VentureFlow Builds AI Technology for Faster and Smarter Venture Investing
In the last decade or so, building and investing in companies has become much more data-driven. Founders usually start with qualitative data which they collect from talking to customers and other stakeholders. These data points are critical in order to be able to specifically address the problems they are going after and are likely the inputs that will inform their initial data strategy.
As soon as a startup gets its first customers who use the product or service, their usage and feedback will probably lead the product development in unanticipated directions. By monitoring usage and validating feedback, startups can ensure that the product development is actually beneficiary for eventually gaining product-market fit. To be able to do this iterative process, founders need to make sure that they create a culture and processes that support such an approach.
“All models are wrong, but some are useful” — George E.P. Box
On the other side, before the rise of more sophisticated data analysis, most investors and financial analysts simply relied on gut feeling and/or theory. Conventional investment models have gone a long way toward the development and “maturation” of financial markets and their creators have rightfully earned academic distinctions. But truth be told, probably all models have at least one identifiable deficit and invariably, it can be found in the assumptions they make. This simply because the model’s assumptions don’t necessarily hold up in the real world. Nevertheless, financial models are truly useful as a source of information and indication and have greatly helped to build today’s economy. However, today, we have opportunities for more granular modeling that is attuned to the real world and much more prescient than probably every previous approach.
VCs spend 22 hours on 1st-round due diligence per company — Harvard Business School
In this context, VentureFlow builds a set of automated due diligence features that accelerate and support early-stage investors, especially in the seed stage up to series B and later, in their decision-making process in order to save time and money and significantly enhance the quality of portfolio companies (see Fig. 1). VentureFlow is committed to pursuing further valuable research projects in collaboration with the University of St. Gallen. Other research is being conducted in collaboration with the University of Lucern and Fachhochschule Nordwestschweiz.
At this stage, the VentureFlow’s solutions consist of an investment risk score (survival prediction), investment attractiveness score (valuation prediction), digital footprint analysis, team diversity score, competition analysis (industry, market size/share, technology, geo-region, funding stage, backers, and growth rate), crowd-curated analytics, legal analysis, and a business setup analysis.
For the assessment of investment risk and attractiveness, the team has adopted an array of supervised machine learning methods, such as support vector machine (SVM), decision tree, random forest and voting methods as well as various software and platform tools for pre-processing and machine learning such as Scikit-learn, Python, TensorFlow, Pandas, and Numpy that analyze founder and company data.
Natural language processing (NLP) is utilized to extract correlations between people and startups. VentureFlow will source an available taxonomy, if a suitable one cannot be found, and create a specialized linguistic taxonomy for investment risk score and investment attractiveness score processing. Other fields of exploration for the digital footprint analysis are deep learning methods such as convolutional neural networks (CNN), recurrent neural networks (RNN), autoencoders, backpropagation, and hyperparameter tuning.
In addition to these analyses, VentureFlow taps into crowd wisdom and crowdsource legal vetting from external experts which is added to the company’s partners who provide services such as detailed due diligence reports by professional analysts.
To train and test the algorithms, VentureFlow uses tools such as Coggle for data plotting and Alteryx for data analytics and Crunchbase as an initial dataset. Other partners such as Pitchbook, Smith and Crown, and Weiss Rating Agency are in negotiation. The Crunchbase dataset was sufficient for building a prototype algorithm based on a dataset of 700K samples, each sample represents one startup company and includes multiple data points, including financials, operation status, the managing team, regional correlation and technology, and market adaptation. The already created algorithms can be customized to meet the individual needs of clients. Some considered customization methods are feature engineering, transfer learning, and parameter tuning.
The tools built and those yet in the making will be integrated on an investment platform where deal flow can be created and curated. To make interactions between investors and founders easier and more organized, VentureFlow also offers a virtual data room that allows for indexing, sharing access to documents, export, and in-file Q&A.
If you wanna learn more about VentureFlow, feel free to reach out to XJ who’s the founder of VentureFlow: xiaojean.chen@ventureflow.ch.
