Surajit Ghosh

That's one good way of doing it.

And if you cannot have the luxury sometimes of doing A-B testing because everyone is having high appetite, give me the product, I don't want to sit aside.

Then you do some sort of causal models, like we say.

So you kind of look at what would have happened if the model was not there.

And then you predict that.

And since the model was there, something else happened.

The difference between the two is the incremental value the model is creating.

A-B testing is more accurate.

The causal models, the other one, like I said, which we call time series models, a little bit less accurate, but directionally, both give you the sense that, yes, it's working.

In that case, we will move on to something else because it means it's already optimal.

Then we say, good, check that.

Now let's move on to something else.

But we need to just make sure that the process is still running optimally.

So time to time, you keep doing A-B testing anyway, every six months or whatever the timeframe is, just to make sure that it's still relevant.

Do you measure something like that?

It's harder to measure.

One way is NPS score, which I said net promoter score.

Are you really happy with the product?

Has it changed your life?

That gives you a good indication.