Key Takeaways
- • Andreessen Horowitz argues that roughly $140 billion spent annually on market research is dominated by human-driven consulting and labor, with software representing only a small fraction of spend, creating an opening for AI to shift budgets into software.
- • The a16z analysis cites valuations of Gartner and McKinsey at $40 billion each, versus Qualtrics at $12.5 billion and Medallia at $6.4 billion, as evidence of the labor-versus-software gap.
- • AI-native research firms are moving beyond AI-assisted surveys to generative agent simulations, with startups such as Simile and Aaru (which announced a partnership with Accenture) building always-on synthetic populations.
- • a16z says success will be judged by a "good enough" accuracy threshold, with many CMOs reportedly comfortable with outputs at least 70% as accurate as traditional consulting, since the data is cheaper and faster.
- • a16z contends AI-native companies are better positioned than legacy firms to own both the data and simulation layers, and that distribution and adoption will determine which startups dominate.
Summary
Andreessen Horowitz (a16z) contends that market research, a field it estimates at $140 billion in annual spending, has been constrained by slow surveys, biased panels, and lagging insights. The firm argues that software captures only a small share of that spend, pointing to valuations of Gartner and McKinsey at $40 billion each compared with $12.5 billion for Qualtrics and $6.4 billion for Medallia, and noting that these figures reflect only external spend.
According to the a16z analysis, early AI-native players are already using speech-to-text and text-to-speech models to conduct autonomous video interviews and apply large language models to analyze results and generate presentations. These startups are reportedly signing large deals and capturing budgets historically allocated to research and consulting firms. Most still rely on third-party panel providers, however, which a16z says limits pricing control and differentiation.
The more significant shift, in a16z's view, is toward generative agent simulation. Drawing on the Stanford-linked paper Generative Agents: Interactive Simulacra of Human Behavior, the firm describes simulated populations of LLM-driven agents that can be queried, observed, and experimented with. Its illustrative example involves a beauty company simulating 10,000 gen Z and millennial consumers in France ahead of a skincare launch, seeded with reviews, CRM data, social listening, and purchase histories. Enabling techniques cited include persistent memory architectures, retrieval-augmented generation, agent chaining, and fine-tuned multimodal models.
a16z notes that no established benchmarks exist for evaluating these systems, so companies are defining their own metrics. The firm argues that perfect accuracy is not the goal; rather, a "good enough" threshold matters, and it says many CMOs it has spoken with accept outputs at least 70% as accurate as traditional consulting. It cautions that startups will need to refine products as benchmarks emerge and pricing rises, and that over-engineering for theoretical accuracy risks stalling companies in pilots.
The article concludes that AI-native firms are structurally better positioned than legacy players, whose workflows were not built for automation, and that distribution and deep integration will determine market leaders. a16z's own investment team, Zach Cohen and Seema Amble, invited builders in the space to get in touch. The piece is an opinion and investment-thesis essay from a venture firm, so its market-sizing and adoption claims should be read as the firm's perspective rather than independently verified data.
According to the a16z analysis, early AI-native players are already using speech-to-text and text-to-speech models to conduct autonomous video interviews and apply large language models to analyze results and generate presentations. These startups are reportedly signing large deals and capturing budgets historically allocated to research and consulting firms. Most still rely on third-party panel providers, however, which a16z says limits pricing control and differentiation.
The more significant shift, in a16z's view, is toward generative agent simulation. Drawing on the Stanford-linked paper Generative Agents: Interactive Simulacra of Human Behavior, the firm describes simulated populations of LLM-driven agents that can be queried, observed, and experimented with. Its illustrative example involves a beauty company simulating 10,000 gen Z and millennial consumers in France ahead of a skincare launch, seeded with reviews, CRM data, social listening, and purchase histories. Enabling techniques cited include persistent memory architectures, retrieval-augmented generation, agent chaining, and fine-tuned multimodal models.
a16z notes that no established benchmarks exist for evaluating these systems, so companies are defining their own metrics. The firm argues that perfect accuracy is not the goal; rather, a "good enough" threshold matters, and it says many CMOs it has spoken with accept outputs at least 70% as accurate as traditional consulting. It cautions that startups will need to refine products as benchmarks emerge and pricing rises, and that over-engineering for theoretical accuracy risks stalling companies in pilots.
The article concludes that AI-native firms are structurally better positioned than legacy players, whose workflows were not built for automation, and that distribution and deep integration will determine market leaders. a16z's own investment team, Zach Cohen and Seema Amble, invited builders in the space to get in touch. The piece is an opinion and investment-thesis essay from a venture firm, so its market-sizing and adoption claims should be read as the firm's perspective rather than independently verified data.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.