Pentathlon Case Conclusion
It's a limit of promotional email activity. You should have answered, we don't have enough evidence to call it either way. Because neither piece of data tells us what effect email frequency has on business outcomes. Anakin Teto's survey evidence measures how consumers feel, but not how they act.
François Cabret's evidence measures business outcomes, but fails to show that more emails actually cause greater consumer spending. In practice, it is common for decision makers to draw the wrong conclusions from the kind of evidence I just showed you in this case.
Causality
- Probabilistic equivalence as magic in data science
Issue w/ non-random allocation and discerning what data shows and what the implications of grouping are:
- But the problem is, that's not how consumers got allocated into these three email frequency groups. In fact, they got allocated in an extremely unrandom way. Namely, that consumers who were better customers to begin with, through this funny association with the independence of the department's email policy, ended up being consumers who received more emails. So, the better a customer you are, the more departments you purchase from, and as a result, the more emails you get. So, if anything, the causality was actually going the reverse here. And as a result of this, this evidence does not in any way, shape, or form show that sending people more emails leads to a bad outcome.
Then remember that google doesn’t treat all customers the same. In fact, they’re biased in the customer’s stated preference.
asdf
Checklist
Dealer Cash example
Expanded checklist
Driver snowfall example where time tells the tale
In the pentathlon case, those who spent more caused themselves to receive more emails
Microsoft Social Engagement
Beauty World Omnichannel
- Retail usually poorer; income are a confound.
- You still might glean insights about customer behavior that could improve UX and increase engagement. So even if retail customers are poorer and converting them to online might not double profits (b/c they won’t spend 2x but just spend elsewhere) affecting behavior like that could still be a useful learning point.
Mind the leap between explanatory to causal
Confound example
