Venture portfolio · Earlier product work
Turning portfolio knowledge into customer opportunities
Built a GPT-3 agent that classified portfolio-company websites and helped founders discover potential customers across the portfolio.
The operating problem
Founders needed help finding customers within a venture portfolio. Relevant information lived across company websites and the knowledge of partners and founder-support staff.
Discovery depended on knowing whom to ask and how to describe the desired connection. The opportunity was to make company information directly searchable so founders could identify relevant businesses themselves.
What was built
Built a GPT-3-powered agent that classified portfolio-company webpages and supported conversational customer discovery. A founder could describe the type of customer they wanted to reach and find potentially relevant companies within the portfolio.
The first delivered milestone focused on customer search. Interface feedback and request tracking supported iteration. Investor search and broader-network expansion remained roadmap items.
Strategic approach
The problem selection drew on experience advising 100+ portfolio companies on product, GTM, fundraising and team building. A recurring advisory need became a self-service discovery capability.
The design aimed to reduce dependence on partner-mediated matchmaking while making the portfolio’s existing relationships easier to discover. It addressed the search step rather than claiming to automate relationship building or sales.
Technical depth and iteration
The implementation combined webpage classification, language-model interpretation and a conversational interface. Unstructured company descriptions became inputs for business discovery.
Testing exposed weak lower-ranked matches and access issues. Search quality and the ability to complete an introduction were separate concerns: working customer discovery did not establish reliable end-to-end introduction facilitation.
Delivered capability
Founders gained a way to search the portfolio for potential customers. The implementation demonstrated an early practical use of GPT-3 in a commercial workflow.
Time saved, completed introductions and resulting sales were not measured. The 100+ figure describes the broader advisory scope, not the agent’s user count.
This work was completed before Measured Intuition as part of Naman's work at another organization.
Discuss a workflow worth changing.
Start with the operating constraint, the current process and the business result that would justify changing it.