Harnessing Synthetic Users: Enhancing Product Management Research Efficiency and Insight

Product management has always been humbling work. We sketch bold ideas, debate roadmaps late into the night and pour energy into features we are sure will succeed, only to watch real customers ignore them. We celebrate the wins that transform a business, but we also live through the launches that land with a thud.

Anyone who has been in the product long enough has seen these moments. They remind us that product management is not just about frameworks, agile rituals, or clever features. At its core, it is about empathy, timing, judgment, and most importantly, about finding ways to truly delight the customer.

This is why synthetic users, artificial intelligence agents, and agentic workflows matter. They are not buzzwords or flashy demos. They are practical tools to address a problem we all carry: the painful gap between what we think we know about customers and what they do.

The Market Research Struggle

Let me paint a familiar picture. You're in a product workshop. The room is buzzing. Sticky notes cover the walls. The product team sketches out a persona on a flip chart: "Sarah, 34, marketing manager, values efficiency." Everyone nods, and it feels good to put a human face to the problem.

But six months later, Sarah is still smiling from that laminated poster while the market has already moved on. Customers changed, competitors launched and poor Sarah stayed frozen — aged out like our decisions.

To be clear, deep user research doesn't always save us either. By the time interviews are scheduled, surveys run and transcripts analyzed, the roadmap has already shifted. The business doesn't wait. So, we're left with a painful trade-off: slow down to learn or move fast and risk being wrong.

Product managers live with this tension every day. It can be draining, and it is exactly why the idea of synthetic users is so compelling — a way to speed up insights without sacrificing depth.

Where Human Participants Fail Us

I'll never forget the first time conducting research failed me. We had a new feature in development, one that we believed customers really wanted. We'd built the user journey, done the surveys and run the focus groups. Customers nodded along with positive facial expressions and said all the right things. On paper, the data collected and real user data said it was a sure bet.

We launched. Adoption was almost zero.

When we went back to the specific user group and asked why, we heard things in these user interviews that they hadn't told us before. In real life, some didn't understand what the feature was for. Different users didn't trust it, and some simply had no use for it in their daily lives. None of these valuable insights had shown up in the original user research or user behavior.

It wasn't dishonesty or lack of skills from our product managers; it was just human nature. People often don't know how they'll behave until they're in the moment. They want to be polite and give helpful answers. They imagine themselves acting one way, but reality pulls them in another.

That experience stuck with me. It convinced me that we need better tools and ways to see a more accurate representation of behavior before exposing real customers to frustration.

What Decades of Building Products Have Taught Me

Over time, product managers come to learn a few hard truths.

First, customers will always surprise you. They'll use products in ways you never intended. They'll ignore features you thought were essential, and they will discover "workarounds" you didn't see coming.

Second, user research is imperfect. Interviews can mislead. Surveys can contradict actual behavior, and though analytics can tell you what happened, they rarely tell you why.

Third, decisions compound. Every backlog choice has an opportunity cost. Every month you delay clarity, you widen the gap between what you're shipping and what customers need.

Synthetic users don't eliminate these challenges, but they give us a way to simulate and pressure-test decisions before the stakes are high. It's all about lowering risk.

The Synthetic Customers that (mis)Behave

Here's the key difference: synthetic users don't just describe, they behave — and sometimes misbehave (more on that later). By that, I mean they don't sit still on a poster. They act out scenarios, log into your app, attempt tasks, make mistakes, recover, succeed, complain and even argue.

Think of them as a persona in motion. They're modeled on real data but expressed as behaviors.

Here's a scenario: you're testing a payments app. You run synthetic users representing audience segments such as freelancers, small business owners and families through onboarding. Some finish setup in under a minute, others stumble over jargon, while one refuses to continue because the permissions feel invasive, another flags that your captcha doesn't work with screen readers, and one totally misunderstood the context because they don't speak the language natively.

That's not theory, that's a rehearsal and what we can do today. Synthetic users give us a safe and inexpensive way to explore how different types of customers might react, while lowering the risk of real frustration in the wild.

Synthetics as a Rehearsal

The future of product management is rehearsal. Just as a cast and crew prepare before opening night by practicing lines, testing lights and setting the stage, product managers will use synthetic customers to rehearse journeys and strategies before they reach real customers.

This idea is, in fact, not new. In medicine, researchers use in-silico simulations to test how drugs might behave before ever reaching a patient. Synthetic rehearsal in product management follows the same principle: learn safely, anticipate risks early and refine before reality sets in. Rehearsal means we launch more confident, more empathetic and ready to delight from the very first performance.

At EPAM, we harness this rehearsal within the AI/Run platform: generating synthetic users, introducing them to our agentic workflow (ADLC) to power our SDLC and PDLC to generate user insights and iterate faster than ever before by bringing the best ideas to market for our clients.

Synthetics with Intents: Predict Customer Behavior

One of the most powerful capabilities of synthetic users is that they can carry intents. This word gets thrown around a lot, but in simple terms, an intent is just a goal or motivation that drives behavior.

Traditional personas capture identity, but they rarely capture intent. You might have a persona that says, "Alex, 29, graduate student, loves technology." That tells you a bit about who Alex is, but it doesn't tell you how Alex behaves when faced with a choice of saving five minutes or five dollars. Motivation, not demographics, drives the decision.

This is where synthetic users shine by giving them intents. Imagine two synthetic personas in the same ride-sharing app. One is driven by the intent to save time, may pay extra for a faster ride and won't hesitate to choose premium options. The other is driven by the intent to save money and tolerate waiting longer, accept detours and will prioritize cost savings over convenience. Same scenario, different outcomes, because the underlying goals (the intents) differ.

Adding intents turns synthetic users from sketches into actors. I've used synthetic users with different intents in product reviews, and the conversations they spark are fascinating. When the "save time" persona races through onboarding but the "save money" persona gets stuck on pricing, the team doesn't debate in the abstract anymore. They see it with the personas in motion, revealing the trade-offs we'd eventually force on real customers.

Why User Behavior Research Studies are So Hard

As a product manager, I am passionate about synthetic users because I've experienced the pain of traditional user research. The backlog, release trains and leadership never wait, and too often decisions must be made, and work must move forward with incomplete insight.

The other reality is fatigue. Research teams burn out trying to keep pace with demand. Everyone wants more studies faster, but the logistics are unforgiving and the costs are high. Recruiting participants takes time, scheduling sessions is hard and transcribing and analyzing feedback is tedious. You can't just snap your fingers and produce deep insights at scale.

This is where synthetic user rehearsals help. They don't replace the product manager or research teams; they augment them and give them breathing room. They let us answer preliminary questions quickly, so product managers can spend their limited time where human depth is essential. Instead of wasting weeks confirming that a flow is confusing, we can run synthetic rehearsals overnight and then ask product managers to dig into the real human emotions behind that confusion.

Qualitative Research: Listening for the Why

Qualitative research has always been my favorite. Hearing customers in their own words hits differently. A single frustrated quote can reframe a roadmap, and a hesitant pause can make you rethink a flow you thought was obvious.

I once watched a customer struggle with a feature we thought was intuitive. Halfway through, she sighed and muttered, "Why is this so complicated?" That one moment reminded us that customers do not care how clever a design is. They care about getting something done without friction.

Synthetic users can't replicate the emotional punch of a human voice, but they can scale the process of surfacing rationales. They can say, "I abandoned this flow because the permissions felt intrusive," or "I hesitated here because the instructions weren't clear." These may not be real quotes, but they mimic the kinds of objections you would expect to hear in the field.

The advantage is volume. Instead of ten interviews, you can run a hundred synthetic sessions overnight. If multiple personas with different intents flag the same step as confusing, you have a clear hotspot. You still validate with real customers, but you already know where to probe.

Quantitative Research: Scaling the What

If qualitative research gives us the "why," quantitative research gives us the "what." It shows how many people drop off at step three, how long onboarding takes, or what percentage upgrade to premium. These numbers are essential for measuring progress, prioritizing fixes, and justifying investments.

The challenge is scale. You need enough customers to detect patterns, proper instrumentation to capture data, and time to gather results. Too often you only discover a broken flow after thousands of customers have already been frustrated.

Synthetic users help by generating early, directional numbers. You can run hundreds or thousands of simulations overnight to produce adoption curves, completion rates, and churn forecasts. For example, test three onboarding designs and compare how many complete setup in under two minutes, how many abandon at step two, and how many retry after failing once.

The numbers are not perfect. They are models, rehearsal metrics. Their value lies in anticipating reality, not replacing it. If synthetics suggest that 70 percent of users will abandon at a step, you can flag it early. Later, when real data comes in, you see if the prediction holds. If it doesn't, you learn the model's limits and improve it.

It is rehearsal, not reality, and rehearsals matter because they let us learn quickly, adjust safely, and launch with greater confidence.

When Human Behavior Misleads Product Managers

Here's an uncomfortable truth: human behavior misleads product managers.

While it's sometimes intentional, often customers just want to be polite, don't want to hurt your feelings, and other times it's unconscious. They misremember or rationalize, or say what they think they want, not what they'll do.

I've been in interviews where the target audience praised a feature enthusiastically, only for adoption to be almost nonexistent after launch. I've seen surveys that suggested strong demand, and when the product shipped, usage was flat. This isn't because customers are liars; it's because humans are complex, emotional, inconsistent and sometimes irrational.

Synthetic users don't eliminate this problem, but can provide a counterbalance. They react consistently under the same conditions, and they won't try to please you (unless you make them). They don't forget what they just did; they hold a context for a long time and simply behave according to their parameters. If the parameters are not producing the expected result, you improve them.

That consistency is valuable because when human feedback and synthetic behaviors align, you gain confidence, and when they diverge, you've again found a signal worth exploring. I've learned to treat these divergences not as errors, but as invitations to dig deeper, to ask better questions and to look for hidden variables we might have overlooked.

How Synthetic Users Complement Humans

By now, it should be clear: I don't see AI-generated personas completely replacing real users or as replacements for UX research or product managers; I see them as complements, rehearsing together.

Human interactions give us depth, bring emotion, nuance and empathy. They show us hesitation, excitement and frustration in ways no model can replicate. Synthetic users solve for breadth, speed, scale and consistency. They let us test scenarios quickly, rehearse edge cases cheaply and generate directional data before the stakes are high.

When you put them together, you get something elementary, as Sherlock Holmes would say. Synthetic users point you toward the pain points; humans help you understand the stories behind them. Synthetic users simulate scale, humans validate meaning.

Hallucinations as Inspiration

One of the most common criticisms of AI is that it "hallucinates." By this, people mean that it sometimes generates content that isn't factually accurate, sometimes not even relevant. In most contexts, like answering a straightforward knowledge question, hallucinations are a problem, and in the context of product research, I've come to see them as an unexpected gift.

When a synthetic user "hallucinates" a reaction, it can surface edge cases that no one in the room had considered. I've seen synthetic personas raise objections that weren't programmed into them, like refusing to continue onboarding because the color contrast felt inaccessible, or objecting to a permission request that seemed too invasive. Were those hallucinations "wrong?" Technically yes. But they sparked conversations that led us to examine those "what ifs" in the real product and have resulted in new product needs.

The key is not to take hallucinations at face value, but to treat them as human prompts that expand the range of scenarios you consider and stretch your imagination. I consider hallucinations as sparks of creativity, and in product work, creativity is often what helps us uncover ways to delight customers.

Critical Thinking and the Person in the Middle

For all their value, synthetic users also carry risk: the risk of being treated as oracles. I worry about teams running simulations, seeing the outputs and treating them as gospel truth. That would be a terrible mistake and frankly lazy.

Synthetic users raise the floor of insight, but they don't raise the ceiling. Humans must still interpret, question and validate, which is why I believe in always having a person in the middle.

A person in the middle means a product manager, engineer or designer reviewing synthetic outputs before they shape a decision. It means asking: Does this align with our customer knowledge? Does this contradict what we've seen in real data? Is this a signal or just noise?

Without a human in the loop, synthetics risk being dismissed as noise or, worse, being trusted blindly. With a human in the loop, they become partners who bring critical thinking, empathy and judgment; things no large language model (LLM) can replace today.

The role of product leaders in a world with synthetic research is one of orchestrator: guiding the interplay between human insight and synthetic foresight, ensuring that evidence from both is harmonized into decisions that delight customers and move the business forward.

Predict Customer Behavior Without Pretending We're Psychics

One of the most exciting uses of synthetic users is prediction. Product strategy often hinges on questions like: What will happen if we raise prices? What will adoption look like if we launch a freemium tier? What will churn look like if a competitor enters the market?

In the past, these were educated guesses at best. Now, with synthetic audience segments, we can run tens of thousands of simulations. We can model likely churn under pricing changes, adoption curves under new features, impact of a marketing campaign or reactions to competitor moves overnight.

Does this make us psychic? Of course not! Predictions are models, not certainties, but they reduce guesswork and give us new perspectives and directional foresight.

Consider an entertainment provider exploring changes to its pricing model. Synthetic users suggested that raising ticket prices beyond a certain threshold would trigger a drop-off in cost-sensitive segments, while other customers remained loyal if bundled offers or premium experiences were included. When these adjustments were later tested in the market, the predictions weren't flawless but proved directionally accurate, giving leaders a valuable head start and the ability to prepare countermeasures before rollout.

That's the real value: not certainty, but preparation with directionally correct predictions. Predictions let us stress-test strategies before committing to get a better sense of risk and allow us to use the collected data to prepare countermeasures in advance.

The Cultural Shift Inside Teams

Whenever I introduce synthetic users into a team, the reactions are mixed. Some are curious, some are skeptical and some want nothing to do with it. I've heard concerns like, "This is fake data," "What if it hallucinates," or "Are you trying to replace product managers?" Those reactions are natural because new ways of working always generate fear.

What I've found is that skepticism fades once teams see the value. When synthetic sessions surface insights that align with real customer feedback, critical thinking kicks in and trust grows. Slowly, the narrative shifts: from fear of replacement to appreciation and excitement for what else can be explored.

This real cultural shift is helping teams see that synthetic users aren't competitors to human product managers. They're partners that handle the repetitive, often time-consuming, scalable rehearsals so humans can focus on depth and empathy, making adoption smoother.

I believe that successful integration of synthetic users isn't just about technology; it's about culture, empathy towards our colleagues and the future of product management as a craft.

With Any Powerful User Research Tool Comes Great Responsibility

We need governance, attention to Personally Identifiable Information (PII), the need to minimize bias and transparency in how synthetic users are created and synthetic customer simulations are run. These aren't checkboxes for compliance but core features of product quality.

I've had conversations with executives who worry that synthetic research might introduce new biases or give a false sense of confidence. Those are legitimate concerns, and the answer isn't to avoid synthetics/AI models, but to manage them responsibly.

To do that means documenting assumptions, labeling synthetic user-generated output clearly, documenting intents, parameters, guardrails and creating patterns. It means always validating critical insights with real customers before making big bets.

In other words, synthetic research needs the same rigor we expect from traditional research, PDLC/SDLC workflows or design thinking practices. Without that rigor, we risk undermining trust and losing the opportunity to design a responsible new tool in the product manager's toolkit.

What the Future of User Research Looks Like

Looking ahead, I believe user research will become continuous rather than episodic.

Synthetic users will handle the breadth: simulating scenarios, surfacing edge cases and forecasting likely adoption. Human product managers will handle the depth: uncovering motivations, validating meaning and capturing the emotional truth.

Together, they'll form a loop:

1. Simulate (Synths)

2. Validate (Humans)

3. Recalibrate (Humans + Synths)

4. Repeat

That loop means discovery and organic research never stops. It becomes a living system, integrated into product development just like automated testing or continuous deployment for engineering. Instead of being a bottleneck, research becomes a flow that provides valuable insights to decrease the time to market with (hopefully) lower risk and better outcomes.

This shift won't happen overnight. It requires new tools, new skills and most importantly, a new mindset. I believe that once you experience it, you won't want to go back.

What It Means for the Product Manager of 2030

The rise of synthetic user research reshapes the role of the product manager. We are no longer backlog administrators or ticket trackers. The product manager of the future is an orchestrator who brings together synthetic user simulations, human insights, AI agents and development teams to turn evidence into outcomes.

This doesn't make the role easier; rather, it makes it more meaningful. Instead of reacting to incomplete data, the product manager of tomorrow will guide teams with foresight and shape outcomes rather than simply managing tasks.

I imagine the product manager of 2030 not buried in spreadsheets or tickets, but running simulated environments, interpreting signals and making judgment calls. Fluent in both empathy and machine reasoning, they will know how to question models, run synthetic interviews, validate with customers and lead growth with confidence.

Part scientist and part storyteller, they will navigate uncertainty with skill and optimism. Most of all, they will balance human intuition with synthetic persona rehearsal, becoming leaders who orchestrate insight at scale and build products that truly delight people.