It could go either way.
Well, what factors might it depend on?
There are plenty of factors, it turns out.
Yeah, for instance, it might depend on a particular city.
Calcutta might behave differently from Paris.
It could depend on the product category.
Pizzas versus apparel laminates.
It could depend on a host of other things, the target segment, for instance,
and so on.
So a bunch of things that are possible and the only way to find out,
the only way to be sure is to let data do the talking,
which brings us to the field experiments which is where we are.
One of the two big approaches in our Toolscape are experiments.
The other being modeling, of course, predictive analytics and
we'll get there in a second half of today.
Let's basically see what actually happened.
The field experiment was conducted [INAUDIBLE] in a management science paper
published in 2014 and
they answered all those questions that we raised in some sense.
So let's go back to the easy questions, the first ones we'd seen.
What happens when in some sense, you have,
what happens to response rates as time varies, for instance?
Clearly, same day had the highest response about 10% redemption.
You had one day delay having a slightly lower response and
the two day delay having an even lower response.
Another simple example.
So, this is the effect of mobile promotions in geographical targeting
along expected lines.
People who are nearby in neighboring geocodes had the highest response.
We're looking at 10% there.
Now people at a medium distance away from the store,
the redemption location had a 7% of response and
people far away had a much lower response, less than 5%.
So what do we see, basically?
The simple questions were answered in a fairly straightforward way.
Figure one, figure two,
we basically get some idea of the relative magnitudes to be had.
So, two days prior is half that of a same day reduction.
Figure two, for instance, we come to see that proximity rises, so does response.
Again, we see that far is about half that of near in some instances.
Now, here's what happened in the slightly more complex cases.
The combination of stimuli and what do we get there?
Interestingly, what we basically find is that when you combine the two
stimuli near is good again for same day.
However, near is not good in some sense as time increases.
So basically, so when you come to far locations, same day is not good.
I'd say, that's kind of expected.
When you come to near locations, same day is good and
medium sits in between those two.
What we actually see?
We'll have a look at that.
So same day, which is so high for near is basically so low for far.
One day prior interestingly is better than two day prior, even in the far location.
So, a lot of interesting insights just coming out from data collected
through a field experiment.
To quickly summarize what we saw,
combination of stimuli does tend to produce insightful results.
The highest redemption condition is nine times that of the lowest redemption
condition,18% versus 2%.
Here are some notes from the motivating example.
See, we were testing response effects of three distinct stimuli,
the three separate stimuli.
So, you have discount debt.
You have geographical proximity and you have the time available.
Consider the first two and recall in some sense,
display this as a two by two table as you can see there.
So you'll high, low there and you have near and far.
The high and low, near and far.
The exact numbers there can be determined by context.
They usually are contextual.
Now, what happens.
How do I add the third stimulus condition?
How do I add in some sense, a time available to this or
in geographical proximity to this?
The way to do that would simply be to replicate this for
different levels of geographical proximity.
High proximity condition, medium proximity condition and there would be one for
low proximity condition, as well, right?
Other interesting questions that could come about.
What other outcomes could be measured about response rates?
We were measuring response rates.
You could actually measure things like say, value profits.
If you're giving a discount, you want to know profit effects.
Post trial retention.
Will people come back?
Does it work?
Does try it? You will like it and
you will come back work?
Loyalty, for instance, and a lot of these things can be tested through experiments.
So in some sense, what this example has aimed to do is motivate
experimentation test and learn as one of the major approaches
in our Toolscape which brings me to a primer in business experimentation.
[MUSIC]