How the Unicorn Nest scoring algorithm works
Unicorn Nest is a startup-centric service. This means that a founder doesn’t need to guess what funds might invest in their startup. We just ask them to tell us about themselves via our questionnaire and our algorithm does all the matching.
All the matching we do is based on the data we have collected about funds. Each fund in our dataset has data that describes it in dozens of categories: investment stages, portfolio company indicators and their geography data, industries of strategic focus/success and their neighboring industries, founding rounds where funds participated, general fund performance data, fund key persons geography data, etc.
After we have collected all the inputs from the user, they are passed to our matching algorithms. All 20+ of our current scoring rules are built on this principle:
- Our fundraising expertise or user feedback generates a request for a new matching rule. The essence of these rules is to associate a metric that is understandable for a startup with a metric that is understandable for an investor. For example, revenue and company valuation.
- Analysts formalize the request in the form of a script, which is adjusted after 1000 test runs of this rule based on real cases: recently raised rounds.
- Before the rule is integrated into the general ensemble, we run a system for calibrating the weights of all our rules, which works on the basis of logistic regression. Thanks to this, we can give real significance to each rule and make rules of completely different dimensions work in a coherent ensemble, for example, geography and the number of founders in a startup.
Our algorithms operate on a scoring system, unlike the filtering system used by our competitors. The advantage of this approach is that it more accurately describes each investor and takes into account the relationship between its various parameters, and does not simply record the fact that there is a client request in the breadth of the investor’s portfolio
After all the rules have worked, the user receives a list of investors that are sorted according to the principle of decreasing relevance. The maximum number of investors you can get is limited to 250 (we consider this number as the optimal for one mailing campaign).
Our scoring system is enriched with a number of additional features. For example, we take into account the size of the fund, since a large amount of assets under management allows the fund to expand its portfolio and defeat more narrowly focused (relevant) funds simply due to the volume of deals. We also impose additional adjustments to the nature of the fund, (for example, not all startup founders are ready to talk to corporate investors).
Despite the high accuracy of our rules, we still work with probabilities. On the one hand, this may seem like a disadvantage, since we do not take into account 100% of all important factors, the proposed funds may not be suitable for a startup founder based on a number of his individual preferences. But in general, this approach allows us to isolate patterns in the behavior of funds that are invisible to the human eye, and, most importantly, are based solely on the verified facts.