerable, and neither of them is about the AI model.
Pilots Flatter The Technology And Hide The Organization
Pilots have become easier to build, but also easier to misread. “Demos and pilots in the AI age can be deceiving,” Gersic says. “They’re easier to build than ever before, and people get impressed because the underlying technology is impressive. But scaling that into something an organization uses every day to produce value takes a lot more work.” One limitation is who participates. A pilot is typically staffed by motivated early adopters who want it to succeed, which means the result measures enthusiasm as much as capability. The business also needs to serve employees who have less time or inclination to experiment.
Scaling to that population means the workflow has to fit how people already work, remove tasks rather than add them, and make the handoff point obvious when a human needs to step in. Gersic adds training, ownership, and a feedback loop to the list, elements that may be informal during a pilot but need to be explicit and repeatable in production. His sales example shows how far short a single capability falls: qualifying a lead is one step, but the lead still has to reach the right seller, the customer’s context has to carry forward, and follow-up has to happen. That connected workflow is also the approach behind Gersic’s company, YouEx.ai, which brings buyer conversations, lead research, and follow-up together to give sellers more time with customers.
That starts with a clear definition of success. “Define the north star before you build, and evaluate against it as you go,” he says. “It’s easy to get excited about what the AI can do. The more difficult question to answer is whether the business is performing better because of it.” Whether the goal is more selling time, faster follow-up, or higher conversion, defining it upfront gives teams a consistent way to judge progress and decide what to change.
Governance As Speed, Not Brake
Agentic AI sharpens all of this because the tools now act on a company’s behalf, and fewer than one-in-five companies have a governance model for autonomous agents. Gersic rejects the framing that governance slows things down. “Governance gives teams the confidence to move quickly,” he says, describing adoption that works from the bottom up: people discover how AI helps them and share what they learn with colleagues, while clear rules from the top set the boundaries. From executive sessions he led on GenAI at the University of Oxford Saïd Business School, two themes recurred: governance and internal champions. Both, not either. What teams need is clarity on what data they can use, what agents can do independently, and where human approval is required.
The warning attached to that is sharper than most governance conversations allow. “People are going to use these tools whether you provide them or not,” Gersic says. “If every new use requires months of review, the process loses touch with what’s happening.” Slow governance can push AI use into tools and workflows the company cannot monitor. The same logic drives what he thinks revenue leaders are underweighting over the next 12 months: selling time is disappearing into tool experimentation. “It’s easy to spend an entire day working on how you sell without talking to a customer,” he says, and he includes himself in the temptation. Ten sellers building ten workflows means ten people solving the same problem and a business supporting ten processes. The leadership job is to give people room to experiment, then evaluate what works, make it reliable, and spread it. “The point is to get people back to selling.”
Follow Tom Gersic on LinkedIn for more insights on AI adoption, workflow design, and revenue operations, or explore YouEx.ai to see those principles applied to B2B sales.
