library(appsflyeR)
library(dplyr)
#>
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#>
#> filter, lag
#> The following objects are masked from 'package:base':
#>
#> intersect, setdiff, setequal, union
library(ggplot2)
Goal
The goal here is to outline in a couple of paragraphs and few lines
of code some simple ways in which we can use the Windsor.ai API and R package
appsflyeR to gain insights into marketing campaign
performance. The nice thing about Windsor.ai is that you can have all of
your marketing channels aggregating in a single place and then access
all data at once using this package. In this case, however, the package
is focused on getting marketing data from AppsFlyer. Of course, once the
data is in R you can do much more than the examples below,
and work on analysis, predictions or dashboards.
Getting data from AppsFlyer into R
After we create an account at Windsor.ai and obtain an
API key, collecting our data from Windsor to R is as easy as:
my_appsflyer_data <-
fetch_appsflyer(api_key = "your api key",
date_from = Sys.Date()-100,
date_to = Sys.Date(),
fields = c("campaign", "clicks",
"spend", "impressions", "date"))
This code will collect data for the last 100 days. Lets take a look
at the data we just downloaded to get a better idea about the structure
and type of information included.
str(my_appsflyer_data)
#> 'data.frame': 14 obs. of 5 variables:
#> $ campaign : chr "retageting APAC" "retargeting UK&CO" "retageting APAC" "retargeting UK&CO" ...
#> $ clicks : num 4 0 5 7 0 0 4 2 3 0 ...
#> $ spend : num 2.57 2.48 2.39 2.54 0.94 0.71 2.59 2.12 2.43 0.13 ...
#> $ impressions: num 806 693 819 689 299 190 682 688 822 135 ...
#> $ date : chr "2022-09-28" "2022-09-28" "2022-09-29" "2022-09-29" ...
Analyzing AppsFlyer marketing data
Now we can analyze data from AppsFlyer data. For instance, let’s
compare the two campaigns we have to see which one performed better the
last 100 days.
ggplot(my_appsflyer_data, aes(y = clicks, fill = campaign)) + geom_boxplot()
It looks like APAC campaign is performing better than UK&CO in
number of clicks. Now let’s see if this difference is statistically
significant by using generalized linear models, as our variable response
is number of clicks, which have a Poisson distribution.
lmod <- glm(clicks ~ campaign, data = my_appsflyer_data, family = "poisson")
summary(lmod)
#>
#> Call:
#> glm(formula = clicks ~ campaign, family = "poisson", data = my_appsflyer_data)
#>
#> Deviance Residuals:
#> Min 1Q Median 3Q Max
#> -2.3905 -1.6036 -0.7599 0.6372 3.5065
#>
#> Coefficients:
#> Estimate Std. Error z value Pr(>|z|)
#> (Intercept) 1.0498 0.2236 4.695 2.67e-06 ***
#> campaignretargeting UK&CO -0.7985 0.4014 -1.989 0.0467 *
#> ---
#> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#>
#> (Dispersion parameter for poisson family taken to be 1)
#>
#> Null deviance: 43.735 on 13 degrees of freedom
#> Residual deviance: 39.456 on 12 degrees of freedom
#> AIC: 66.147
#>
#> Number of Fisher Scoring iterations: 6
We can see that differences among campaigns are statistically
significant and that the campaign UK&CO have a mean that is 0.79
lower than the APAC campaign.