Three things a venture capitalist must ask when a startup employs machine learning
During a pitch presentation, here are some important things to consider.
What exactly is Machine Learning?
Now let’s start with the basics. What is machine learning (ML) exactly? There are several interpretations available online, but they might be confusing. SAS, a data processing program, provides a simple definition:
Machine learning is a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention.
In other words, ML is a subfield of artificial intelligence rooted in the premise that machines can learn from data, spot trends, and make choices with little or no human interaction.
Ask the startup these questions
1. Do you use hybrid intelligence or go all-in on end-to-end machine learning?
What this is asking:
The phrase “hybrid intelligence” refers to utilizing a mix of humans and AI to finalize the selection of a target variable. When you question a company if it uses hybrid intelligence or entirely machine learning, you are attempting to determine the extent of reliance on AI to achieve optimum results.
Why is this an essential question to ask:
Imagine a social networking company that utilizes machine learning to predict which users require therapy. Assume that the suggested platform is entirely reliant on machine learning. In such an instance, it implies that no human intervention is involved in any process (besides historical datasets). On the other hand, if a startup uses hybrid intelligence, it indicates that it uses a combination of human evaluation and machine learning to identify who requires therapy. Of course, there is not only one correct answer here; only the VC’s choice for what constitutes a strong business plan. Nonetheless, the questions are critical in understanding the degree of AI application in the company’s fundamental reasoning.
2. Which machine learning algorithm(s) will this system employ, and why?
What this is asking:
This question inquires whether the startup has examined different solutions for performing predictive analytics in their business model.
Why is this an essential question to ask:
There is no one solution to an issue in machine learning. For instance, if a company’s business model uses ML to estimate a binary result (such as fraudulent vs legitimate transactions), the problem is most likely classified as categorical. Nevertheless, even in a categorical problem, several models could be used. Categorical algorithms include logistic regressions, K means clusters, and SVMs, to name a few.
To complicate matters further, data might be labeled or unlabeled, which substantially impacts the sorts of algorithms that work the best. Finally, as a general rule, the more straightforward the model is, the greater the bias. The more significant the variability, the more complicated the model. If precision is important to you, you may have to sacrifice speed.
I highlight this to emphasize that the purpose for posing this question is to determine how well the founder and team understand the data they are analyzing. Of course, comprehension of one’s information is essential for validating the business strategy, but it’s also crucial for the VC to understand what they are about to invest in.
3. Could you elaborate on your feature selection process?
What this is asking:
You can select several attributes for most machine learning procedures to utilize as model inputs. Feature selection minimizes the quantity of these input variables to ensure that the model performs optimally. By asking them to go through the feature selection process, you’re probing the thought process used to determine which factors matter most in estimating a result against those not considered necessary.
Why is this an essential question to ask:
Most datasets include many more features than would be required to develop a model. Adding too many superfluous elements can pollute and slow down the model while omitting other crucial ones might jeopardize its reliability. Using variables with a linear connection, on the other hand, might produce multicollinearity, which can have a detrimental impact on the analysis.
Discovering why and how the startup selects particular predictive analytics features is another important indicator of how well the team comprehends the data.
Thank you for your time! Feel free to contact me at ayaspencer.com if you want to learn more about machine learning and venture capital.
