What Is Conversation Intelligence? How Contact Centers Turn Every Call Into Data
Conversation intelligence turns every customer call, chat and AI agent interaction into data. Here’s how United Airlines, CVS Health and others use it
Make CX Current News one of your go-to sources on Google
Conversation intelligence is AI that analyzes every customer conversation, including calls, chats, emails and AI agent interactions. It identifies why customers make contact and which agent behaviors lead to good outcomes, and it links each conversation to results like resolution, sales and retention. Contact centers use it in place of sampled quality reviews and low-response surveys, and more and more to evaluate human and AI agents against the same standard.
Why United Airlines' customers kept calling
United Airlines sells a fare type that doesn't allow travel changes unless the customer pays to upgrade to a different fare. Customers could do that upgrade themselves online, yet many still called the contact center to have an agent handle it.
For years, United had limited ways to find out why. According to a session recap by Metrigy CEO Robin Gareiss, the airline's view of call drivers moved from paper tallies kept by agents, to customer surveys, to a keyword tool its team found inflexible and slow. None of those methods showed why customers were calling about this fare.
The calls themselves held the reason. Customers were asked to pay for the upgrade before the self-service flow told them whether the change they wanted was possible, so they phoned to check first. United changed the flow so customers could see their options before paying. Live interactions for that contact type fell 50% immediately, which the airline said saves millions of dollars a year.
United found the cause by analyzing what customers said across all of those calls. The airline now has a dedicated Insights to Action team that uses conversation intelligence products from Cresta, which publishes CX Current.
What conversation intelligence is
Conversation intelligence is AI that analyzes customer conversations at full scale, across voice, chat, email and messaging, and increasingly across conversations handled by AI agents. It identifies why customers make contact, how they feel, what agents said and did, and how each conversation ended. It then links those details to business results such as resolution, sales, retention and satisfaction.
Contact centers have had other ways to study conversations. Call recording stores them, but someone still has to listen. Speech analytics transcribes calls and searches for keywords, so it only finds what someone thought to search for. Quality assurance teams score a small random sample by hand, and surveys hear only from customers who choose to respond.
Cresta has estimated that manual QA rarely covers more than 2% to 3% of conversations in a large contact center. Before CVS Health adopted AI scoring, its customer surveys had a 5% response rate, and it took weeks to collect enough responses to analyze.
In Metrigy's AI for Business Success 2025-26 study of 1,104 companies, more than 90% of respondents rated conversation analytics data as the most valuable data in the company or among the most valuable. Only 49.8% were using conversation analytics.
Forrester treats conversation intelligence for contact centers as its own software category. In The Forrester Wave: Conversation Intelligence Solutions for Contact Centers, Q2 2025, the firm evaluated the 10 providers it considered most significant and named Cresta a Leader.
How conversation intelligence works
Most conversation intelligence systems work in four steps.
Capture every conversation. Calls, chats, emails and messages are transcribed and indexed in one place, and sensitive information such as payment details is redacted.
Understand what happened. Language models identify contact reasons, sentiment and specific agent behaviors, such as whether an agent confirmed the problem before offering a fix. Because the models work from meaning, they recognize the same issue however a customer phrases it.
Connect conversations to outcomes. Each conversation is tied to a result, such as a sale, a resolved issue, a saved account, a transfer or a predicted satisfaction score. That link shows which contact reasons and behaviors affect the metrics a business reports on.
Act on the findings. Insights feed quality scorecards, coaching plans, real-time guidance for agents, fixes to digital journeys and decisions about what to automate. The effect of each change shows up in the next round of conversations.
Gareiss has written that many companies collect insights and customer feedback and then don't act on them. United assigned a team to do that work. Before proposing a change, the team estimates the return, and after the change it documents the savings. Rachel Walker, United's managing director of customer support and advocacy, described the approach as "real data, real volumes, real cost savings."
How contact centers use conversation intelligence
Contact centers use conversation intelligence to understand why customers call, measure quality, coach agents and fix the processes that create avoidable contacts.
Finding out why customers make contact
United has other examples. Customers could see the upgrade waitlist online about 72 hours before a flight, and conversation data showed many calling earlier to ask an agent where they stood. United now shows the waitlist 30 days ahead, so agents field fewer of those calls.
During a 2025 air travel disruption, United expected safety to be its customers' main concern. Its conversation data showed customers were more worried about rebooking, and the airline used that to give customers and agents the information they needed.
Scoring every conversation instead of a sample
Snap Finance, a consumer lender, moved from random sampling to automated QA on all of its calls and now checks every call for quality and compliance. Brinks Home, a home security provider, cut its quality management costs by 50% after automating scoring and coaching plans.
At Oportun, which moved from sample-based QA to monitoring every interaction, supervisors stopped arguing over scores from a handful of calls and began coaching from patterns across all of an agent's work. Chief Operating Officer Ryan Helwig said the team left a "gotcha" approach behind and now spends more coaching time on what agents do well.
Measuring satisfaction without waiting for surveys
CVS Health now uses predicted customer satisfaction scores on all of its calls instead of relying on surveys with a 5% response rate. Srikant Narasimhan, its vice president of enterprise customer experience and insights, has said that this lets the company understand each agent's performance from the customer's point of view. When sentiment drops, leaders can see it quickly and respond. With surveys alone, gathering enough data took weeks.
Linking agent behaviors to results
Brinks Home used conversation insights to learn why customers were calling and which agent behaviors mattered most to them. Transfers fell from 30% of calls to 8%, and net promoter score rose 30 points. At Oportun, Helwig has said that coaching built on full conversation data began closing the gap between top-quartile sales agents and the rest of the floor, and that the improvement showed up in the company's financial results.
Where conversation intelligence is heading
Early conversation intelligence products mostly produced reports on past performance. Newer ones are used throughout the working day, and as AI agents handle more customer conversations, they are also being used to oversee automated work.
Questions instead of reports
Leaders can now ask questions of their conversation data without waiting for an analyst to build a report. Cresta's AI Analyst, which United uses, answers plain-language questions, such as why a contact type spiked this week, and cites the conversations behind each answer. Cresta describes the latest version as an agent that plans and carries out its own analysis and drafts recommendations.
Catching what nobody is tracking
Traditional analytics only tracks categories someone set up in advance. Cresta says its Insights product now also flags phrases and events that aren't in a company's existing categories, such as outages, regional disruptions or a new product defect, so teams can respond early.
One standard for human and AI agents
Contact centers running both human and AI agents need a way to judge both. Conversation intelligence can apply the same quality framework to every conversation, whoever handled it, so teams can compare performance by contact type, find where an AI agent is struggling and decide when a person should step in. CX Current has covered why containment and FCR fall short for AI agents and who should own them, and both questions depend on this kind of conversation data.
Cresta has also described a new role, the AI supervisor. When it launched its Agent Operations Center, the company said AI supervisors are human experts who monitor AI agent conversations and step in to guide or take over when needed. To do that, they need to see which conversations are going wrong while they are still in progress.
Conversation data also helps teams choose what to automate. It shows which contact types are routine enough for an AI agent and which need human judgment. United has said it plans to use Cresta's automation discovery tools to meet growing demand with a team of the same size.
Customer data for the whole company
Some companies now share conversation data well beyond the contact center. In Metrigy's research, 83.7% of IT and CX leaders said insights data should be included in companywide dashboards. United's Insights to Action team shares findings with the airline's digital, communications and automation teams and tracks the savings from each change.
How to evaluate a conversation intelligence platform
Many analytics products are now marketed as conversation intelligence. These checks test what a system can do with a company's own conversations.
Check the coverage. The platform should analyze every conversation on every channel, including those handled by AI agents.
Test how it understands conversations. Give it contact reasons and behaviors described in plain language and see whether it finds them without a keyword list someone has to maintain.
Ask how it connects conversations to outcomes. Insights should be ranked by their effect on resolution, conversion, retention or predicted satisfaction, as well as by volume.
Look for evidence behind every answer. When the system reports a trend or answers a question, a reviewer should be able to open the conversations that support it.
Follow an insight to an action. Findings should flow into quality scorecards, coaching and real-time agent guidance within the same system.
Confirm it fits the existing stack. Most enterprises need it to work with their current telephony, CRM and contact center platform. Brinks Home deployed across in-house and outsourced teams, including legacy on-premises phone systems, without replacing them.
Review the controls. Ask how the platform handles redaction, access permissions and the accuracy of its generative AI features, and how its automated scores are calibrated against human reviewers.
Forrester's 2025 evaluation of the category used similar criteria. Cresta received the highest possible scores in 16 of them, including insight discovery, real-time guidance, outcome analysis, GenAI safety and controls, and insights-to-action.
What this means for CX leaders
Contact center leaders have long made decisions based on a small share of what customers said. Conversation intelligence gives them all of it. The companies in this article got results by picking a specific, costly problem, assigning people to act on what they found and measuring the effect of each change.
AI agents make full visibility more important. A contact center that can't review every interaction has no reliable way to know whether its AI agents resolve issues, how they compare with human agents or which contact types to automate next. Conversation intelligence gives leaders that record and lets them hold human and AI agents to the same standard.
Frequently asked questions
What is conversation intelligence?
Conversation intelligence is AI that analyzes every customer conversation, across calls, chats, emails and AI agent interactions, to identify contact reasons, sentiment and agent behaviors. It links those details to outcomes such as resolution, sales and retention, so contact centers can see what drives results and act on it.
What is the difference between conversation intelligence and speech analytics?
Speech analytics transcribes calls and searches them for keywords someone defined in advance. Conversation intelligence covers voice and digital channels, understands meaning in context and connects each conversation to a business outcome.
How is AI quality scoring different from traditional QA?
Traditional QA scores a small random sample of conversations by hand, often weeks after they happen. AI scoring evaluates every conversation automatically, so rare problems such as a missed disclosure are caught and agents are judged on all of their work. Human reviewers spend their time calibrating the scoring, handling appeals and coaching.
Does conversation intelligence work on AI agent conversations?
Yes. The same analysis and quality standards can apply to conversations handled by AI agents, which lets a contact center compare human and AI performance on the same contact types and spot where an AI agent needs a human to step in.
Do agents see full conversation monitoring as surveillance?
It depends on how the data is used. When every conversation is scored, one bad call doesn't define an agent's rating, and supervisors can coach on what agents do well. Telling agents what is measured, and why, also helps.
How accurate is AI conversation scoring?
Accuracy depends on how clearly scoring criteria are written and how well models are tuned to the business. Metrigy's research puts predicted CSAT models at 85% to 95% accuracy. Teams should compare automated scores with expert human reviews before relying on them, and keep calibrating as scorecards change.
How do you measure the ROI of conversation intelligence?
Count the value of the changes it leads to, such as fewer avoidable contacts, lower QA costs, fewer transfers, shorter handle times and higher conversion or retention. Estimating each change's value before acting and documenting the result afterward makes the return easy to show to finance and other teams.
