RUM Archive Holiday Traffic 2023
By Amit Singh on
Table of Contents
Black Fridays of the past, those when I was growing up, had two unmistakable hallmarks – they took place 1) in person and 2) on a Friday. Today, the prevalence of the Internet has not only transformed Black Friday commerce, but ushered in Cyber Monday, where retailers have shifted their focus from brick and mortar stores to online marketplaces.
Any analytics surrounding web sales occurring around these two days will tell you what you instinctively know, that cyber traffic spikes on the weekend after Thanksgiving every year.
In reality, it may feel like Black Friday sales are advancing earlier each year. While a robust economic study may help corroborate this change, an Akamai mPulse-centric view of the world[1] suggests that this transition is certainly happening.
RUM Archive
I investigated this advancing holiday sales theory with the help of RUM Archive data. I queried for overall beacons segmented by day and device type for each day in September and October. I then noted the day of the week and found the average daily beacons over these two months for a Sunday, for a Monday, for a Tuesday, and so on...
Next, I looked at the daily beacon count (i.e., customer website visit count) for each day per device type (e.g., Desktop, Mobile, or Tablet) in November, and divided it by the average beacon count from September and October for that day of the week and device type. For example, if I saw 100 beacons on Mobile devices on 11/1/2023 (which was a Wednesday), and average daily beacons on Mobile devices on Wednesdays from the prior two months was 80, I would get a value of 100/80 = 1.25. Following this analysis through for the entire month of November led to the creation of a frequency table apt for visualizing in a heatmap.
After doing this exercise and visualizing the results, I saw two interesting, albeit not entirely surprising, phenomena.
- A gradual rise in Mobile and Tablet usage throughout the month of November
- A drop off in Desktop usage on Thanksgiving day
The first phenomenon supports the theory of sales creep, where retailers are inviting customers to their sites earlier each year. Further, it may be the case that mobile usage increased as holiday shopping occurred on personal phones during traditional work hours.
The second phenomenon is very likely a dispersion away from Desktop usage towards Mobile or Tablets as families gather for Thanksgiving festivities.
Appendix
Below is the query I used in BigQuery to access the RUM Archive data. I then exported the results to CSV and visualized the results in Python via a Jupyter notebook using Seaborn.
WITH baseline_data AS (
SELECT DATE, EXTRACT(DAYOFWEEK FROM(DATE)) AS dayofweek, SUM(BEACONS) as total_beacons
FROM `akamai-mpulse-rumarchive.rumarchive.rumarchive_page_loads`
WHERE DATE BETWEEN "2023-09-01" AND "2023-11-01"
GROUP BY 1,2
ORDER BY 1 ASC
),
baseline_data_agg AS (
SELECT dayofweek, AVG(total_beacons) as avg_beacons
FROM baseline_data
GROUP BY 1
ORDER BY 1 ASC
),
november_data AS (
SELECT DATE, EXTRACT(DAYOFWEEK FROM(DATE)) AS dayofweek, SUM(BEACONS) as total_beacons
FROM `akamai-mpulse-rumarchive.rumarchive.rumarchive_page_loads`
WHERE DATE BETWEEN "2023-11-01" AND "2023-11-30"
GROUP BY 1,2
ORDER BY 1 ASC
),
joined_data AS (
SELECT t1.DATE, t1.DAYOFWEEK, t1.total_beacons, t2.avg_beacons
FROM november_data AS t1
INNER JOIN baseline_data_agg AS t2 ON t1.DAYOFWEEK = t2.DAYOFWEEK
)
SELECT DATE, total_beacons, total_beacons / avg_beacons as day_ratio
FROM joined_data
ORDER BY 1 ASCAn important caveat to this analysis is that both retailers and non-retailers are present in the RUM Archive data, with the former accounting for anywhere between 50-70% on a daily basis. ↩︎