How CVS Health created an always on customer using agentic twins

When it comes to customer listening, many organizations make bold claims about their customer centric culture, or the organizational drive to put customers at the heart of everything they do. They run surveys and focus groups, spend budget and time asking real people to share their thoughts and experiences, and after taking time to extract the Voice of Customer (VoC) insights and strategize next best moves, they might – eventually – roll out an experience or service improvement.

At CVS Health, analogue VoC methods have now been augmented with an AI-powered solution that can give deeper insights in real time, and provide the organizational agility to act on customer demand before it becomes outdated.

Since October 2025, the company has been rolling out agentic AI twins; a simulation technology that is used to test and improve customer and patient experiences before they go live. The simulations are calibrated on millions of data points to pressure test customer journeys, messaging, and service design. Built on 2.9 million consented responses from more than 400,000 participants across more than 200 behavioral scenarios, the agents are modeled on data from real customers and patients and are particularly useful for understanding hard-to-reach populations. 

“We used to have a chair we put in every meeting to represent the customer,” says Sri Narasimhan, VP of enterprise customer experience and insights for CVS Health. “We don't have to do that anymore because we literally have hundreds of thousands of AI representations of our customers available to us.”

The agentic twins are modeled on real individuals – rather than averages – enabling more precise insights across different audiences and giving CVS Health the ability to uncover friction and understand the “why” behind behaviors and demand. What once took weeks of research can now be done in hours and, because there is no fatigue, Narasimhan and his team can drill down into the responses and behaviors they need to uncover virtually unlimited insights. 

Don't miss any news, updates or insider tips from CX Network by getting them delivered to your inbox. Sign up to our newsletter and join our community of experts. 

Learn More

Amplifying the customer voice

As with all the best applications of AI, this project is the result of needing to solve a real-world problem, namely, how can the customer voice be amplified? 

“Traditionally, with all these CX tools and platforms you have to go out and get feedback after the fact. Even for advert or message testing, you’re running pilots and questioning customers retrospectively. But we wanted to do something where we could say, how do we truly bring the consumer voice into our processes from start to finish? Both from the development of product all the way through to how it's impacting consumers, and how we can get ahead of the cycle?”

Narasimhan and his team started to explore how they could build an always on customer using digital twins that would allow the insights team to have “literally hundreds of thousands of AI representations of our customers available”. They worked in collaboration with AI startup Simile; it’s the first company to build and bring this technology to market and CVS Health is an investor.  

The project augments the extensive VoC work the insights team at CVS Health conducted, including customer experience surveys, focus groups, ethnographic studies, NPS tracking and transactional relationship studies that generated 18 million pieces of solicited feedback across the enterprise annually. To give the best results possible, the agentic twins are used in tandem with predictive modelling.

“What's interesting about the agentic simulation capability is it goes through how people will think but drilling down into the reasons why. We're actually building the customer’s decision calculus, which means we can apply that decision calculus to new scenarios,” Narasimhan explains. “Now it’s no longer a case of predicting how you are going to behave based on what you did in the past. I'm now predicting what you would do based on this situation and how you think and make decisions. That's a big change for us because that is how you extend prediction and understand consumers before an action happens,” he adds. 

Ensuring the accuracy of digital twin models

One of the greatest challenges in using AI is avoiding hallucinations and inaccurate outputs. When the AI is representing customers and guiding strategy, confidence in the outputs is essential. 

The data that feeds the agentic twins is derived from four sources with a focus on zero and third party data, and customer consent always captured. The zero party comprises surveys and behavioral interviews sourced from traditional panels, while the third party purchased data covers spending habits, media diet, internet behavior and consumption patterns, as well as how customers engage with various search engines. 

On how this data is used to build the twins, Narasimhan says: “When you put that into the Simile engine it generates what's called a decision profile – the customer’s contacts, behaviors, how they respond to stimuli and situations – and it builds what I would call decision engine, or their decision calculus.”

Now executed at scale, CVS Health has an army of around 100,000 digital twins and aims to create more as the program grows. And while the “rubbish in equals rubbish out” mantra that applies to all AI is still applicable here, data quality is far from the only way the insights team ensures the accuracy of outputs. 

The models used by CVS Health are “85 to 95 percent accurate” Narasimhan says with a specific accuracy figure also given against each output and all results pressure tested against a population level model.

“In addition to the individual twins we build up, Simile built a population model of the entire US population. If you put those two together, you have a very powerful decision engine that knows how people are making decisions and behaving, and we can validate with existing studies, existing research, and data. If the two start to deviate, we also take a look at the model and assess if we need to adjust things and improve it,” Narasimhan says. The human intervention in these adjustments is intended to catch inaccuracies before they cascade. 

“If my team is running the model and we get something we didn't expect, I don't tell them to discount it because it could be a new insight that we didn't understand. We don't want to live in an echo chamber. But because it's research-led, we can see if a result is unexpected or not. Maybe we need to go take a look at a specific output a little more carefully and make sure it's right. There are a couple different ways we do it,” Narasimhan explains. 

It’s also critical to ensure the modeling isn’t extended beyond its capabilities. “You have to apply it properly with the right guardrails,” Narasimhan says. “That's where some of the testing and the humans being involved is really helpful because you're not going to want to extend this thing beyond what it can do.”

Identifying the business case for AI 

As all practitioners know, AI must be deployed to solve specific real-world problems, rather than being deployed simply because others are doing so, or the organization demands it. At CVS Health, the business case didn’t stop with the need to amplify the voice of the customer. Specific use cases included message and advert testing and scenario testing to inform operational changes.

Narasimhan explains: “If you want to understand how people are going to respond to different messaging or different advertising, you can get this information almost instantaneously. We've also been doing a lot of what you would traditionally call simulating choice. It's a really interesting use case because humans, again, cannot respond to more than like three or four choices. If I put 30,000 choices in front of a person, their brain would know how to process that, but these agentic twins can do that.

“It means can pressure test capital investments in different things that you would want to do in terms of where you're investing in operations and CX in a retail environment. You can put those scenarios in front of the customer, all the different permutations, and they can actually respond and you will know what will be the most impactful operational changes to make,” Narasimhan adds.

The ability to road test a hypothesis in real again against the models – whether to operations, experience, product, or other factors – not only saves time, but also money, and potentially even brand reputation.

“Think about how companies spend millions learning lessons after programs go live, rather than validating them upfront. You don't need to do that now. It's a dress rehearsal then you go live with a lot of confidence that it's going to work,” Narasimhan says. 

“In some of our meetings, we're able to bring up the tool and say, ‘hey, you had this question, well here's what ten thousand people are saying about what you asked’. Being able to do that live in a meeting is a massive unlock in terms of visibility because we're able to get those insights so quickly. You don’t respond by saying ‘I’ve got to go back to and go run that’, you say ‘here’s the answer’. This AI capability is unique in that it's amplifying human voices and they’re now with us the whole time,” he adds. 

Preserving customer trust 

Customers are increasingly aware of how AI works and uses their data. CX Network’s annual research into the state of CX in 2026 found that customer trust and transparency is essential to AI-powered CX. In the top challenges for 2026, consumer demand for data security emerged as the fifth most selected response from a list of more than 14 choices, while consumer privacy entered the top 10 CX trends for the first time this year, and awareness of how AI works and uses customer data was the leading customer behavior trend influencing the work of practitioners.

It means ethical AI use and governance are imperative for any AI use case, let alone something as sophisticated as CVS Health’s pioneering use of agentic AI twins.

“Trust is critical in health care and consumer trust is really important to us,” Narasimhan says. “For everything we do we capture consent – that is really important – the consumers agreed to share this information when we built these twins.”

In CX, one way to ensure transparency and preserve trust is to make the reasons for data collection clear and ensure customers understand the what they get in return for sharing their data. Narasimhan says the twins give customers “a bigger say in what we do than they've ever had by creating these always on personas”. He adds: “It's being used specifically to improve the experiences that we have, the offerings we have, and make sure that we are delivering the most consumer centric offering that we can in health care.”

Consent isn’t the only consideration. Ensuring customers understand that the use of agentic AI twins will not replace the direct communication CVS Health has always sought to have with its customer base. “We’re not creating a bunch of robots so we never have to talk to people – that's not the way this works. It's creating an amplification of that voice,” Narasimhan says. “We can extrapolate what that customer told us to a series of other scenarios and now the person who gave us feedback in that one moment can give us feedback in every moment. We've made customers more powerful in our organization and the advocacy we do as an organization I take incredibly seriously, and that's the unlock here.”

In breaking down the transaction cost in having to find and survey customers, CVS Health has a new means by which it can measure and meet the needs of customers and patients.

Quick links 

• The Future of Customer Listening: dynamic, active, and AI-powered
• The AI governance maturity model for CX leaders: Where does your organization stand in 2026?
• How Lidl built a 92% accurate AI model to transform customer service

48021.004 All Access: The AI Revolution in CX 2026

All Access: The Revolution of AI in CX will explore how to demonstrate ROI from your AI initiatives, establish governance frameworks that ensure transparency and accountability. Join us for your chance to stay at the forefront of the CX evolution.

Register Now