A Signal Taxonomy from a Noisy Dataset
Founder quality signals for investment research, designed and refined through five iteration cycles
All case studies are shared under NDA for portfolio review purposes only. The platform serves investment professionals conducting research, deal sourcing, and portfolio monitoring. Exact metrics values are withheld and replaced with proxy phrasing due to confidentiality constraints. Happy to walk through the details personally.
Problem given
The platform gave investment professionals access to a very large people dataset. A search for 'founder' returned everyone who had ever listed the title on their profile. Users drowned in undifferentiated results and had no way to distinguish quality from noise.
Problem framed
The filtering tools operated at the wrong level of abstraction. Users needed to evaluate founders on composite quality signals (pedigree, expertise, network credibility, warm paths), but the interface only offered raw profile attributes. Bridging that gap required a new layer between the raw data and the user.
Constraints
Noisy source data (professional profiles with unreliable titles) with over 700M profiles; signals had to be pre-computed and tagged, not computed at query time
My role
Owned taxonomy design, signal definitions, cross-surface integration, iteration process, stakeholder alignment (no PM on this project)
What shipped
Fourteen signals organized into four categories (Founder Background, Industry Expertise, Strategic Network, Warm Pipeline), deployed across four platform surfaces, combinable with existing filters, refined through five iteration cycles
Impact
People screener transformed from results page into a genuine filtering tool; users added filters on profiles with signals as the base filter; organic extension requests validated the framework; taxonomy accommodated custom firm-specific signals naturally