Estimated CTR

Adthena uses a complex machine learning model to estimate nightly both CTR (click through rate) and CPC (cost per click) for every single search term, for every competitor in every position we have seen them appear in that day.  

This model is constantly learning from the 10 terabytes of new data that Adthena is collecting daily, but essentially it uses the following inputs:

• What is the country?
• What is the device? (there is a separate model for mobile)
• Who is the competitor? (their size*, overlap, appearance frequency)
• What is the search term? Is it brand or generic?
• How many advertisers compete on this search term?
• Is their most frequent position above the fold (top of the page or bottom)?
• What times of the day and days of the week does the advertiser generally appear on this term?
• Are there other items on the page (site links, other extensions, knowledge boxes, image search, PLAs, comparison boxes) that could draw clicks away from this advert?

*the size of their search term universe - how many search terms have we seen them appear on (both paid ads and organic listings)?

 

Average CPC

The model estimates the CPC for each competitor in each position in a very similar way to estimating CTR. Is it a brand term? Is it the brand owner? How many different competitors are appearing? etc. (see above for full list).

Average CPC values are predicted on a monthly basis for each search term and competitor.

 

Average Monthly Search Volume

How many people search for the term?

Adthena uses Google’s average monthly search volume as one of its underlying data sources. The search volume is updated once a month as it becomes available from Google. 

This also affects the Estimated Clicks.

 

Frequency

Is the % of appearances Adthena sees a competitor out of all times searched (includes the number of times users searched for a term but no ads were shown), weighted by Search Volume.  

Example

If Adthena searches for "black dress" 100 times and sees competitor retail-brand.com showing an advert on 50 of those occasions, then retail-brand.com will have a frequency = 50% on that specific term.

If retail-brand.com is appearing on multiple terms, then the aforementioned % will be weighted according to the average monthly search volume of each term. 

Using the two search terms "black dress" and "black cocktail dress", which Google tells us have an average search volume of say 3,000 searches per month and 300 searches per month.

Retail-brand.com appear on "black dress" 100% of the time and "black cocktail dress" 50% of the time. Rather than showing a resulting average frequency of 75%, this example scenario will be weighted based on the fact that "black dress" is searched for 10x more often, so the frequency will be 95%.

Calculation: (100% x 3,000 + 50% x 300) / (3000 + 300) = 95.45%

Note however that Frequency only includes the terms that a competitor appears on. 

 

Estimated Impressions

1. We calculate the Frequency that we see your relevant competitors on your relevant search terms.
2. Using Frequency (for each competitor and each search term) and Search Volume for each search term, we calculate the number of impressions of a competitor on a search term for the given period.
3. If you then sum the impressions of all search terms we've seen a competitor on, you will get the total number of impressions of the competitor for the given period. This is Estimated Impressions.

Using the above example, Estimated Impressions is the sum of Frequency x Average Monthly Search Volume (for each competitor and each search term)

Calculation: (100% x 3,000 + 50% x 300) = 3,150.

  

Share of Impressions

= Estimated Impressions / ALL available impressions. The latter is the total Google search volume for all relevant search terms.  A competitor who appears every time Adthena searches for a specific term will have 100% Impression Share.

 

Estimated Clicks 

Multiply Estimated Impressions by the CTR for the competitor and the search term. At the end of this procedure you'll get the Estimated Clicks for each competitor on all of the relevant search terms they've been seen on.

Example:

Let’s say a search term gets 1,000 searches per month and a competitor’s advert appears on the term 60% of the time, so they generate 600 impressions. They always appear in position 2 with a CTR estimated at 5% which equates to 30 clicks per month.

The following week the competitor is seen 80% of the time fluctuating (70:30) between position 2 and position 3. The model has re-estimated the CTR for position 2 to 6% and for position 3 estimated  3.5%.

1000 searches / 30 * 7 days * 80% frequency = 187 impressions that week

(187 impressions x 70% p2 frequency x 6% CTR) + (187 x 30% x 3.5%) = 10 clicks

Estimating Clicks gets more complicated when you include additional competitors on the same search term and additional search terms (that have different search volumes and sets of competitors).

 

Share of Clicks

= Estimated Clicks / Sum (Estimated Clicks for all relevant competitors)

 

Share of Spend

A competitor’s spend is estimated by multiplying their Estimated Clicks for each search term by their Estimated CPC for that search term. The total spend for all competitors can then be calculated and each competitors spend divided by the total is their Share of (the total) Spend.