Years ago, a manufacturer of protective phone cases had a support volume problem:
• Customers were calling about the charging port cover because it wasn’t intuitive, didn’t always seat properly, and, in some cases, failed.
• Calls led to troubleshooting, then to replacement cases and, for the company’s waterproof products, potential phone warranty exposure.
The root cause existed in exactly one dataset in the company: what customers said on the phone. It wasn’t in returns data, wasn’t on warranty forms, and it wasn’t in the ticket taxonomy. Nobody was filing a report that said the cover design was confusing. Instead, customers were describing the problem in their own words to the agents.
Conversation analysis then surfaced the issue; the finding went to Product Design and to Manufacturing, and the product changed.
The company reportedly avoided millions in warranty claims, but the more lasting change was procedural: product teams started coming to the contact center first because they understood that customer conversations contained intelligence the rest of the organization couldn’t always see.
This incident happened years ago, not as a pilot or proof of concept, but as normal work for a company that was simply paying attention.
Removing Barriers to Obtaining Intelligence
I started my career in a contact center supporting customers for a software company before moving into product management, so I knew from both sides what product managers were missing and what support conversations already contained.
The barrier was never technical feasibility as much as the effort, specialization, and imagination required to determine what the data could answer.
As those barriers fall, the question for contact center leaders shifts as well. It’s becoming less about how efficiently the contact center is staffed and more about whether technology investments are delivering better customer and business outcomes.
The pattern also goes back decades.
Roughly 20 years ago, a cable operator used speech analytics not for quality monitoring (QM), but to detect signal degradation in specific service areas and quantify what that degradation was doing to customer satisfaction.
That was network intelligence derived from conversations.
• The case manufacturer I mentioned earlier turned them into product design intelligence.
• More recently, a direct-to-consumer pet supply company traced a cluster of low-star Amazon reviews back to a root cause identified in support calls and changed the product.
In each case, the organization wasn’t simply asking how efficiently the contact center handled an interaction; it was asking what those interactions could tell the rest of the business.
How AI Makes A Difference
What’s actually new isn’t the capability. Organizations have been doing this for years. The value of that intelligence has been demonstrable for a long time.
What’s changed recently is that AI has made it dramatically easier and faster to understand what’s inside the data, while raising general awareness that this understanding is even possible.
Historically, extracting meaningful insight from large volumes of customer conversations required setup time, specialized expertise, and a fairly clear idea of what you were looking for in advance.
AI changes that equation, reducing the specialized skills and time to surface insights. Through faster analysis, organizations gain a broader ability to understand why customers are contacting the business, where recurring issues may be taking shape, and more.
Conversation data also becomes more accessible beyond the contact center. Product, Marketing, Operations, and other teams can use AI to easily interact with conversational data to identify themes and trends in customer feedback that would have only been accessible to specialized teams.
The underlying customer signals were always there; AI is lowering the effort required to find them, connect them, and put them to use across the business, truly democratizing intelligence.
The Contact Center is One of Several Intelligence Sources
The key to delivering those outcomes is ensuring that you are tapping all possible sources of data from which you extract intelligence and actionable insights.
The contact center is one of the richest sources of voice of the customer (VOC) insight, but it isn’t the only source.
Organizations should instead be aggregating contact center interactions with every other stream of customer interaction and feedback, including digital, survey, review, chat, field, and sales data.
This is especially important in large enterprises with multiple contact centers, multiple BPOs, and dozens of customer touchpoints, each holding only a partial view.
What’s changed recently is that AI has made it dramatically easier and faster to understand what’s inside the data…
Every channel holds valuable customer signals, but the insights become even more powerful when those signals are connected across channels and feedback sources. Fragmentation, not scarcity, is the real obstacle.
There is a competitive dimension to this as well. Algorithms are purchasable, and models are licensable, but your customer interaction data can’t be replicated by a competitor. That’s the durable asset.
When an organization genuinely makes this shift, ownership of conversation intelligence begins to migrate beyond the contact center and becomes a capability serving Product, Marketing, Billing, Compliance, and Operations.
If the intelligence still begins and ends with contact center operations, there is considerably more value to unlock, regardless of how much the organization has invested in technology.
Intelligence Earns You The Right To Be Proactive
There is significant focus now on predictive systems that can detect early signs of churn or disruption and initiate outreach automatically, with success increasingly measured by problems avoided rather than problems resolved quickly.
That is an important shift, but it’s worth stating what makes it possible: you cannot prevent what you haven’t diagnosed. Proactive outreach without root-cause intelligence is just more outbound noise, and customers have little patience for interactions that don’t address something they actually need.
The progression moves from reacting to a customer’s issue, to understanding their intent, to predicting what they may need next, and ultimately to preventing the issue from arising in the first place. Each stage requires more understanding of why contact happens, not just how much of it there is.
Inbound volume isn’t simply a workload number; it is often a lagging indicator of upstream breakdowns in product, communication, billing, and policy.
The port cover I discussed at the beginning of this article is the cleanest possible example because the calls themselves weren’t the fundamental problem.
Instead, the calls were evidence of a product problem that happened to surface in the contact center. Understanding that distinction allowed the organization to address the cause instead of simply becoming more efficient at handling the symptoms.
What’s genuinely new is the economics. Near-zero marginal cost per conversation makes net-new engagement viable, whether that’s proactive notification, preemptive explanation, or confirmation at moments of uncertainty.
These aren’t necessarily automated versions of existing calls; they’re conversations that were never affordable before.
But the goal shouldn’t simply be deflecting demand from a human to an automated channel. It should be understanding why the demand exists and, where possible, eliminating it. The highest-value conversation may be the one the customer never needs to have.
The Measurements That Matter
Containment quality, automation accuracy, and reduction of downstream rework are all useful measurements, but downstream rework points toward something much bigger: the value created by the contact center often shows up somewhere other than the contact center.
Go back to the port cover. Those calls were effectively invisible to the metrics the contact center owned because they looked like routine troubleshooting on its dashboards.
The real value surfaced as warranty claims avoided, in another department’s profit-and-loss (P&L) statement, on another team’s budget line, months later. The measurement that mattered wasn’t a contact center measurement at all.
Accuracy raises a similar question: accurate against what standard? An AI agent can execute a designed flow with high fidelity and still be walking customers down a path that was never the best route to resolution.
Accuracy without a ground truth derived from real conversations measures compliance with our assumptions, not necessarily effectiveness. Containment has the same limitation.
The risk isn’t the metric’s existence, but what it counts. Containment records where the interaction stayed; it says nothing on its own about whether the customer’s need was met.
I would put three questions at the center of measurement.
1. Was the need actually resolved?
2. Did it resurface?
3. How much effort did it take?
Resolution should be verified through conversation intelligence or contextual post-interaction follow-up rather than inferred from the absence of a transfer.
Organizations should also know whether the same issue, for the same customer, resulted in another contact within the next 24 or 48 hours, including through another channel, which requires journey-level visibility.
Finally, repeated explanations, escalation requests, and channel switching all provide important measures of effort, which can help distinguish a resolved contact from a survived one.
Yet that avoided interaction may represent more value than making the original contact cheaper or faster to handle.
We’ve run the experiment of optimizing around easy measurements before. Average handle time (AHT) was easy to measure and easy to report.
But when organizations over-optimized around it, agents could be pushed to rush customers, avoid research, and dodge escalations.
Easy-to-measure has beaten meaningful-to-measure before, with the costs eventually showing up downstream as repeat contacts, increased effort, and churn.
AHT is a good example: optimizing for shorter calls could make performance look better on paper while driving repeat contacts, greater customer effort, and churn.
Escalation was never inherently failure; it’s adaptation, and metrics should reward it when it’s the right call.
The same caution should apply to AI, because if we optimize primarily for keeping customers inside automated experiences, systems may become very good at containment without becoming equally good at resolution.
Measure the Demand That Never Occurred
There is another measurement that is harder to see but potentially much more valuable: demand that never occurred. Reduction in downstream rework points directly toward it.
When customer intelligence helps an organization identify a confusing product feature, fix an unclear policy, improve a communication, or intervene before a known problem occurs, a future interaction may disappear entirely.
Preventable demand is difficult to measure because nobody gets credit for the call that didn’t happen unless someone deliberately builds the baseline to claim it.
Yet that avoided interaction may represent more value than making the original contact cheaper or faster to handle. To address this, establish a baseline for repeat or avoidable contacts, link those interactions to their root causes, and track whether volumes decline after process changes.
That is why the measurement challenge matters so much today and especially going forward.
The difficulty isn’t new technology arriving; it’s the business change that several years of technology have forced, and the reckoning is with what we choose to count.
The port cover insight was available years ago to any organization willing to look at what customers were already telling them. What’s changed is that looking has become far easier and far more widely understood as possible, which means the excuse for not measuring what matters is disappearing.
Get the measurement right, and the contact center becomes an intelligence source capable of helping the enterprise become more proactive, prevent problems, and generate value well beyond customer service.
Get it wrong, and we risk spending the back half of the decade unwinding a new generation of metrics that are optimized for the spreadsheet rather than the customer.
This article appears in October 2026.
