BFP #OpenLP Series (10) — In the first part of this article I discussed the use of AI and ML scoring models for sourcing and filtering investment opportunities.

In venture capital, we are also seeing an increasing number of managers using scoring methodologies to make their final investment decision. Such scoring models basically consist of creating a weighted average of the expected value of an investment (or a proxy of it), which can be built in very sophisticated ways.

**Three Problems**

We find these scoring systems very intriguing but have identified three major problems in them:

1. The first problem with these frameworks is that, while having the purpose of enhancing decisions with a rigorous quantitative method, they can be subject to completely arbitrary parameters, in terms of both independent variables as well as their weights.

2. The second major problem is that, even when trying to build the algorithms based on empirical data, the data is mostly irrelevant due to the dynamism of the environment. This changes the relative importance of a parameter versus the others (and consequently its weight) quite rapidly over time.

For instance, having a certain degree (e.g. Computer Science from an Ivy League University) in the early 2000s probably was more important than what it is today for building a successful startup. Hence, if a parameter of the algorithm is “type of degree”, its weight in the average today should be very different from what it used to be. I also discussed this point in the first part of this article.

Consequently, if there was a hypothetical algorithm to predict the success of a startup, this should change significantly at the same pace of the evolution of its parameters in the real world. Clearly, this leads to a lack of relevant data, which makes it necessary to use arbitrary parameters as per the previous problem.

3. The third problem we see is that, independent from the reliability of the data, the scoring models we found in the market do not at all consider the interdependence of variables (related to both the startup and its external environment). The parameters should not be considered as single elements, but as a whole set of organized resources (similar to biological organisms) and the real key is the “strategic fit” between the way such resources are organized and their goal, which is extremely difficult to be written in an equation format.

**VC Scoring Models: The Example of Chess**

To better explain problem 3, let´s take an example with chess (yes, I am watching Netflix´s *The Queen´s Gambit *too), assuming that we would like to build a scoring algorithm to predict the probability of a player winning a game that has already started.

Since the players have already moved their pawns and the game has started, the remaining pawns from both sides (black and white) would be positioned in the exchequer.

Now let´s try to build our algorithm, for which we would need to define a) what the input is and b) which weight we want to assign them.

Apart from the skill of the players, such algorithm could be built, for instance, in this way:

a. The independent variables could be the number of pieces per type (pawns, horse, tower, etc.) left in the exchequer for each player

b. The weights could be a score that we assign to each piece according to its strength (e.g. 1 for pawns, 3 for towers, 10 for the queen)

We would now have an algorithm, built rationally, that we could use to predict if a player will win the game.

For the readers who are familiar with chess, it is already clear that this algorithm would not go very far, because it is missing two crucial elements:

· The strength of a piece is not due only to the strength of the piece, but also due to its position in the board.

· Even more importantly, such strength would also be dependent on the position of the piece relatively to the position of the other pieces, which is the result of strategy and is far more important than what and how many pieces are on the board.

This is what VC algorithms are missing: In the same way that our chess algorithm would need to substantially adapt to the individual game to capture the strategic position of the pieces as a set of organized resources, a “VC algorithm” would need to substantially adapt its parameters all the time to the individual investment cases, which would make impossible to find a standard model.

Weights are difficult to be standardized because the *“right”* weight of a certain variable is completely different case by case. In chess, such a difference is due to the position of the pieces, while in startups it is even more complicated because:

a. The number of variables (and their constraints) related to building an organization is enormous and includes exogenous factors

b. The number of ways such variables can be combined is even bigger

c. The variables and their combinations are dynamic and change rapidly

d. Differently from a chess game, which can be described as a completely deterministic process where probability is not involved, real life includes an element of randomness

The way players win in chess is by adopting a dynamic strategy to move the pieces coherently. At the same time, understanding who is going to win in chess and in startups is about recognizing a dynamic pattern of coherence in the different elements that make a set of organized resources position themselves organically towards a goal.

I have not yet seen an algorithm that can do this for VC, but I am optimistic about the future (indeed, computers can be quite amazing at chess and *“chess algorithms”* work very well).

Meanwhile, think of VC as a checkmate: it takes time, passion, you may lose some pawns along the way, and whether you win or lose, there is always a new game to play.

—

*About Marco Cesare Solinas**Marco is an Analyst at Blue Future Partners, where he is responsible for sourcing and analyzing new investment opportunities. He is passionate about Technology and Venture Capital and making an impact with investments*. *He focuses on both direct and indirect investments.**Previously, he has built an international and multicultural background across Italy, US, Germany, Turkey and Malaysia.**Marco holds a CEMS Master’s in International Management and a Bachelor´s in Economics and Finance from Bocconi University.**Linkedin** — **Twitter** — **Medium*Blue Future Partners