Cresta's Synthetic Customers Point to the Next Evolution of CX Analytics

Cresta has announced Synthetic Customers, describing it as a first-of-its-kind capability that creates realistic, representative, and evolving customer personas directly from an enterprise's own conversation data. The personas are designed to power simulated conversations for AI agent testing and human agent training, while surfacing deeper behavioural insights across the business.

Analytics in CX has long been defined by outcomes. Which customers are likely to leave? Which interactions carry escalation risk? Which prospects are closest to a purchase? The industry's efforts have focused on forecasting future events through what is called predictive CX. Cresta's latest launch, however, is not focused on what customers will do, but how they are likely to behave.

Cresta argues conventional approaches to persona building tend to draw on surveys, CRM data, and support records. The problem is that these sources capture a static and often incomplete picture that reflects internal assumptions about customer behaviour rather than how customers actually communicate in live interactions.

The problem, according to Cresta, is one that goes to the heart of how enterprises use customer data. The capability works by mining historical calls, chats, and emails to build profiles that capture the full texture of real customer interactions, including emotional states, communication patterns, and the kind of unpredictable behaviour, such as sudden topic changes or expressions of frustration, that scripted personas rarely reflect. Those profiles are designed to evolve continuously as new conversation data comes in, rather than representing a fixed point in time.

Ping Wu, CEO of Cresta, reminded how close this critical data lives to us: "The data that enterprises need to build accurate customer personas for better testing, training, and decision-making is right in front of them, at their fingertips. It lives in every call, chat, or email conversation."

Wu added that the launch represents only a starting point. Accurate AI representations of customers could be applied to everything from incident planning to market research, with use cases he described as "endless."

Other vendors are pursuing related ideas, though from different angles. LivePerson's Syntrix platform, launched in March, allows brands to stress-test AI agents against diverse synthetic customer personas and edge-case scenarios, and to train live agents against realistic simulations before they interact with real customers. PumpCX, which expanded into North America in the same month, positions its platform as an independent AI assurance layer, using synthetic interactions to continuously validate AI-driven customer journeys in both pre-production and live environments.

What these approaches share a common premise that scripted testing is no longer sufficient as agentic AI takes on more autonomous roles in customer-facing workflows, and that something closer to genuine customer behaviour is needed before deployment. What separates Cresta's approach is the data source. Where PumpCX and Syntrix use synthetic interactions to test and assure AI systems, Cresta's personas are derived directly from an enterprise's historical calls, chats, and emails, capturing real language, emotional patterns, and behavioural tendencies observed in those interactions. The ambition, as Cresta frames it, is not just better testing but a living model of the customer base.

Larger CX vendors, including NICE, Genesys, Observe.AI, and Five9, have invested heavily in conversation intelligence, predictive analytics, and agentic AI orchestration. Yet none have publicly launched capabilities that closely resemble conversation-derived synthetic customer modelling. Their focus appears to remain on building the data foundations and continuous learning systems that could eventually support this kind of behavioural intelligence, rather than the intelligence itself.

If behavioural modelling grounded in real conversation data were already an established category, the largest platforms would likely have competing products. Instead, the terminology remains inconsistent, the approaches vary, and the market has yet to converge on a standard model. Whether this ever develops into an established category remains to be seen, but the concept and the technology are already there.

For years, predictive CX has represented the frontier. What Cresta's announcement indicates is a possible evolution beyond forecasting what customers will do, one that might be called simulated CX, where enterprises can model what could happen under different scenarios before any decision is made or any AI agent goes live. Descriptive CX asked what happened. Diagnostic CX asked why. Predictive CX asked what will happen. Simulated CX asks what could happen, and does so before deployment, not after.

Although many of the largest platforms have yet to follow, this could mean that the phase is only just beginning. CX leaders should keep an eye on this potentially transformative space and the performance of solutions like Cresta’s Synthetic Customers.