A customer notices something wrong with their account. They think about contacting support, then decide it isn’t worth the trouble. The problem is small, and they don’t feel like explaining it, waiting for a response, and potentially explaining it again.
No ticket gets created. No agent spends time on it. The support dashboard looks exactly the same.
But the customer still has a problem.
Now imagine that customer can get help immediately. They explain what happened in their own words, receive an answer specific to their situation, and get the issue resolved in a few seconds. The next time something comes up, they’re more likely to ask for help.
We’ve improved the experience. We may also have increased demand for support.
That possibility complicates one of the most common business cases for AI: fewer contacts, fewer agents, and lower costs. Those can be valuable outcomes. But if they define the entire ambition, we risk undervaluing what better support can do.
The demand our dashboards don’t capture
Contact volume measures how many people sought help. It doesn’t measure how many needed it.
Customers make a judgment about whether contacting us will be worth the effort. Their expectations come from our previous interactions and their experiences with other companies. Sometimes they tolerate a problem. Sometimes they use the product less. Sometimes they leave.
AI creates an opportunity to change that calculation. As help becomes easier to access and more useful, customers may bring us problems they previously would have abandoned.
This is a hypothesis leaders should test, rather than a benefit to assume. More contacts can also signal product defects, confusing communication, or failed resolutions. But when we evaluate automation primarily through contact reduction, we can miss an important distinction: preventing a problem and discouraging someone from seeking help can produce the same apparent result.
Moving beyond answering questions
At Fetch, our AI and automation journey has focused on making support better through improved understanding, more accurate and consistent answers, and faster resolution.
Generative AI was an important turning point because it improved the system’s ability to understand what customers were asking. People don’t describe their problems using our internal categories. They tell us what happened, often with incomplete information or several concerns mixed together.
Understanding the question is the starting point. The opportunity grows as we connect that understanding to customer context and the ability to act.
Explaining how a process works is useful. Understanding what happened in a particular customer’s account is more useful. Taking an appropriate action to resolve the issue is better still. Identifying a problem and reaching out before the customer asks can remove the need to initiate contact altogether.
Each step requires investment in engineering, reliable information, clear policies, and quality controls. Each also expands what support can accomplish.
Three things leaders can do
- Understand the demand before deciding what to automate.
Look at why customers need help and what a successful outcome would be. Some contacts should disappear because we fix the underlying product or communication. Others represent legitimate service needs we should make easier to resolve.
A repeat contact about an unresolved issue and a first contact from someone who previously would have given up are very different events. We need enough understanding of customer intent and interaction history to distinguish them.
- Measure successful resolution alongside efficiency.
Deflection tells us whether an interaction reached an agent. By itself, it doesn’t tell us whether the customer received an accurate answer, completed the task, or simply stopped trying.
Review actual interactions. Assess answer accuracy and whether the intended action happened. Look for repeat contacts and unresolved needs. Use customer feedback to understand what the operational metrics leave out.
Also recognize that as automation handles simpler issues, the work reaching people becomes more complex. Expectations for handling time and productivity need to reflect that change.
- Decide where the capacity you create should go.
Efficiency gives leaders choices. We can absorb growth, improve response times, give people more time for difficult cases, or serve needs that were previously too expensive to address.
The cost of resolving an individual issue could fall while total investment in support rises. That can be a good business decision if the additional service creates sufficient value.
Build the evidence for that decision. Test whether better support contributes to continued product use, retention, or completion of an important customer activity. Those outcomes deserve a place in the business case alongside labor savings.
AI may allow us to resolve the same problems at a lower cost. It may also allow us to help customers in ways we couldn’t afford before.
If we make support useful enough that more people want to use it, our scorecards should help us recognize the achievement.
Brian leads a 200-person global organization across User Support and Platform Integrity at Fetch. His experience with AI spans 15 years across analytics, machine learning, AI hardware, and now GenAI-powered customer support, where he helped Fetch move from near-zero self-service to roughly 70% AI-enabled self-service resolution while increasing CSAT. His work focuses on building teams and operating systems that use AI to make things better, not merely cheaper and faster.
