The challenges of no-code ML

Using machine learning to tackle complex problems has never been easier. Today, numerous tools enable you to utilize or build models based on your domain-specific data with minimal ML expertise and almost no coding skills. However, as one aspect of building ML-based products becomes easier, another becomes the bottleneck.

Imagine you have a dataset that (partially) captures user behaviors and problems, and you wish to apply ML to enhance your services. Numerous tools exist to help you understand your data. These tools can identify features that are non-normalized, missing, or cyclical. They can also build models using AutoML. However, what these tools often don’t address, and what many teams struggle with, is deciding what to build  and how to effectively meet customer needs. This isn’t merely an engineering challenge; it’s fundamentally a product challenge.

Before you even begin building models, you might use analytic tools to identify patterns in your data, hoping to discover insights that could help you define specific subproblems. With the rise of no-code tools, you can effortlessly extract statistics and trends that point you towards potentially valuable directions. However, as the saying goes, “it's useless to have the right answer to the wrong question”. Much like a search engine, the usefulness of these tools heavily depends on the quality of the questions you pose. Learn to ask better questions.