How to mock the value of an ML solution

Sell your solution before training your model.

data science

Bryan Whiting


December 16, 2020

IBM once launched a demo of speech-to-text in the early day’s of the technology. But there was a twist: it was smoke and mirrors. A typist in behind the stage was translating and typing.

This is a mechanical Turk.

In data science, we can present results or dashboards to audience members before they see anything. If you see a big reaction, you know you found a number they cared about.

Go ahead, write the conclusion first. Fake the chart first. Then back out the code that could produce the insights they need.

Sometimes, unique insights appear along the way. But most of the time you can properly gut-check whether someone will care about what you’re doing. Save time. Work backward.