TL;DR
As it becomes harder to collect customer feedback at scale, CX teams are turning to AI-modeled data—such as synthetic survey responses and digital twins—to fill insight gaps, move faster, and protect privacy. This article explores what modeled data is, why it’s gaining traction, and how to use it responsibly.

When applied thoughtfully, modeled data doesn’t replace the voice of the customer—it strengthens it. By predicting likely sentiment at key moments in the customer journey, it helps teams act on directional insights to reduce friction and market smarter decisions—even in the absence of direct feedback.

Customer Experience is entering a new era—one where AI is not just powering chatbots and routing systems, but reshaping how we listen, learn, and act on customer feedback at scale.

Earlier this July, I attended the Quirk’s 2025 Conference in New York City, where I took in nearly two dozen presentations from industry researchers, technology leaders, and CX innovators. One clear theme emerged from across the sessions: AI-modeled data—including tools like digital twins and synthetic survey responses—is poised to transform how companies measure, understand, and respond to their customers.

---

Today I wanted to reflect on the most forward-looking ideas I encountered at Quirk’s, along with my own perspective on how CX teams can implement them responsibly and effectively. It’s not just a matter of adopting emerging tools, but understanding how to apply them strategically so they generate meaningful insights, enable faster decision-making, and respect customer privacy in the process.

The message from Quirk’s was clear: if you are leading or advising a CX program today, now is the time to go beyond traditional survey programs and start building the data foundations for what is coming next.

What is Modeled Data?

Modeled data refers to information predicted or simulated by AI, using your company’s existing customer data. Instead of collecting a new survey response, an AI model estimates how a customer is likely to respond based on patterns found across similar accounts.

In a CX context, areas where this might be applied include:

• Quantitative metrics like predicted NPS or CSAT scores
• Qualitative content such as simulated open-text feedback
• Conversational tools like digital twins—simulated AI personas trained on real data that you can interact with

These predictions are not random. And they are more than just multiplying the number of respondents across your customer list. They are based on rich behavioral data and signals from customers, including:

• Demographic data like industry, region, company size, service package
• Product engagement like product usage metrics, features adopted, frequency of activities
• Support interactions like ticket volume, CSAT ratings, resolution time, self-serve access
• Conversational signals like survey verbatims, call transcripts, chat/email logs (only if customer gives consent to use)

By feeding this information into a model, the AI is trained to estimate how customers are likely to feel, think, and respond to common survey questions or prompts—whether or not that individual has recently provided feedback.

Why Organizations Are Turning to Modeled Data

A common refrain from other CX practitioners at Quirk’s was that their CX programs face growing pressure to deliver insights faster, with fewer resources, to the utmost privacy standards. Many expressed that traditional research methods are reaching their limits:

• Response rates are down—survey fatigue is real
• Fielding times are slow
• Respondent fraud is growing
• Recruiting niche audiences is time and money intensive

Modeled data offers a modern alternative. It enables CX teams to:

• Increase speed to insight and speed to action for a competitive edge
• Close feedback gaps in underrepresented segments
• Generate directional insights where traditional methods fail
• Identify friction trends earlier, before issues escalate
• Scale feedback coverage without over-surveying real customers
• Save time and money versus panel recruitment or 1:1 interviews

Forward-looking CX teams are recognizing that modeled data isn’t just a workaround, but a way to reduce friction, respect customers’ time, and prove that we are listening even when they don’t speak up.

Two Strategic Applications: Digital Twins and Synthetic Survey Responses

There are two primary ways that organizations are beginning to operationalize AI models for CX:

Applying Modeled Data Responsibly and Privately

One of the key strengths of modeled data is its ability to surface meaningful insights at scale without exposing individual-level predictions.

Internally, modeled scores (eg, NPS, CSAT) are generated at the account level for aggregated analysis. However, these modeled scores should never be reported or displayed at the individual customer level as representative of one account. In aggregate they provide directional feedback, but individually they may create risk of misunderstanding, misuse, or mistrust. Only actual survey responses should be tied to specific accounts.

Instead, the best use case for modeled data is to aggregate it to the cohort, persona, or lifecycle stage to surface directional trends across segments. This protects customer privacy while still delivering valuable insight.

The basic rule should be:

• Use real feedback for individual customer insight.
• Use modeled data for group-level insight.

If a real survey response is received from an account, the modeled result can be compared privately to calculate a model fidelity score. This helps the insights team monitor and improve model performance over time without ever using modeled data as a proxy for real customer voice at the individual level.

In a privacy-conscious world, modeled data provides a scalable way to listen better, act faster, and respect the boundary between directional insight and customer truth.

Six Safeguards for Smart Use of Modeled CX Data

To make the most of modeled data, it’s critical to design your program with care. Here are six safeguards to guide responsible and strategic implementation:

1. Guide, Don’t Replace
   Modeled data provides directional insight. It should guide attention, not act as a final decision-maker. Use it to identify customers that fall into a group that requires further attention, not as a substitute for closing the loop.
   
   
2. Measure Fidelity
   Track how well modeled scores align with real survey responses for the same accounts. This helps establish the model’s reliability and guides model iteration over time.
   
   
3. Keep Data Fresh
   Customer behaviors and journeys evolve. Modeled data must be refreshed frequently with new usage patterns, account interactions, and recorded sentiment to remain relevant. Remember: garbage in, garbage out.
   
   
4. Ensure Transparency
   Avoid black-box modeling. Internally document what data sources are used, which prompts are used, and how predictions are made. Be up front about where you are confident or uncertain in your model. This builds trust and may soon be a legal requirement in some markets in Europe.
   
   
5. Respect Privacy with Cohort-Level Reporting
   Apply modeled data at the account level internally—but when sharing results with other teams, only report at the cohort, persona, or segment level. This ensures insights remain anonymous and actionable without exposing individual customer assumptions.
   
   
6. Mitigate Bias and Monitor Input Quality
   Even the best models are only as good as the data they are trained on. Be deliberate about ensuring diversity and representation in your inputs. Monitor for data gaps or skew that could replicate bias or wrong information at scale.

Summary: A Responsible Path Forward for Modeled Data

Modeled survey data and digital twins are not gimmicks. When built on clean, behavioral data and guided by smart design, they can help CX teams understand their customers faster, more broadly, and with greater context. They also allow organizations to test and explore without exposing premature ideas to the market or over-surveying already-fatigued users.

But, as with any insight tool, the impact of these tools depends on how thoughtfully they are used.

If they are used well, modeled data doesn’t eliminate the voice of the customer—it strengthens it. It gives you a head start on what customers are likely to say, so when they do—you’re already prepared to act.

Timothy Brown, CCXP