When we set out to build our quick valuation estimate, based on only 10 questions, we didn’t want it to be shallow, static and simply an average of our data that would not be based on any theory and could not be analysed. It is one of our core tenets at Equidam that data is of critical importance, but it should always be paired with theory. When data is analysed alone, and put on a pedestal, there is a strong risk of us just blindly contributing to peaks and troughs in the market, rather than improving conditions for the whole early stage funding environment.

The focus of the development of the 10 question tool was then in trying to extract as much information from the limited input we receive from the user, and augment it as much as possible to try to perform a valuation that is as rigorous as it can be.

That’s why the way the calculation of that works is split in three processes:

Let’s see how each works.

The first step in expanding the data of the company under analysis is to find other companies in our proprietary database that look like it. For this, we’ve trained a machine learning algorithm that roughly follows the theory of “Collaborative Filtering”. We’ve used part of our data as a training set, with the aim of defining the importance of each of the 10 questions in defining companies that are similar to each other.

With the training done, we have an algorithm that knows which are the most important factors and can identify companies and company-observations that are the most similar to each other

Once we’ve identified the most similar company-observations, their data is used to fill in for the limited data of the company that we are trying to estimate. This is done differently for financial and qualitative information.

For financial information, we gather from the 10 questions the current and future revenue of the company. This allows us to have a starting point that is not only useful to identify similar companies but to also tailor the financial projections to the specifics of this company.

We use the revenue growth as a guideline for all the other projections, that are determined as ratios from the comparable companies.

For the qualitative information instead, we use the median or average data of the comparable companies when appropriate.

Companies that perform the initial valuation estimate with us don’t get a reduced version or a simplified version of our algorithm, they get the full, most up-to-date one. This is by design, we want the estimate, which is necessarily limited by the fact that we know only superficial information about the company, to not be limited further by applying old or generic data.

Thanks to steps one and two, in our process so far we’ve been able to augment the data of the company and have estimates of what it would have answered if it underwent the full valuation with us. Our algorithm then doesn’t need to run in a partial or different capacity, which is convenient for us from a development point of view, but it also serves users who can benefit from the latest updates both to our data and to our methodology.

Valuation is, in a sense, the art of averages. Since every price is determined in the context of other prices, the valuation of a specific company cannot differ too much from the average. However, the whole goal of valuation is to determine, with as much accuracy as possible, how far it departs from the average.

This activity is based inevitably on the specific information of the company, that, by definition, makes the company different from the average.

So again inevitably, an estimate that knows so little about the specifics of the company cannot be extremely specific. What we really aim for here is to preserve as much information value and bias it as little as possible by adding and calculating its valuation.