Stetos.co: How AI Voice Agents Are Scaling User Research

Product Overview

Stetos.co is an AI-powered user research platform positioned as "insight infrastructure," built around the core concept of "Listen at Scale" — hearing user voices at massive scale. The platform deploys AI voice agents to conduct large-scale qualitative interviews, transforming thousands of conversations into actionable insights that help businesses deeply understand user needs — including things you didn't even know you needed to learn.

As a product backed by YC (Y Combinator), Stetos.co directly addresses a long-standing pain point in user research: traditional qualitative research is time-consuming, labor-intensive, limited in sample size, and difficult to scale.

Background on Y Combinator: Y Combinator is the world's most influential early-stage startup accelerator, having incubated tech giants like Airbnb, Stripe, and OpenAI. YC's endorsement means more than just funding (typically a standard $500K investment) — it represents strong validation of both the product direction and founding team, significantly boosting credibility with enterprise customers. Notably, YC has been investing heavily in AI infrastructure and vertical AI applications in recent years, and AI user research tools are one of its key focus areas. The rise of this category is driven by the maturation of large models like GPT-4, which enabled AI agents to evolve from "understanding text" to "leading complex conversations," making automated qualitative research technically feasible for the first time. This window of opportunity began opening around the second half of 2023.

Core Features & Technical Highlights

AI Voice Agent-Driven Qualitative Interviews

Stetos.co's core capability lies in deploying AI agents to automate qualitative user interviews. Unlike traditional surveys, these AI voice agents engage users in natural conversations, dynamically adjusting the line of questioning based on responses — simulating the experience of a real human researcher.

Technical Principles Behind AI Voice Agents: AI voice agents are a fusion of large language models (LLMs) and speech synthesis/recognition technologies. The core tech stack typically includes: Automatic Speech Recognition (ASR) to convert user speech to text in real time; a large language model to understand semantics and generate the next question; and Text-to-Speech (TTS) to convert responses into natural-sounding audio output. Latency control across the entire pipeline is a critical challenge — the industry standard typically requires end-to-end response times under 500 milliseconds to maintain a natural conversational rhythm. Platforms like Stetos.co must also layer an "interview logic engine" on top, dynamically deciding whether to probe deeper, switch topics, or conclude the interview based on the user's emotional tone, keywords, and topic depth. This is essentially a reinforcement decision system based on conversational state.

Evolution of Voice Interaction Quality Standards: It's worth noting that the "naturalness" of AI voice agents is undergoing a generational leap. Early TTS systems (such as concatenative synthesis before Google WaveNet) sounded mechanical with unnatural pauses, resulting in extremely low user acceptance. After 2022, end-to-end neural network-based speech synthesis (such as ElevenLabs and OpenAI TTS) achieved near-human voice quality, including tonal variation, emotional coloring, and speech rate changes. This technological breakthrough is particularly critical for user research scenarios — respondents tend to give more genuine, detailed answers when interacting with an AI that "sounds like a real person," rather than giving perfunctory responses. Research shows that an interviewer's vocal characteristics (speech rate, intonation, pause rhythm) significantly affect respondent answer depth, meaning voice quality itself is part of data quality.

This approach delivers three major advantages:

Scalability: Conduct thousands of interviews simultaneously, completely breaking through human resource bottlenecks
Research Consistency: Every interview follows the same core framework, eliminating individual researcher bias
Natural Interaction Experience: Supports both voice and chat channels, allowing users to express their true thoughts in whatever way feels most comfortable

Dynamic Outcomes & Instant Insight Generation

The platform goes beyond data collection, emphasizing the concept of "dynamic outcomes." AI agents can adjust subsequent questions and action paths in real time based on user feedback during the conversation, rather than mechanically following a fixed script.

After interviews are completed, the system instantly transforms large volumes of unstructured conversational data into structured, actionable insights — compressing the cycle from data collection to decision-making from weeks to hours.

Technical Pipeline for Converting Unstructured Data into Insights: Transforming thousands of conversations into actionable insights requires a sophisticated NLP (Natural Language Processing) analysis pipeline. This typically includes the following layers: first, Topic Modeling, using LDA or Transformer-based clustering algorithms to automatically identify recurring core themes in conversations; second, Sentiment Analysis, determining user emotional tendencies toward specific topics; third, Entity Extraction and Frequency Statistics, identifying frequently mentioned product features, competitor names, or pain point keywords; and finally, Insight Summary Generation, where a large language model synthesizes the above structured data into natural language reports. The key challenge in this process is ensuring "insight quality" — how to prevent AI from generating conclusions that are superficially correct but lack actual decision-making value. This is a shared technical challenge facing all players in this space.

Potential Application of RAG in Insight Generation: To improve insight accuracy and traceability, leading AI research platforms in the industry are adopting RAG (Retrieval-Augmented Generation) architecture. The core idea behind RAG is that when generating insight reports, the system doesn't rely solely on the LLM's parametric knowledge but also retrieves from the original conversation corpus in real time, ensuring every insight can be traced back to specific user quotes as supporting evidence. This "evidence-based" approach to insight generation effectively addresses enterprise customers' concerns about the credibility of AI-generated analytical conclusions — product managers and researchers can click on any insight item and jump directly to the original conversation excerpts supporting that conclusion, enabling bidirectional navigation from summary to raw data.

Use Cases: Who Needs AI User Research Tools

Three Major Bottlenecks Facing Traditional User Research

To understand Stetos.co's value, it's important to first clarify the basic framework of user research methodology. Quantitative research (such as surveys and A/B testing) excels at answering "how many people do this," relying on large samples and statistical significance, but cannot explain "why." Qualitative research (such as in-depth interviews and focus groups) uses open-ended conversations to uncover users' true motivations, emotions, and latent needs, typically with sample sizes of 5–30 people, following the principle of "theoretical saturation" — stopping when new interviews no longer yield new insights. The gold standard for traditional qualitative research is the Grounded Theory method, which requires researchers to maintain high sensitivity during interviews and adjust strategies in real time.

It is precisely these methodological characteristics that have long constrained traditional qualitative user research with the following challenges:

High Cost: Recruiting respondents, scheduling interviews, and transcription analysis all require significant human resource investment
Limited Sample Size: Typically only a few dozen people can be interviewed, making it difficult to cover diverse user groups
Long Analysis Cycles: Going from completed interviews to deliverable reports often takes weeks of waiting

Stetos.co solves all three problems simultaneously through AI voice agent technology, giving qualitative research the scale advantages of quantitative research while preserving the insight value of deep conversations — creating a previously nonexistent paradigm of "scaled qualitative research."

Methodological Debate and Responses Around Scaled Qualitative Research: It should be noted that the "scaled qualitative research" paradigm is not without controversy in academic and professional UX research communities. Traditional qualitative research methodology holds that the value of qualitative research lies precisely in the researcher's keen perception of interview context — including body language, moments of silence, emotional fluctuations, and other signals that cannot be captured by voice transcription. AI agents have inherent limitations in this regard. However, proponents argue that the value of AI interviews lies not in replacing deep ethnographic research, but in filling the gap of "large-scale exploratory research" — using AI to quickly scan thousands of users before formal in-depth interviews, identifying topics and segments worth deeper exploration, followed by human researchers conducting precisely targeted deep dives. This "AI broad screening + human deep research" hybrid research paradigm is becoming the new standard workflow for enterprise UX research teams.

Typical Use Cases

This AI user research tool is suitable for the following teams and scenarios:

Large-scale user feedback collection before product iterations
Target user needs exploration before entering new markets
Ongoing user satisfaction and experience monitoring
Agile development teams that need to quickly validate product hypotheses
Startups with limited resources but high-frequency research needs

Data Compliance Challenges for Enterprise-Grade Research Tools: In enterprise procurement decisions, data privacy compliance is often a critical threshold for AI research tools. User interview content involves personal opinions, consumption behaviors, and even emotional states, which under the EU's GDPR (General Data Protection Regulation) and China's Personal Information Protection Law must meet strict informed consent and data localization requirements for collection, storage, and analysis. For enterprise customers operating in global markets, AI research platforms need to provide clear Data Processing Agreements (DPA), support data residency options, and ensure access controls for interview recordings and transcription texts. This compliance cost is often the main barrier facing small and mid-sized AI research startups when entering heavily regulated industries like finance and healthcare, and is also an important dimension through which leading players build competitive moats.

Business Model & Competitive Landscape

Stetos.co offers a free trial option and uses a "2 months / double usage" promotional strategy, effectively lowering the trial barrier for enterprises. Based on its YC background and product positioning, its target customer base is mid-to-large enterprises with ongoing research needs and fast-growing tech companies.

Competition in the AI-driven user research space is intensifying. However, Stetos.co's choice to position itself as "insight infrastructure" carries deep strategic intent.

Full Competitive Landscape of the AI User Research Space: The space has developed several distinct competitive tiers. The first tier consists of comprehensive research platforms like UserTesting and Maze, which have mature enterprise customer bases and are rapidly adding AI capabilities. The second tier includes AI-native research tools such as Synthetic Users (which use AI to simulate users rather than interview real ones), Outset.ai (AI interviews + analysis), and Marvin (AI-assisted interview analysis), competing directly with Stetos.co. The third tier involves cross-domain penetration from general-purpose AI tools — some companies are beginning to use ChatGPT or Claude to directly analyze interview recordings, creating low-cost substitution pressure. Within this landscape, Stetos.co positions itself as "insight infrastructure" rather than a standalone tool, emphasizing the platform's system integration value — deeply connecting with enterprises' existing product management tools (such as Jira, Notion, Productboard) via API, so that research insights flow directly into product decision-making processes rather than remaining siloed in standalone research reports. This "infrastructure" approach represents a strategic differentiation.