Mar 18, 2020 - 07:35 AM
I have always questioned the reliability of Alexa data. In fact, I wrote a piece on it way back in 2007 called Please Stop Quoting Alexa Data. The root of the problem with Alexa data is how it gets data. Like most traffic estimating tools, it uses a panel. Back when I wrote the article it relied on only users who installed an Alexa toolbar, which is an unusual panel that does not represent the U.S. audience very well. Panels that use ISP data (like Comscore, Nielsen or the recently shutdown Hitwise) are much more accurate.
In that article, I provided a comparison of real data from two popular sites that showed a site with 102,000 users getting the same traffic as a site with 11,000,000. The smaller tech-focused site registered the same traffic as a top 100 property because of its audience, which was more likely to install the Alexa toolbar.
A year later Alexa announced it would no longer rely purely on the toolbar, but collect aggregate data from "multiple sources." The new sources haven't been divulged, but many believe they were buying data from other widgets and browser plug-ins.
Regardless, the data hasn't ever been very accurate and is biased towards long time internet users in the startup space. Alexa even states, on their help page that "We do not receive enough data from our sources to make rankings beyond 100,000 statistically meaningful."
Where Alexa data can be accurate is on certified sites that install Alexa code on their pages. Most sites prefer to use regular analytics platforms that don't make their data publicly available.
Going back to your situation, it makes sense that Alexa may not have caught the dip like it should have considering people may have still tried visited your site while it was essentially down (if they had the toolbar installed). Your own analytics is obviously a much better measure.
Out of curiosity, I pulled data for your site on Ahrefs & Spyfu to see how the traffic was shown during the transition. I tried pulling SimilarWeb, but they must have temporarily removed your site due to the dip. They look more accurate. Here's what they look like: