[simonhamp]

If you’re going to ask what someone _will_ do, it’s best to keep it to a simple binary decision and always remember that that’s only representative of a single moment in time/the current customer service cycle for that individual.

As this article identifies correctly, the more important question is the harder-to-quantify “why?”.

Armed with this, the fluctuating and heavily biased score will have more meaning and context. Using that data to improve the service/product for the next cycle may do more to realistically reflect sentiment change than any weirdly random maths.

Of course, this still ignores what someone actually does do. But it’s possible to track certain cases of referral and actually to encourage trackable referrals using the correct approach. This is a more solid measure (if possible) than any hypothetical gauge of future behaviour - again as the article points out.