The Complete Guide to the New Algorithm

On September 8, 2019, Similarweb updated the algorithm it uses to generate traffic and engagement stats for every one of the millions of sites that it tracks around the world.

The change to our algorithm comes after enormous research and development efforts to build on and enhance our platform. Similarweb’s market intelligence platform delivers the highest standards of benchmarking excellence, the most valuable insights and the widest possible angle for visibility of your industry, your market and your competitors.

Brief Reminder About Our Methodology

Similarweb’s data methodology uses a multidimensional approach to data sourcing to ensure the reliability of the numbers and proportions we display. The platform relies on four primary sources to give a comprehensive view of the digital footprint of your market and your competitors’ sites.

● Global Panel Data
● Data Partnerships
● Public Data Sources
● Direct Measurement Data

To learn more, read our Knowledge Center article, Similarweb Data Methodology.

How Did The Algorithm Change

Similarweb’s insights are based on various data sources. Each source exhibits a different, known bias due to different groups of users with divergent interests. Similarweb’s algorithm has always worked to reduce the bias and provide the most reliable measure of desktop and mobile web traffic and engagement for every site.

Due to Similarweb’s access to direct measurement for a wide range of sites, we were able to use direct measurement data to identify correlations between sites with similar audiences which exhibited similar biases. The sites providing Similarweb with their direct measurement data make up the learning set from which we have derived the improvements to the algorithm.

Before now, Similarweb’s data algorithm took into account the specific source of the data and the country that data is coming from. The update to the algorithm added a further dimension and also adjusted for the behaviour of visitors to similar kinds of sites. To achieve this we mapped every site by its audience correlation with the sites in the learning set. The mapping generates a set of features that use a machine learning algorithm to apply the appropriate adjustment for every site.

Why Did We Make This Change?

The short answer is that we are always looking for ways to improve the dependability of the platform. We pride ourselves on the reliability of the insights Similarweb provides, but when we uncover a way to upgrade even further, we are keen to pass that on to all our users. From September 8, 2019, the updated algorithm has been applied to, and replaced the data from the past 37 months, to ensure consistency and accuracy across all your reporting.

Which Metrics Will Be Impacted?

The new algorithm is likely to have an impact on absolute traffic volumes of sites. This will be reflected in changes to the following metrics:

● Visits
● Unique Visitors
● Page Views
● Bounce Rate
● Visit Duration
● Ranking
● Desktop vs Mobile share
● Site and Channel Conversion

For the traffic sources to the desktop version of a site the impact is likely to be felt more in the absolute numbers from each channel rather than in the proportions.

The new algorithm will impact the traffic to a site for each country from which it sees traffic. As a result, the share of each site’s traffic by country may be affected.

For monthly keywords, we anticipate that the share of, mainly branded, keywords will see some changes.

The changes to the algorithm will impact the insights available on our platform as well as through the Similarweb API, Digital Insights and any custom reports.

How Much Has The Data Changed?

The nature of the changes to the algorithm is that the size of the impact will vary across different kinds of sites. With adjustments applied for groups of similar sites by category, industry or vertical, each group is likely to experience different types of adjustment.

This means that some groups of sites will have adjusted up and some adjusted down. We have already noted that traffic volumes to Fashion and Travel websites are likely to increase while traffic numbers to Technology websites may go down.

Which Features Were NOT Affected By The Update?

● App Analysis
● Company Data
● Categories
● Similar Sites
● Demographics
● Competitive Landscape
● Keyword Volume and CPC
● Ad Creatives and Videos

Conclusion

The purpose of the update to our algorithm was to increase the reliability of the insights you get from Similarweb. Following the change, Similarweb’s platform offers an even more accurate vision of the competitive landscape for millions of websites, in countries around the world.