Basic work with WindsoraiR

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 windsoraiR to gain insights into marketing campaign performance across channels like Google and Facebook. 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. 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 Facebook and Google ads 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:

library(windsoraiR)
my_data <- 
  windsor_fetch(
  api_key = "your api key",
  date_preset = "last_7d",
  fields = c("source", "campaign", "clicks",
             "medium", "sessions", "spend")
)

Lets take a peek at the data we just downloaded to get a better idea about the structure and type of information included.

str(my_data)
#> 'data.frame':    1677 obs. of  6 variables:
#>  $ data.campaign: chr  "Website visits - May 25, 2019" "(ID)<00_mat>[id-cat]{eb}: mattress" "(ID)<00_mat>[emma mattress]{eb}: emma mattress" "Kampanja #1" ...
#>  $ data.clicks  : int  1 0 0 0 0 0 0 0 0 0 ...
#>  $ data.spend   : chr  "4" "0" "0" "0" ...
#>  $ data.medium  : chr  "Unknown" "Unknown" "Unknown" "Unknown" ...
#>  $ data.source  : chr  "linkedin" "google" "google" "google" ...
#>  $ googlesheets : chr  "'spreadsheet_id'" "'spreadsheet_id'" "'spreadsheet_id'" "'spreadsheet_id'" ...

Analyzing our Facebook and Google ad campaign data

First, lets try to find the campaigns with most clicks. To do this, we’ll filter only those rows that have clicks, then group the dataset by campaign (data.campaign column) and sum up the click count per campaign.

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)

top_10 <- 
  my_data %>% 
  filter(data.clicks > 0) %>% 
  group_by(data.campaign) %>% 
  summarise(n_clicks = sum(data.clicks)) %>% 
  ungroup %>% 
  arrange(desc(n_clicks)) %>% 
  slice_head(n = 10)

knitr::kable(top_10)
data.campaign n_clicks
AO_Smart_Shopping_Running_Jogging 56670
city_display_awareness_hyd_april_2021 54711
AO_Dynamic_Remarketing_Cart_Abandoners 54089
AO_Dynamic_Remarketing_Cart_Abandoners_Website 39969
AO_Smart_Shopping_Yoga 27551
AO_Search_Brand_Exact 24535
AO_UAC_app_install 21958
AO_UAC_app_install_Hyd 20155
AO_Shopping_Cycling 18852
AO_UAC_app_install_chennai 16905

Thereafter we can quickly visualize our data using ggplot2

ggplot(top_10, aes(x = n_clicks, y = data.campaign)) +
  geom_col()

We can gain further insight by grouping by other variables add mapping them to the plot aesthetics. For example, in this case we might want to keep track of the data source (linkedin, google, …). So we’ll modify our magrittr chain to add another grouping variable before tallying the clicks.

top_10 <- 
  my_data %>% 
  filter(data.clicks > 0) %>% 
  group_by(data.source, data.campaign) %>% 
  summarise(n_clicks = sum(data.clicks)) %>% 
  ungroup %>% 
  arrange(desc(n_clicks)) %>% 
  slice_head(n = 10)
#> `summarise()` has grouped output by 'data.source'. You can override using the `.groups` argument.

knitr::kable(top_10)
data.source data.campaign n_clicks
google AO_Smart_Shopping_Running_Jogging 56670
google city_display_awareness_hyd_april_2021 54711
google AO_Dynamic_Remarketing_Cart_Abandoners 54089
google AO_Dynamic_Remarketing_Cart_Abandoners_Website 39969
google AO_Smart_Shopping_Yoga 27551
google AO_Search_Brand_Exact 24535
google AO_UAC_app_install 21958
google AO_UAC_app_install_Hyd 20155
google AO_Shopping_Cycling 18852
google AO_UAC_app_install_chennai 16905
ggplot(top_10, aes(x = n_clicks, y = data.campaign, fill = data.source)) +
  geom_col()

In this case the vast majority of data (all of it for the top 10 ad campaigns) comes from Google, so only this source is labeled on the graph.

We can apply the same type of data manipulation and plotting to check the data.spend values.

my_data %>%
  filter(data.clicks > 0) %>%
  group_by(data.campaign) %>%
  summarise(sum_spend = sum(as.numeric(data.spend))) %>%
  ungroup %>%
  arrange(desc(sum_spend)) %>%
  slice_head(n = 10) %>%
  ggplot(aes(x = sum_spend, y = data.campaign)) +
  geom_col()

Finally, for a direct comparison, we can aggregate both the clicks and spending per ad campaign and plot them jointly:

library(tidyr)

my_data %>%
  filter(data.clicks > 0) %>%
  group_by(data.campaign) %>%
  summarise(n_clicks = sum(data.clicks), sum_spend = sum(as.numeric(data.spend))) %>%
  arrange(desc(sum_spend)) %>%
  slice_head(n = 10) %>%
  pivot_longer(cols = c("n_clicks", "sum_spend"), names_to = "aggreg", values_to = "values") %>%  
  ggplot(aes(x = values, y = data.campaign, fill = aggreg)) +
  geom_col() +
  facet_wrap("aggreg", ncol = 2) + 
  theme(legend.position="bottom")