How to quantify the business value of machine learning?

5 pointsThePhysicist3 comments

Dear fellow data scientists / ML experts: How do you explain the impact and value of your work to the business people in your organization? Do you have to explain your work to them, do they check it and do they understand what you're doing?

View on HN

That’s all contextual. If you work in adtech you are probably linking your work to the targeting performance.

If you’re in a research function it’s probably patents trade secrets or publications

Those for me are the two examples on opposite ends of the spectrum.

And usually yes I need to explain value but it really depends on what level that discussion happens. If I’m team lead looking for budget it’s gotta be measurable, if I’m a IC in a team it’s more did I help the team he successful

Thanks! Do you track the performance over time or do you even have continuous measurements for this?

I gave a talk on this topic! See https://www.infoq.com/presentations/opendoor-machine-learnin...

The tl;dr is to try to incorporate the ML models into a business simulation, so instead of maybe-abstruse metrics like AUC and log loss, you have "businessy" metrics like margin and volume and can better directly tie the ML models to business value creation.

Reply on news.ycombinator.com