# 8 — Continuous Benchmark in ML projects

The importance of enabling the competition to improve ourselves and the product.

Data projects have a feature that make them very powerful, and it is that decisions about their evolution are strongly based on metrics. If our KPIs are well defined and we have not left A/B testing in production as complex as to be part of our technical debt, we now face a clear path to benchmarking and making those solutions compete.

Mature machine-learning projects lower the barrier for the introduction of new models enabling the competition among different solutions. By granting incentives to the teams that generate models to keep innovating and improving, we contribute in turn to enhancing the final performance of the system. In addition, since we have clear metrics, there is an instant mechanism for transmitting information about the best model.

Encouraging these competitive spaces has clear business advantages and may be less complex than it seems. Let’s explore some ideas on how to generate them:

• Getting two approaches proposed by the same team to compete and measure them in production.
• Finding another company team with experience in machine learning that can make a model using a different approach.
• Creating an internal challenge (like a Kaggle) within the company, with prizes!
• Creating a public challenge where everyone can take part (this is not always so simple, specially for data privacy aspects).
• Benchmarking with an external company that can make a model for us.

It is important to point out that, by building these spaces, we are not only making a positive impact on the business, but also on the team. Ongoing challenges avoid stagnation, generate incentives to improve ourselves, keep us away from the bias of “falling in love” with our solutions, and even avoid a certain intellectual lethargy or laziness. This is what we experience every day. Or don’t we find it motivating to compete with talented rivals to push us to improve?

This approach has resulted in the improvement of business metrics and of ourselves as professionals, and in the capitalization of the company. This is valid in different types of projects in life, and in business in general, but it is especially applicable to machine-learning projects.

Now it’s your turn: Are you taking advantage of this opportunity?

We are near the end of this series of articles, only two more flight checks ahead. Keep up & fly safe!