Despite years of investment, most enterprises still can’t easily unify voice, chat, and customer relationship management (CRM) data to generate meaningful customer insights.
Unifying this data is difficult because organizations have bought or consolidated applications, not the underlying data layers and platforms.
They are left behind with incompatible schemas, duplicate identities, and point-to-point or custom data integration/ETL workflows that were never built for a shared view of customer journeys.
These data sources have different data models – CRMs think in accounts/opportunities, ticketing systems have cases, chats are in sessions and threads. The same customer who appears in all these sources requires a strong, reliable customer identity graph.
“The most valuable CX signals live in call transcripts, notes, and chat logs – these are unstructured, free-text fields, often noisy, multilingual, and full of shorthand,” explains Tapan Patel, research director, AI-enabled customer data and analytics, IDC.
Organizations underestimate the effort needed to transcribe, parse, tokenize, classify, and normalize to cluster similar conversations, detect patterns over time, and map those signals back to customers and journeys.
“They end up with structured events and attributes you can join to your behavior and operational data rather than just a pile of text or audio,” Patel said.
Enter AI Discovery
AI-powered data discovery and semantic search tools represent a significant leap beyond traditional Business Intelligence (BI) or keyword analytics.
Traditional BI typically relies on structured data and predefined queries, offering insights into “what” happened based on known metrics. Keyword analytics, while useful, are limited to exact matches and often miss context, synonyms, or implied meaning.
In contrast, AI-powered tools tapping into clean and unified data, especially those leveraging semantic search, can understand the meaning and intent behind unstructured data.
“They go beyond keywords to grasp context, identify relationships, and uncover patterns that are not explicitly stated,” explains Chang Chang, senior director, product, cloud CX solutions, Cisco.
For example, they can detect emerging trends in customer complaints, identify root causes of issues, or predict churn risk by analyzing the sentiment and topics discussed across thousands of interactions, even if customers use different phrasing.
“This allows CX teams to move from simply reporting on data to actively discovering actionable insights—and importantly the rationale behind those insights—that drive proactive improvements,” he said.
Lenses.io head of AI Tun Shwe says the true differentiation in AI-powered data discovery emerges when the engines supporting them come through real-time data pipelines.
Unlike traditional tools that might analyze a call log 24 hours later, AI tools integrated with real-time data can proactively interpret and analyze metrics that can further improve the customer experience, like sentiment and customer intent.
“This allows an AI model to be adaptive to new behaviors rather than just reporting on old ones,” Shwe said.
Governance Guardrails
Deploying AI tools for interpreting customer intent, sentiment, or churn signals requires robust governance guardrails to ensure ethical, responsible, and effective use.
Shwe says IT leaders responsible for improving customer experiences should pivot towards real-time workload visibility.
“It’s vital that organizations have dashboards that are able to visualize exactly what data stacks their AI systems are touching, and how the specific data is interpreted,” he explains.
This visibility acts as the primary guardrail, allowing teams to spot vulnerabilities and potential AI hallucinations in real time.
“Additionally, these AI systems ensure that only approved agents gain access to sensitive customer data, which avoids risk and ensures compliance,” Shwe said.
Assessing AI Discovery Effectiveness
To measure whether AI-driven discovery is genuinely improving CX outcomes rather than just generating more data or dashboards, enterprises must focus on quantifiable improvements in key CX and business outcomes.
Patel says he advises enterprises to tie together the discovery and insights to find out whether behavior changed, or if outcomes improve for both the organization and for the end customer.
“Ultimately, AI discovery should help you fix things faster, recommend improvements and design better experiences, not just visualize problems,” he said.
If customers aren’t feeling the difference in reduced effort and clearer and contextual interactions, the technology isn’t doing its job yet.
“My practical suggestion to many enterprises is that every insight surfaced, understand what process or step needs to change and which KPI they did expect to move,” Patel said.
If possible, measure specific metrics pre- and post-action (e.g., new conversational AI flow vs. existing) to identify clear signals and success.
Patel cautions that enterprises also tend to get carried away in the value/ROI measurement pressure and do not involve CX agents and supervisors when it comes to finding out whether AI-driven recommendations are usable and useful.
“If the people closest to the customer do not find it credible, adoption will stall and impact will be limited,” he said.
Chang says organizations can evaluate the effectiveness of AI in anticipating customer needs and issues through proactive insights derived from consolidated data, providing a comprehensive view of each customer’s current and past interactions.
He points to increased first contact resolution (FCR) and call containment/deflection metrics, which show how AI insights enable agents to resolve issues more effectively on the first contact, as well as how AI contributes to deflecting calls or containing issues within self-service channels.
“In some cases, improved CX can directly lead to increased customer loyalty and upsell opportunities, which can be tracked,” he adds.
Shwe says IT enterprises should prioritize measuring the speed of contextual innovation, which is essentially a reduction in time-to-resolution for systems that operate with real-time context.
“The main question is: Is your organization’s AI agents accessing live data to proactively solve problems or are they just summarizing data workloads?” he asks.
From his perspective, true AI ROI is unlocked when models are continuously adaptive
“If AI discovery tools reduce the basic operational issues for IT teams, human developers can prioritize more complex data issues,” he said.
link