Customer Centricity - BMME144
Module 2
Targeting Customers
Recap M1
- you wanna be Customer Centric over Customer Friendly because in CC you find a
select set of customers to focus on > TARGETING
- CC is better because the key metric is ROI therefore you focus on LIFT instead
of pleasing everyone
- being CC means providing additional value only to the most special costumes.
- CF is hardly sustainable if focusing on all customers
- Good Data - A/B
- Bad Data - Churn >
- CC: estimate how our intention can impact objective set up for the campaign (e.g.
sales). To do this, need to estimate the LIFT
- To maximize LIFT you can count on non-monetary incentives. For example, you can
nudge customers, by making them feel empowered. the OTHER customers, that we
do not wanna focus on, we don't spend marketing dollars to keep them (negative
ROI)
Experiments in Marketing
Facebook Amazon Apple Netflix Google (FAANG): started idea of experimentation >
advantage of being online to gather more data on customers and can easily adapt web pages
to customers (website morphin: understand from customers what they like and adapt to
provide them)
- RCT: Randomized Control Trials
- A/B test: 2 conditions
- A/B/n test: 3 or more condictions
Why do we need experiments?
ebay case:
- free search vs paid search > paid search had a CTR of 80% BUT compared to what? >
you need a different condition (A/B)
- after Treatment condition, the CTR changes from80 to78% > is a 2% difference worth
the costs of intervention? NO > pay more than it delivers
- BUT if competitors do it, it might be worth it to attack back
Only compare groups when they are
the same > observed variables are
things that influence experiments
, but we don't actually see > observed variables can influence the DV
!! Confounder variable (they confuse) > “the devil of decision maker”: they influence both IV
and DV and you do not see them
There is a reason why you see the ad. if there is no reason, the marketing strategy is not
effective > condition of exposition is not random (> not very reliable)
Correlation vs Causality
- correlation = what will happen if nothing changes? (> we move away from correlation
to causality) ! you want to change not just to predict
- causality = what’s the next best action as a company? > you do not wanna know
what happens if nothing changes. you want to know WHAT to do so that you can do
better
- for normative insights
What can we do with experiments?
- test global change > not what we interested in
- A/B testing: know difference between A and B but does not give answers on
how to act accordingly > to test an average treatment tes: how the treatment
works on average
- we want to understand HOW DIFFERENTLY people react from one condition
to the other
- Bezos Type I vs Type II decisions > ?is the decision reversible?
Kohavi (to check experiments)
- optimize personalization > how can you segment the audience?
Lift-Base Targeting A/B Test and Randomization
1st rule of CC: Allocate your marketing resources in priority to your most responsive
customers > LIFT of an Intervention
- Profit LIFT: CLV if you target specific customers compared to the CLV if we do not
target and remove action cost > because you wanna know if more or less expensive
-
> Problem: You never observe both (targeted and not targeted) at the same time, as
people are exposed to 1 condition each
Module 2
Targeting Customers
Recap M1
- you wanna be Customer Centric over Customer Friendly because in CC you find a
select set of customers to focus on > TARGETING
- CC is better because the key metric is ROI therefore you focus on LIFT instead
of pleasing everyone
- being CC means providing additional value only to the most special costumes.
- CF is hardly sustainable if focusing on all customers
- Good Data - A/B
- Bad Data - Churn >
- CC: estimate how our intention can impact objective set up for the campaign (e.g.
sales). To do this, need to estimate the LIFT
- To maximize LIFT you can count on non-monetary incentives. For example, you can
nudge customers, by making them feel empowered. the OTHER customers, that we
do not wanna focus on, we don't spend marketing dollars to keep them (negative
ROI)
Experiments in Marketing
Facebook Amazon Apple Netflix Google (FAANG): started idea of experimentation >
advantage of being online to gather more data on customers and can easily adapt web pages
to customers (website morphin: understand from customers what they like and adapt to
provide them)
- RCT: Randomized Control Trials
- A/B test: 2 conditions
- A/B/n test: 3 or more condictions
Why do we need experiments?
ebay case:
- free search vs paid search > paid search had a CTR of 80% BUT compared to what? >
you need a different condition (A/B)
- after Treatment condition, the CTR changes from80 to78% > is a 2% difference worth
the costs of intervention? NO > pay more than it delivers
- BUT if competitors do it, it might be worth it to attack back
Only compare groups when they are
the same > observed variables are
things that influence experiments
, but we don't actually see > observed variables can influence the DV
!! Confounder variable (they confuse) > “the devil of decision maker”: they influence both IV
and DV and you do not see them
There is a reason why you see the ad. if there is no reason, the marketing strategy is not
effective > condition of exposition is not random (> not very reliable)
Correlation vs Causality
- correlation = what will happen if nothing changes? (> we move away from correlation
to causality) ! you want to change not just to predict
- causality = what’s the next best action as a company? > you do not wanna know
what happens if nothing changes. you want to know WHAT to do so that you can do
better
- for normative insights
What can we do with experiments?
- test global change > not what we interested in
- A/B testing: know difference between A and B but does not give answers on
how to act accordingly > to test an average treatment tes: how the treatment
works on average
- we want to understand HOW DIFFERENTLY people react from one condition
to the other
- Bezos Type I vs Type II decisions > ?is the decision reversible?
Kohavi (to check experiments)
- optimize personalization > how can you segment the audience?
Lift-Base Targeting A/B Test and Randomization
1st rule of CC: Allocate your marketing resources in priority to your most responsive
customers > LIFT of an Intervention
- Profit LIFT: CLV if you target specific customers compared to the CLV if we do not
target and remove action cost > because you wanna know if more or less expensive
-
> Problem: You never observe both (targeted and not targeted) at the same time, as
people are exposed to 1 condition each
