[syed99]

Hey! Co-founder/CTO of Tara here.

We take privacy around your data very seriously. We're thinking about the ML training model as a "walled garden", ie recommendations are based on your past sprint activity, effort load, tasks completed during a sprint, etc. and are exclusive to your organization.

In the future, if we decided to do benchmarking (for eg quick recommendations on how companies in your industry are running sprints), we would have a double opt-in. This would mean anonymizing the data, and providing the recommendations an opt-in. Basically, very similar to how google's autocomplete email recommendations work.

[viach]

Interesting, just had this idea - would it be possible to feed the model with wrong data and make it giving bad suggestions to competitors who opted in?

[syed99]

Ha! Yes to some extent it would but that would really mean either a) we decided to run recommendations using a sparse model or b) we never validated the data or kept a watch on models when they were trained/retrained. This type of manipulation is typically likely when you're much larger and have alot of bots involved on the platform.