This post was written by Yousuf Bhaijee. Additional insights provided by Tomas Pueyo (VP Growth @ CourseHero), Michael Taylor (Co-Founder of Ladder) and Phil Carter (Growth Product Lead @ Quizlet).
Hi Yousuf, first, I expected this to be info-packed, and it was just that.
I feel like there is some degree of inaccuracy that is inevitable for any brand using the coeffecient.
Here's what I mean:
The customer journey is more complex than ever with users going across devices and going from one channel to another, which is recking attribution as we know it.
So if a customer sees an influencer campaign, doesn't click on the link to go the website, goes to google to check the brand on their own and then clicks from Google (using a branded search) to view your website then decides to signup. Here we'd attribute them as organic users although they came from an influencer collaboration we paid for, they just weren't attributed to that campaign.
Now imagine we are running multiple paid acquisition channels, affiliate marketing and partnerships, etc. Their will be a good amount of users that will come to us were they first heard about us from a paid channel, but then are attributed as organic since they didn't click on the link, they visited us directly/using branded search.
So as a company scales more channels, the degree of inaccuracy of the coefficient will increase. That will make it hard to see the tipping point for a company, the number of organic users who are actually coming from WOM.
And as you previously stated, attribution surveys have biases so they aren't a good measure of WOM.
As you said, with Zynga, isolation of channels (only using WOM to fuel acquisition) made the metric very accurate.
However, if a company has paid CPA goals (as stated in this blog post) and are testing paid acquisition channels, so going beyond WOM, it will start to increase that degree of inaccuracy. So Zynga would have increased that degree of inaccuracy had it introduced new paid channels.
Your thoughts on this would be greatly appreciated. Thank you.
1) there is a broader point that you are referring to, which is marketing model inaccuracy. This is true. All attribution models are inaccurate in their own specific ways. However what you are looking for with any attribution model is the ability to make good business decisions. If you can’t, you should not use the model. Alternatively, I have used multiple attribution methods simultaneously to triangulate making a business decision (eg if methods 1,2,3 all tell me I should spend more, let’s spend more!).
2) The word of mouth coefficient and tipping points are no different, you need to calculate and see if it makes sense for your product. The second article goes in depth on how to calculate and refine for your product. However if you can’t get a good enough r^2, it’s not a good fit for your product and marketing mix. https://open.substack.com/pub/yousuf/p/calculate-word-of-mouth-coefficient?
3) when your marketing mix gets too complicated (like you are describing above) it’s best to use media mix modeling methodologies for making decisions. There is a good visualization of that in the second article as well.
4) I know some companies who outgrew the WoM coefficient. It was extremely valuable for early development and then when they added TikTok marketing it became way too noisy to work. Which is totally ok
5) some clarifications to your comments: Zynga had a lot of paid ads whereas ClassDojo had nothing. We used WoM coefficient for Zynga and tipping points for ClassDojo
Hi Yousuf, first, I expected this to be info-packed, and it was just that.
I feel like there is some degree of inaccuracy that is inevitable for any brand using the coeffecient.
Here's what I mean:
The customer journey is more complex than ever with users going across devices and going from one channel to another, which is recking attribution as we know it.
So if a customer sees an influencer campaign, doesn't click on the link to go the website, goes to google to check the brand on their own and then clicks from Google (using a branded search) to view your website then decides to signup. Here we'd attribute them as organic users although they came from an influencer collaboration we paid for, they just weren't attributed to that campaign.
Now imagine we are running multiple paid acquisition channels, affiliate marketing and partnerships, etc. Their will be a good amount of users that will come to us were they first heard about us from a paid channel, but then are attributed as organic since they didn't click on the link, they visited us directly/using branded search.
So as a company scales more channels, the degree of inaccuracy of the coefficient will increase. That will make it hard to see the tipping point for a company, the number of organic users who are actually coming from WOM.
And as you previously stated, attribution surveys have biases so they aren't a good measure of WOM.
As you said, with Zynga, isolation of channels (only using WOM to fuel acquisition) made the metric very accurate.
However, if a company has paid CPA goals (as stated in this blog post) and are testing paid acquisition channels, so going beyond WOM, it will start to increase that degree of inaccuracy. So Zynga would have increased that degree of inaccuracy had it introduced new paid channels.
Your thoughts on this would be greatly appreciated. Thank you.
Couple thoughts!
1) there is a broader point that you are referring to, which is marketing model inaccuracy. This is true. All attribution models are inaccurate in their own specific ways. However what you are looking for with any attribution model is the ability to make good business decisions. If you can’t, you should not use the model. Alternatively, I have used multiple attribution methods simultaneously to triangulate making a business decision (eg if methods 1,2,3 all tell me I should spend more, let’s spend more!).
2) The word of mouth coefficient and tipping points are no different, you need to calculate and see if it makes sense for your product. The second article goes in depth on how to calculate and refine for your product. However if you can’t get a good enough r^2, it’s not a good fit for your product and marketing mix. https://open.substack.com/pub/yousuf/p/calculate-word-of-mouth-coefficient?
3) when your marketing mix gets too complicated (like you are describing above) it’s best to use media mix modeling methodologies for making decisions. There is a good visualization of that in the second article as well.
4) I know some companies who outgrew the WoM coefficient. It was extremely valuable for early development and then when they added TikTok marketing it became way too noisy to work. Which is totally ok
5) some clarifications to your comments: Zynga had a lot of paid ads whereas ClassDojo had nothing. We used WoM coefficient for Zynga and tipping points for ClassDojo
Hi Yousuf, yeah that makes sense.
I just finished reading the second article you linked to. It was really insightful, really deep, and just amazing.
Thank you very much for your detailed articles and your in-depth replies.