You've decided Machine Learning (ML) can help your customers? Great! But ML can be difficult, time consuming, and expensive. Agile practices can reduce these costs and improve outcomes in the same way they've always helped: by making sure we're building the right thing every step of the way.

We've identified, and our presentation will speak to, the following anti-patterns for machine learning and agile approaches that avoid these anti-patterns. We value: * Iterative over incremental * Always be releasable over "toss it over the wall" at the end * ML as a means over ML as the end goal

For each anti-pattern, we will present several techniques to mitigate it.

Avoiding Incremental Development In Favor Of Iteration * Find 'minimal learnable concept' and get a complete ML pipeline quickly * Improve your ML pipeline from data preparation to deployment, don't waterfall it * Realize when something isn't feasible early on to avoid wasted effort

Integrate ML In The Product From The Start, Not At The End * Avoid answering the wrong question by focusing on outcomes and stories * Prevent overwork by only doing what's needed to get the desired result * Experts may identify easier-to-solve adjacent problems that have comparable outcomes for users

ML is a feature like any other - it is a means, not an end * Identify when it is not the right tool for the job * Know when good enough is good enough * ML stories should be prioritized, have story points, etc.