Heiko Specht Profile picture
Jul 16, 2022 23 tweets 12 min read Read on X
/Thread about possible #ad_fraud and how I analyzed it.
I am a big fan of twitter and until now I thought it is a great marketing platform to reaching a good target audience - but I seem to be wrong. But read yourself
For this analysis I used the @newrelic telemetry data platform and the data analyzed are access logs from @fastly where I got some excellent support from.

So our access logs told us: 138k visits in the past week from @Twitter - wow. NOT BAD !!
My motivation in the first place was to help our #marketing team getting more insights into the success of our paid #socialmediamarketing campaigns to make them more successful - getting faster in removing those who don't perform and allow them to experiment.
But this turned out to be a tough task. With a very very worrying result at the end. Because the first look a the last week traffic from twitter looks like this.
(all traffic filtered by utm_source=twitter over the last week - redacted) a chart showing the traffic coming from Twitter grouped by t
It seems that our employee marketing is working well. We use @postbeyond and all employees are invited to post prepared content on the internet. The other campaigns are paid #socialmarketing campaigns - top ten. (whatever buffer is...)
So I was keen to see where the traffic is coming from and to look for other attribues @fastly 's access logs provide. Such as country, city, Bot classification, client proxy information. (the proxy description and proxy type ? are not classified = probably real user) Pie Charts showing the distribution by city and country of tPie charts showing classification for Bot or not, client pro
By looking at the charts recognize a weird city and country distribution (given that most campaigns are EMEA campaigns) but also - it is clear that most of traffic is orignated from clients already classified by @fastly as bots.
@newrelic allows me to narrow the results down to where I expect real user to be. So I filter out the bots (click on false), concentrate on real users (proxy type = ? and proxy description = ?) which brings me down to 51k clicks - sure - with that many clients classified as bots
Now I gotten rid of the bots in my charts look at the geo distribution again. Uh....so much traffic from Singapore? And so much traffic from Singapore on the @Postbeyond campaign and our EMEA campaigns...weird - isn't it. Let's dig deeper geo distribution by country showing massive amounts of traff
A nice feature in the logs from @fastly is that they provide a "client_as_name" attribute that I ingested to @newrelic as well - and it shows me that most of the Singapore originated traffic comes from a single ISP. chart shows that most of the traffic comes from 1 dedicated
Ok....this is strange - so I wanted more details from the data. And narrowed the traffic down to this provider. So I look a the user agents of the clients requesting our page from a Twitter Marketing campain - not classified as bots.
And here is the result of those "users" requests from Singapore. Hey Singapore people - didn't you know that Chrome has a version 103 out there? Look at the distribution - clearly you can see - these are bots. Lets kick them out for further analysis. showing old (very very old) user agents with a weird (equal)
So, we are now down at 13.2k Traffic from our investment into twitter. 10% of what we'd seen in the first place. And we have 4 filters applied only.
our @Postbeyond campaigns are no longer in our top ten. And we see a pretty equal distribution of the traffic coming from our #digital Marketing campaigns with an expected dip today (Saturday). Very equal distribution area chart of traffic on our pages g
But...looking at the distribution per geo I appear not to be completed with my mission - why are our EMEA marketing campaigns clicked from South Africa? I was not aware that there is so much interest in #o11y . Let's dig deeper.. Pie charts showing traffic to mostly come from South Africa
Honestly now things get a bit weird. I focussed on the ZA traffic, I repeated the 'look for the user Agent play' and I checked further things and the only thing worrying me is the equal traffic that coming from the cities and that campaigns generated equal interest clusters.
Would it be a fair question @TwitterMktg to ask why South Afrika has such an interest equally distributed over 4 Campaigns meant for EMEA and equally distributed over the cities in ZA ? If I had to guess:
There are real people motivated by whatever clicking on #DigitalMarketing campaigns.
But why? I hope this is not for *making it appear* as if the Twitter campaigns are successful and generate traffic.
I walked done the road a bit further and identified a couple of other countries with exactly (!!) the same pattern: high distribution of client agent, equal interest in every campaign, weird distribution over cities. And filtered all out. Plus some obvious bots.
I know for sure I excluded maybe 20 or 30 Real Real user clicks with this filter but...now I am down to ....
1.6% of what showed up as Traffic from Twitter originated by Twitter campaigns - or roughly 5% after excluding pre-identified bots.
And if I had to guess: Out of these 2.64k "visitors" > 50% are human bots.
Oh, and whom it might interest. During the same time frame we have identified 4164 organic clicks from @Twitter

/end

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Heiko Specht

Heiko Specht Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(