5 Steps for Building Machine Learning Models for Business

Summary

The article presents a five-step workflow for deciding when and how to build machine learning products in a business setting. It recommends delaying machine learning for new products, establishing data-collection foundations, and testing simple heuristics as baselines before investing in a model. For an initial model, teams should favor simple algorithms, limited feature sets, and off-the-shelf solutions where appropriate, then measure business-aligned impact before optimizing. The workflow also calls for planned model iteration and explicit comparison with existing versions, while accounting for prediction stability when changes could affect user trust. The authors say they used these practices at Shopify to help build and scale machine learning products and to productionize models intended to drive impact.

Context

Technical guidance on building and updating machine learning models often does not explain how model development fits within a business context. The article addresses when machine learning is worth investing in, how to get started, how to measure business impact, and how to update models without breaking the product.

Approach / What changed

The article proposes five practices: assess whether machine learning is timely, keep the first solution simple, measure before optimizing, plan how to iterate and compare model versions, and consider model stability over time. It recommends heuristics as baselines, simple interpretable models and feature sets, off-the-shelf models where suitable, business-aligned metrics, and checks for prediction changes between versions.

Takeaways

• For new products, the article recommends prioritizing product launch, user engagement, feedback, and data-collection foundations, while initially delaying machine learning; proven products may be better candidates for an algorithm.
• Simple models such as shallow decision trees, linear regression, and logistic regression can speed iteration, improve interpretability, and reduce maintenance costs; complexity can be added later if needed.
• Model evaluation should include business-aligned metrics, comparisons with existing versions, and stability over time because changing predictions can frustrate users or undermine trust even when aggregate accuracy improves.

Related reading