How CX leaders can choose one AI win, assign ownership, and retire the work it replaces
5 min read
Published on November 10, 2026
A contact center leader pulled me aside at a recent CX meetup with a problem she had no clear way to solve. Her executive had given her one instruction: "I need you to do something with AI." He set no outcome, no owner, and no measure of success. She left with expectations to meet and no definition of them.
That sentence captures the bind many CX leaders live in. One pressure says move on AI now or fall behind. The other says their people cannot absorb one more rollout that promised relief and delivered extra work.
AI fatigue in the contact center refers to the exhaustion teams can feel when AI tools arrive without clear outcomes, ownership, or relief from existing work. A practical response is to choose one measurable use case, define who owns it, and keep people involved in oversight and decisions.
FOMO and fatigue grow from the same root. Both can intensify when an organization says yes to AI before it decides what the technology owes the business, who owns it once it goes live, and which work it takes off people's plates. The answer to both is what I call a defensible yes.
A defensible yes commits to one AI use case and settles the hard questions before launch: the business outcome it serves, stated in the executive team's terms; the person who owns it; the specific task it removes from someone's day; and the measure that a skeptical executive and a skeptical agent will both accept. A disciplined leader makes one defensible yes, proves it, and holds off on everything else until it delivers.
Holding off sounds simple until the requests start arriving. Imagine a leader whose team just launched a chatbot for order-status questions. Within a month, her operations director wants to pilot a tool that coaches agents during live calls. She asks him to wait until the chatbot hits its target, since both tools pull from the same content and her team can fix one set of problems at a time. Marketing asks for a dashboard that helps analyze customer sentiment, and she moves it to next quarter because nobody has time to act on what it would show. A vendor pitches a fourth tool, and she tells him her team cannot evaluate it well right now. Each idea gets a reason it waits and a date to revisit it, and her team gets the room to make the first commitment work.
I wrote this piece for the leader who owns the AI decision, and the work involves everyone around that leader. If someone else controls your budget, the filter and the Ownership Test below give you a stronger case to bring upward. If you work on the front line, as a supervisor, or in quality, your workarounds and error reports belong in the one-hour session at the end. Leaders can retire hidden work only after someone shows it to them.
When I ask leaders whether AI met their expectations, many say yes. When I ask what those expectations were, the room goes quiet. That vagueness lands on teams that already run on empty. Service teams carry hidden labor long before any AI tool appears: the team-chat searches for a current policy, the duplicate entry that quality assurance requires, the supervisor who repairs the same broken handoff every shift. I explored that pattern in AI Fatigue Starts Before the AI Arrives, and those tasks become the first candidates to retire in the plan at the end of this piece.
When leaders tell me their teams have AI fatigue, I ask what those teams felt tired of before AI showed up. They describe staffing and process problems nobody solved before AI landed on top of them, mandates so vague that the leader carries the risk of guessing wrong, and a flood of new output with no pause to settle. "Exhausting" is the word I hear most.
A CEO once told me he lets his contact center team deliver its quarterly presentation, then moves the room on to "actual work." For several quarters, that team celebrated quality scores, satisfaction ratings, and containment rates while he watched churn climb and revenue fall. He concluded that the contact center's world had no connection to the business of the business.
Executives need no technical background to see why measurement has to shift. The buckets of value have stayed the same for decades: cost efficiency, customer loyalty, and strategic impact. Traditional contact center metrics counted activity along the way: calls answered, minutes spent, surveys returned. AI now helps handle a growing share of that activity, so counting it tells leaders less every quarter. The questions executives already ask (Did this protect revenue? Did it reduce cost? Did customers stay?) become the measures that matter.
Containment shows the problem clearly. Many contact centers judge chatbots and automated phone menus by their containment rate, the share of customers who finish without reaching a person. Completion offers a sharper measure. It tracks whether the customer got what they came for without contacting you again within a set window, such as seven days.
Picture a customer who asks a chatbot about a double charge. The bot offers three help articles, none of them reverses the charge, and the customer closes the window in frustration. The containment report records a success. Two days later, the same customer calls, waits on hold, explains the problem again, and mentions canceling. The completion measure records a failure, and so does the customer.
The change, from activity metrics to outcome metrics, turns the contact center leader into a translator. Teams using Zoom Contact Center can apply this outcome-first approach to help connect AI initiatives with customer and business goals. The work starts with what matters most to the organization this year and moves backward to what has to happen inside customer interactions to deliver it. For a utility under regulatory scrutiny, that means outage and billing questions that can be resolved without complaints reaching the state commission. For a health system working on patient access, it means referral calls that lead to a booked appointment. An insurer focused on retention cares whether a claims-status call leaves the policyholder confident enough to renew, and a consumer brand such as Cricut uses customer feedback captured by AI to inform product development. Each case illustrates how organizations can connect the interaction to an outcome the executive team already tracks.
Each leader I talk to has watched a clever AI experiment impress in a demo and then prove nothing that mattered to the business. Aim for a defensible win, even when it looks plain, and run each candidate through three questions.
Volatility invites opinion, so settle it with evidence. Pull 20 recent cases and look for three things: customers in distress, such as someone calling after a death in the family or for the third time about one issue; consequences that cross a line you set in advance, such as a large refund or a required compliance disclosure; and moments when the agent needed a supervisor's approval to finish. An order-status check rarely shows any of them, while a fraud claim may involve several. The counts end the debate faster than any meeting.
Predictability depends on what the AI draws from. Many organizations feed the same knowledge base (the library of articles, policies, and approved answers the organization maintains) to two tools: the virtual agent customers talk to directly, such as Zoom Virtual Agent, and the assistant that suggests answers to employees during live conversations, such as Zoom AI Expert Assist. When that knowledge base sits in disarray, both tools may underperform. Audit the source content before you automate, and assign a named owner to keep it current.
The filter also forces honest tradeoffs. One leader at a subscription business told me her company automated part of its cancellation path over her objection; she saw that call as the last real chance to save the relationship, and the economics won. I give leaders in her position the same advice: care deeply, fight passionately, and hold loosely. You will lose some of these arguments, and your credibility on the next decision depends on how well you made the case for this one.
In my work with Metric Sherpa, I keep hearing the same pattern. Often the first sign of an AI mistake is a customer complaint or an agent raising it. Many lack a reliable way to measure the factual accuracy of what their AI tells customers. Agents become a safety net for bot errors, and many tell me AI has created more problems for them to solve.
A weekly sample of a handful of interactions misses many of the interactions AI handles across thousands of conversations, and better monitoring starts with clear ownership. Before you launch, or this week if you already have, answer three questions in writing.
A name on a chart protects no one unless that person follows a routine. Build the routine around four elements, and it can help catch issues before customers see them.
Leadership can buy AI tools faster than an operation can absorb them. New launches land while agents still maintain the old checks and supervisors still coach against last month's instructions. Many agents also carry ethical, environmental, or accuracy concerns about AI, and resistance follows. Treat it the way you would treat resistance to any other change. Start with clarity: explain why the change matters, what leadership believes, and what it means for people's jobs, because silence invites people to fill the blanks with their worst fears. Then let people see the tool in practice and talk openly about what works and what fails. Many frontline employees tell me AI felt inflicted on them, with nobody asking their opinion.
People prefer familiar misery to uncertain hope, especially after years of predictions about AI replacing them. Leaders earn adoption through candid conversations, and they keep talking with the people who do not feel fine about it yet.
Book one hour this week, or propose it to the leader who owns the decision, and invite five people: an operations leader, a supervisor, a frontline agent, your quality lead, and the person who owns your knowledge content. Add someone from finance or the executive who asked you to "do something with AI" if you can.
Capture the answers on one page. That page becomes your Defensible Yes brief, and it gives your executive something concrete to approve, challenge, or fund.
The leader from that CX meetup now has a way to respond. She can walk into her executive's office with one page: the business outcome, the interaction, the owners, the work that disappears, and the ideas the team will set aside for now. That page turns a vague mandate into a decision the business can judge.
FOMO loses its grip when you know exactly what you said yes to and why. Fatigue eases when your people watch that yes deliver before the next one arrives. Make one decision you can defend, prove it, and let the results set the pace for everything that follows.
What is AI fatigue in a contact center?
AI fatigue refers to the strain teams may experience when AI deployments add work or uncertainty without clear outcomes, ownership, and support.
How can contact center leaders reduce AI fatigue?
Leaders can start with one measurable use case, assign an accountable owner, involve frontline employees, and retire the old task when automation takes it on.
How should contact centers measure AI outcomes?
Choose measures tied to the business goal, such as repeat contacts, customer resolution, cost, or retention, and define the measurement window before launch.
What is a defensible yes for AI?
A defensible yes is a decision to pursue one AI use case after defining its business outcome, owner, task, and success measure.
How should teams oversee AI in customer interactions?
Assign people to review output, receive agent reports, monitor accuracy, and decide when to correct or pause a deployment.
About the author:
Justin Robbins leads Metric Sherpa, a research and advisory firm that helps people leaders close the gap between what their organization promises and what employees experience every day. He is also the author of More Than A Motto.
7 min read
7 min read
8 min read
