Tailored Audience size smaller than it should be

[MySocialData]

We are having issues with the Tailored Audience sizes. Every time we upload a tailored audience based on Twitter ID numbers, the audience matched by Twitter is much smaller than expected based on our data. It’s our core business to upload tailored audiences and we want to have maximum impact in campaigns for our clients. Sometimes the audiences even drop with 75% of the database audience size. Therefore we ran many audience upload tests, please find the steps of one of these test below:

•     We created an audience consisting of Twitter users who Tweeted on the 9th of April 2018
  

•     Protected, suspended and accounts with more than 5000 followers were excluded from our list
  

•     The list of 15146 public, unique and active Twitter users were uploaded into Twitter Tailored Audiences
  

•     We are sure that all users were active as they tweeted the day before.
  

•     After uploading, the audience size dropped to 11.7K users
  

•     We understand that some people opt-out for tailored audiences but we doubt that the number is that high.
  

•     We see the same or larger audience size drops in other audiences that we upload.
  

We have already discussed this at the Twitter office in New York, they advised us to inform you about this in the community. We would like to understand how this is possible and what the exact methodology is behind uploading and matching tailored audiences.

Let us know if you have any questions.

[imit8me]

Hey @MySocialData

When it come to match rates there’s not really much we can do, and these are what they are. Generally speaking we match users if:

1. They’ve logged onto Twitter in the last 30 days
2. They are an “Active” Twitter user.

We determine “Active” users based on several criteria, and not just whether they’ve Tweeted or not, which is why you may see smaller Audience sizes than expected. Additionally, the drop in the Audience size is an artifact of the current pipeline. When Audiences are created/updated we will provide the “full” match rate, i.e., excluding the conditions described above. Once we layer on the additional filtering (i.e., Active users), the tends to drop since users are filtered out.

We completely understand that this has been frustrating for advertisers and causes confusion, and so we’re making a concerted effort this year to revamp our Audiences product from the ground up and alleviate some of these concerns. We don’t have firm timing on launch dates, but expect changes to be visible in Q3 of this year.

Hope that helps!

[MySocialData]

Hi @imit8me,

Thank you for your explanation - the audience drop is clear for us. Good to hear that you are working on revamping the audience product from the ground up.

We have detected something very strange now, we have uploaded the exact same CSV list with different audience names on different moments to understand what’s wrong. This is what we see:
May 16: CSV upload results in an audience of 563.000 users
May 22: CSV upload results in an audience of 435.000 users
June 11: CSV upload results in an audience of 403.000 users
June 20: CSV upload results in an audience of 388.000 users

This means that in a month, the match rate dropped syncificantly while we are using the exact same CSV file with the same Twitter User IDs. The audiences are still in our account so I can share screenshots if you would like to see what’s happening. We see the same thing happening for all other audiences that we upload.

This is very alarming for us because the new audiences that we upload for our advertisers are now becoming too small to deliver the budget. Could you please help me out here?

Thanks,

Thomas

[imit8me]

Hey @MySocialData

Can you provide the tailored audience ids for the all the lists that were uploaded?

Given the focus on the new Audiences product, I’d recommend holding off on any changes and re-uploading these lists via the new endpoints when they’re released (likely around August).

Thanks!

[MySocialData]

Hi @imit8me,

Thank you for your reply, please find the audience IDs below:

2mlj6
2mlo6
2l7fu
2l9pg
2ljmp
2lkzt
2lq6h
2m56e
2me9p

As explained earlier, we are uploading the exact same audience ID list, but the actual audience size that is matched becomes smaller every day. If we wait a few days and we upload the same ID list, the audience will already be significantly smaller than before.

We are facing this issue with all audiences that we upload, accross all Twitter Ads accounts (100+) from our clients. Our clients are already complaining that they cannot spend their budgets because of the smaller audience sizes. It would be great if you could look into this / help us out.

Please let me know if you need any further information.

Thanks,

Thomas

[imit8me]

@MySocialData

Can you provide some additional data around the size of the CSV file uploaded, the number of user identifiers in the file as well the user identifier type as well?

From an initial review it looks like the number of entires in the input file is 4800061 and the total number of users matches (i.e., both active and inactive users) generally varies between 3173361 and 3995244. This is generally consistent with what we see using other list uploads as well.

Have you ran any tests with different identifier types, and noticed similar results?

Thanks!

[MySocialData]

Thank you for this review @imit8me.

the number of users in the file is: 4.8M. The type is: Twitter user IDs. The match that you found sound consistent to us as well (between 3.1M en 3.9M) However the final audience size that is displayed varies between 400K and 500K, and becomes smaller every time that we upload the same list a day later. That means that 1 out of 10 matched users seems targetable, which is not consistent with what we used to see. We have tested with Twitter @Handels, we see the same result when we do that.
Based on our data we see that the number of active users within this list is significantly higher than 1 out of 10. Is it possible to revieuw the process that filters active users after the audience is matched?

Thanks a lot for your help

[imit8me]

Is it possible to revieuw the process that filters active users after the audience is matched?

This will be part of the changes to the new Audiences endpoints slated to launch in the next month or so, so we would recommend holding off until those are available.

[MySocialData]

Hi @imit8me, just checking-in to see if you have an update regarding this issue, will the endpoints launch soon? Let me know, Thanks Thomas

[mjbowell]

Hi,

Will the tailored audience upgrade be available in the UI or will it only be available via API?

We are using the UI and our match rates are as low as 8% - I’m questioning whether the API route would be better.

Mike

[imit8me]

Hey all,

A quick update around the new audiences API. We’re working through some final QA and integrations tests prior to launch. We don’t have a defined date as of yet, however we’re targeting the end of August for the launch.

Additionally, the UI will be built upon the new APIs so you should see the same functionality and feature sets on the UI as with the API.

Thanks for the patience here everyone!

[MySocialData]

Hi @imit8me,

We have access to the new audience API and we have done many audience uploads through both the new audience API and the Twitter UI. In the beginning of september we have seen improvements in both processing speed and match rates. However, over the last two weeks we have seen similar issues as we had before, in some cases worse than before.

This is what we see: Audience uploads with the exact same list of Twitter IDs show different results. Match rates are more than 2 times lower than what we have seen in the beginning of september. Furthermore, the same audience ID lists result in significantly different audience sizes across advertising accounts. In some cases the size of the same audience differs with 10 thousands of accounts. Finally, the audiences are stuck on processing for the last 3 days, but this is something that is probably going to be fixed soon as we have seen that before.

We understand how Twitter is matching users and we know that you only match active accounts (active in last 30 days) and exclude suspended accounts, protected accounts and users who do not want to receive tailored advertising. However, we work with Twitter data and Tailored Audences for more than 5 years and we really feel that there is still something going wrong when the uploaded lists are matched with Twitter users.

It would be great if you could look into this and please let me know if we can help you with more information around this issue. We are more than happy to jump on a call with you if that helps.

Thanks,

Thomas

[MySocialData]

Hi @imit8me,

Dou you have an update on the match rate issue mentioned in the comments above? We are still seeing 2.5 times lower match rates than we have seen before which limits all of our advertisers in their spend on Twitter. We are in contact with other Twitter partners / advertisers who noticed this issue as well.

Let me know.

Thanks,

Thomas