New AI-assisted research separates customer needs from technologies to reveal a different structure beneath the venture economy.
Listen to this article:
When AI investigates AI, it’s like a microscope focused somehow to look at itself. There is certainly something self-reflexive about using a general-purpose technology to measure just how general-purpose that technology has become, but it’s also evident that AI enables human researchers to tackle problems and see patterns that would be difficult to measure at scale without it. That is the premise of the new working paper “Beyond a Technology Lens: Characterizing the Problems Entrepreneurs Solve,” co-written by a team at the HBS AI Institute. They used AI to analyze nearly 90,000 U.S startups, constructing a novel problem-first view of entrepreneurship. Their findings show both AI’s extraordinary reach and the limits of treating technological capability as destiny.
Key Insight: AI Can Make Startups Measurable
“The problem statement is solution-agnostic: it describes the problem the startup solves for its customers, and the startup’s solution is extracted separately.” [1]
Instead of beginning with what a venture sells, the technology it uses, or the industry it occupies, the researchers reorganized the startup economy around the underlying need (business problem) each company is trying to address. Then, for each of the startups in their dataset, they used GPT-5.2 to read three kinds of public material: the company’s website, third-party coverage such as press releases and news articles, and PitchBook’s company description. The AI model extracts three fields that define the problem: the customer segment (who experiences the problem), the life/business context (the situation in which the problem arises), and the customer need (the progress the customer is trying to make). The next stage turns language into structure, finding patterns that best explain each startup’s problem statement. The result was 265 distinct, coherent problem clusters, such as last-mile retail delivery, space and defense operations, and urban shared electric mobility. To confirm that AI captured reality, the research team conducted 331 structured validation interviews with startup founders. Across the 1,324 individual components reviewed in those interviews, founders accepted the model’s wording unchanged 71.8% of the time, and just 1.7% of the model’s wording contained information that founders rejected as hallucinations. Even when founders changed the wording, their revised versions remained highly similar in meaning, giving the researchers evidence that the automated problem statements were reliable enough to serve as the basis for further analysis.
Key Insight: AI is Everywhere
“This finding shifts attention from the breadth of tasks that AI can perform to the breadth of problems it can help entrepreneurs solve.” [2]
Once startups are organized by problems instead of products or industries, AI emerges as astonishingly pervasive across the entrepreneurial ecosystem. It appears in 235 of the 265 problem clusters, or 89%. That doesn’t mean that 89% of startups are AI companies. The finding is about distinct customer problems. AI shows up in ventures addressing many unrelated needs rather than remaining confined to one industry or application. Only cloud and computing infrastructure reaches farther, appearing in 240 clusters (91%). But the authors draw an important distinction. Cloud and AI achieve their breadth in different ways. Cloud supports the underlying computing environment used by businesses pursuing all kinds of customer needs, so its ubiquity partly reflects the infrastructure modern digital companies share. AI can sit closer to the intervention itself: the same broad family of capabilities can be applied to problems ranging from workplace automation to tutoring and media production. Its reach therefore suggests that AI is being repurposed across remarkably different kinds of customer demand. Capital has noticed: deploying and operating enterprise AI at scale emerged as the study’s top problem cluster by funding, attracting $295.8 billion, or 10.6%, of all disclosed capital in the sample.
Key Insight: Technology Expands the Possibilities, but People Direct the Effort
“The problem space is anchored by slow-moving human and organizational needs, against which technological change is better read as a shifting supply of solutions.” [3]
If technological breakthroughs dictated entrepreneurial activity, the newest capabilities would be expected to generate most of the problems startups tackle. The data complicates that view. The authors classify 180 of the 265 clusters (67.9%) as fundamental needs that predate a particular technology, regulation, or social shift. Those clusters contain 66.8% of the firms and include persistent challenges such as treating cancer, hiring the right candidates, and improving agricultural yields. Seen through this demand-side lens, technological change looks less like a continual generator of new needs and more like a force that repeatedly expands and reshapes the set of solutions available to needs that were already present.
Why This Matters
Most AI strategy is written in the technology’s vocabulary: adopt the models, add your connections, and find use cases. This research argues for inverting the order. If AI is meaningfully present in 89% of the problem space, “we are doing AI” stakes out no position at all. It says about as much as “we use electricity.” What differentiates is the problem you pick and whether you can credibly own it. Only then should AI compete with other possible solutions. A problem-first approach gives executives and leaders a clearer basis for comparing opportunities across functions, matching domain expertise to implementation, and deciding where investment can produce measurable value. AI may be general purpose. Execution still depends on being specific.
References & Notes
[1] Bernstein, Shai, Mimi Chen, Jacqueline N. Lane, and Zilin Ma, “Beyond a Technology Lens: Characterizing the Problems Entrepreneurs Solve,” Harvard Business School Working Paper, No. 27-016 (August 2026): 10. https://dx.doi.org/10.2139/ssrn.7339099
[2] Bernstein et al., “Beyond a Technology Lens,” 4.
[3] Bernstein et al., “Beyond a Technology Lens,” 27.
This article was created through a human-led, AI-assisted process.
Meet the Authors
Shai Bernstein is the Steven R. Fenster Professor of Business Administration at Harvard Business School.
Mimi Chen is a Predoctoral Research Associate with Shai Bernstein and Jacqueline Lane at the HBS AI Institute.
Jacqueline Lane is an Assistant Professor of Business Administration at Harvard Business School and co-Principal Investigator of the Laboratory for Innovation Science at Harvard (LISH) at the HBS AI Institute.
Zilin Ma is a Postdoctoral Researcher at the HBS AI Institute.