-
598. Harsh Sahai, AI-powered Due Diligence
- 2025/02/03
- 再生時間： 50 分
- ポッドキャスト
-
サマリー
あらすじ・解説
Show Notes: In this episode of Unleashed, Will Bachman interviews Harsh Sahai, CEO and co-founder of Bridgetown Research, a company that has built an AI tool and he talks about it in this episode. Harsh previously worked at McKinsey, where he focused on commercial due diligence. He also ran a machine learning lab at Amazon, where they researched sequential decision-making algorithms. AI Pricing Algorithms and Convex Optimization Harsh talks about his work at Amazon where main use cases were pricing products, as people tend to remember old prices and make decisions based on what they remember. For example, planning the sequence in which to launch products or introducing new shows on Prime Video could be done in a multi-step planning process. Harsh talks about his background in convex optimization, which is a mathematical model that can be used to represent various outcomes. Convex optimization is often used to model price versus volume, and it helps in making more sequential decisions for more than just pricing. Bridgetown Research Explained On founding Bridgetown Research, many of Harsh’s former colleagues joined him in the mission to build tools for the consulting industry and more. Bridgetown Research developed a platform that automates data collection and analysis, allowing them to curate these analyses and deliver value to clients. The firm developed software products that can conduct interviews at scale at a fraction of the cost, run 300 common analyses, evaluate approximately 10 decisions, and work alongside clients to build interactive documents. The firm primarily serves investors in the software industry, similar to McKinsey due diligence. Automating Consulting Groundwork They use AI agents to conduct interviews, breaking down high-level questions into sub-questions that can be answered by the AI agents. The agents then map the best sources of data for each analyze, such as Gartner or Forrester, and compile secondary research. The AI agents are integrated with a few expert networks, which they recruit on the company’s behalf. They have a fully adaptive conversation, similar to a consultant's conversation, and then parse out the analysis to answer the main questions. The cost of these interviews is lower than a normal human-to-human interview because they can do it on their own schedule. Harsh also discusses the benefits of owning a research platform for consultants. They have researched this topic extensively and have 1000 interview transcripts of both people who hired a consultant and like consultants. The platform offers voice-based conversations, text prompts, and interactive screens for additional context. Using AI Agents in Surveys The AI agent in the discussion is similar to a traditional survey, but it allows users to answer questions directly on their screen. It can also embed multiple choice or ranked sorting questions, and can follow a different chain of questioning depending on the user's response. The agent constructs a hypothesis based on secondary research and uses adaptive questions to collect enough data to either prove or disprove these hypotheses. If it disproves the hypotheses, it goes back and looks at all transcripts to come up with new hypotheses and start collecting more data. One of the reasons for the cost efficiency is that, unlike regular surveys, the AI agent doesn't ask the exact same questions, reducing the length by about 20 to 25% once statistical conviction is reached. This flexibility allows for discounts from the person taking the interview, as it's extremely convenient for them. Examples of AI Agent’s Responsiveness The agent's responsiveness works by comparing the user's responses to previous answers, such as asking about the main reasons they chose a particular software versus another. The agent then moves on to the next question based on the user's response. Harsh offers a few examples and verifies that the agents have received positive feedback from experts who are willing to interact with the voice agent, but they also interviewed people with slightly different profiles than consultants at McKinsey. More Information about the AI Tool The AI tool used in this discussion is a work in progress that aims to provide insights into competitor archetypes and their strategies. It is designed to be more efficient than traditional human interviews, as it can gather data from mid-tenure professionals and frontline users closer to the business operations. This approach allows for a more comprehensive understanding of the business, reducing the need for frequent human interviews. The tool is fully scalable, allowing for 100 interviews in three days, which is the time it takes to recruit individuals rather than the time it takes to interview them. This allows for the creation of compelling projects within a week. Before the interview phase, the AI tool asks a set of questions and breaks them down into sub hypotheses. The tool then constructs sub questions...
activate_buybox_copy_target_t1