In markets where a handful of blockbuster products drive most of the revenue, companies live or die by their next big launch or critical feature change. But traditional new product development is too often slow, expensive, and unreliable, which helps explain why so many launches miss the mark. Meanwhile, companies are under pressure to deliver more personalized, innovative, and timely products without expanding their budgets.

Enter synthetic customers. Built with generative AI and machine learning, synthetic customers offer a smarter, faster way to test and iterate. By layering insights gleaned from synthetic customers on top of feedback from real customers, companies uncover deeper truths, cut research costs, and move faster. It’s a breakthrough that unlocks levels of experimentation that were previously out of reach.

What are synthetic customers?

Synthetic customers consist of AI-generated proxies that emulate human behavior, preferences, and decision making. These digital agents are built from a mix of internal company data (transactional, behavioral, demographic, and voice-of-the-customer research) and external sources such as product reviews and market-level analysis. Synthetic agents can be used to evaluate new product concepts, test marketing campaigns, and simulate buying behavior.

While early uses were mostly qualitative, quantitative evidence is growing. A recent study led by Stanford University and Google DeepMind found that digital agents trained on interview data matched human survey responses with 85% accuracy and mimicked social behavior with 98% correlation, demonstrating the potential of this approach to approximate real-world behavior at scale. Our own experience shows early promise: Tests take half the time and cost one-third as much as traditional methods. Because these models can run continuously and learn, they fill in knowledge gaps and enable scenario planning and forecasting that simply aren’t feasible with traditional methods.

Where they change the game

We’re seeing five use cases stand out across industries (see Figure 1):

• Value proposition design and forecasting: Simulate how customers might respond to new features, pricing, and bundles before committing to a full launch.
• Persona development and segmentation: Build nuanced, data-rich personas that sharpen customer understanding.
• Marketing and ad testing: Rapidly test campaigns—especially those that are high-stakes, experimental, or target hard-to-reach audiences—using synthetic customer segments.
• Predictive Net Promoter Score℠ (NPS®) modeling: Model how changes in experience or offering could affect customer sentiment.
• Frontline training: Train sales or call center teams using synthetic personas that mimic real-world customer reactions and objections.

Figure 1

One major telecom provider recently used synthetic customers to break into underpenetrated value-first segments without cannibalizing its premium brand. By pairing a synthetic capability with traditional research, the company tested features, pricing, and promotion options to pinpoint optimal launch strategies. Over time and with new data sources, sophisticated prompt engineering, and iterative testing, the model’s predictions increasingly aligned with real-world outcomes, demonstrating how iterative testing and smarter prompts can run a huge number of permutations at high speed, thus saving costs and improving accuracy before the final test with real customers.

Getting started

Synthetic customers can be powerful, but they’re not a plug-and-play replacement for real customers. Success depends on a clear-eyed, rigorous approach:

• Know the limits. AI can introduce bias or miss nuance. For instance, bots can be overly positive and agreeable. Use synthetic research to augment, not replace, critical insights.
• Don’t cut corners. Building the synthetic capability may require breaking down data silos or backtesting past launches to train more accurate models.
• Design the bots. Which specific bots to build will depend on the goal, whether that’s making new products to launch, understanding new segments, or finding out why customers churn.
• Sweat the data. To power the bots, what data does the company need, and how can it gather the data? These models are only as strong as the training data and architecture behind them. Proprietary, well-integrated data makes the biggest difference.
• Start small. Deploy synthetic customers first on low-stakes use cases to see if the results are logical.
• Get creative. Try open-ended use cases such as assessing usability of a website.
• Understand bot quirks. For example, in longer surveys a bot sometimes will replicate responses from earlier questions or skip questions altogether.

As companies get more comfortable with synthetic customers, bold ideas become less risky. Marketing teams can refine campaigns before spending on media. Design teams get immediate feedback at every step of the customer journey. And finance teams can model revenue potential with greater accuracy. Organizations that master this capability will gain an edge in deeper customer understanding, faster time to market, and more resilient innovation pipelines.

ネット・プロモーター・システム℠

顧客・従業員の優れたロイヤルティを獲得できる文化を構築します。