Guys, we need to talk about attribution modeling. It’s a hot issue in our industry and most of us (SEOmoz included) aren't doing it as well as we want to be. It's tough stuff. Mike P from Distilled gave a great MozCon presentation on the topic, but most of us aren't anywhere close to that sophisticated - and even his model is impacted by Google Analytics' limitations.
It’s been covered in far more detail elsewhere, but in a nutshell: attribution modeling attempts to solve the problem of which channel gets credit when a user touches multiple channels prior to converting. Many marketers simply throw up their hands and say the last touch gets all the credit – but then we have to live with the knowledge that some of our efforts are far more effective than we give them credit for.
Not-so-super modeling
Supermodel by Soggydan on Flickr
Unfortunately, attribution modeling is very hard to do well for a lot of reasons:
- Any site to which users return daily (like, for example, SEOmoz.org) quickly fills up with touches that may or may not be related to conversions.
- Channels like social media and community building are often a first touch but rarely the only touch before conversion, meaning they tend to get less credit than they deserve.
- Attributing offline sales to online efforts can be very painful, not to mention tracking one user’s conversion path as she uses multiple devices during her buying decision.
- In our post-Panda world, we’re spending a ton of time and effort on content that may end up on third party sites, opening us up to the near-impossible task of tracking view-through conversions.
In my opinion, however, the biggest problem with the attribution models available to us today is that their roots lie in web analytics tools like Google Analytics. This means that attribution models tend to be biased toward on-site efforts. The bulk of our marketing efforts doesn’t happen on-site, so why should our measurement? Our competitors certainly aren’t doing things on our site, so why should we content ourselves with on-site data?
Web-analytics-based attribution models also tend to break up sources at the channel level: organic search, social media, direct traffic, etc. Anyone who’s worked for months on driving traffic from Twitter and then had one tweet from Rand break their site can tell you not all social media touches are created equal, so why lump them all into Social Media?
.jpg)
Finally, attribution models are incredibly difficult to implement for success metrics beyond conversion (more on that later).
Marketing analytics is about campaigns, not channels
Here at the MozPlex, we’ve been talking a lot about marketing analytics: the way we measure and optimize our marketing activities. I think Joanna put it best in her post: “Marketing analytics is the act of looking past mere website results, and asking yourself, ‘How did that marketing campaign really go?’”
Marketing analytics means going beyond the data we can get from our web analytics tool so you can measure off-site and even offline activities. Capturing that additional data about how your off-site and on-site marketing activities are performing allows you to test with greater confidence, and as marketers, we should always be testing. It’s probably not as simple as “social media doesn’t drive as many conversions as organic search.” Instead, we can test how to spend our time and money - which levers to pull at which time and in which way - to attract, keep, and delight our customers. At the same time, we can take a cross-channel, holistic view of our efforts to see what messages are resonating best.
All conversions aren’t created equal
Of course, one thing we want to do with our marketing efforts is make more money. ROI-driven modeling is always going to be part of what we’re measuring. However, modern marketers are driving for more than just the lead or the sale or the free trial. We’re looking at micro-conversions like newsletter signups. We’re watching and participating in conversations about our brand. We’re investing in customer happiness. We’re tracking shares, tweets, mentions, and views – and we’re keeping an eye on how are competitors are doing, too.
In addition to major conversions, marketing analytics is about tracking customer loyalty.
Forever Friends by dprotz, on Flickr
We can often gain as much revenue from keeping our existing customers happy as from getting new ones. What happens after the conversion?
Marketing analytics is also about tracking brand identity. This is becoming more and more important as the major search engines focus more and more on brand strength as a quality indicator. This is another area where typical attribution models just don’t go far enough. Brand-centric campaigns are as much about generating conversation and positive feelings as they are about directly causing more conversions – this makes it harder to prove value if conversions are your only KPI. Branding has an influence on direct traffic, but it also has a big influence on organic search traffic from branded keywords.
So, should that traffic still count as organic search, if branding efforts are what inspired the search in the first place? This is another area where a more campaign-centric view can provide more insight than simply attributing conversions to channels.
Getting closer to marketing analytics
We’re still in the early days of true marketing analytics, which means we’re still mashing up data from a bunch of different tools and struggling to find the right ways to track campaigns. In the meantime, we can start hacking our web analytics’ attribution monitoring tools to go beyond simple channel attributions:
Advanced metrics for attribution modeling
- Top referrers (separated out from the rest of referral traffic)
- Top keywords (separated out from the rest of the keywords)
- Long-tail keywords (same deal)
- Top partners and/or affiliates
- (not provided) search traffic
- Branded and non-branded search traffic
- Individual social networks (A friend and a follower may not be the same!)
- Individual feeds
- Individual paid advertising sources
We can also start thinking of (and tracking) our data with a marketing analytics mindset:
Advanced metrics for marketing analytics
- Messages
- Type of touch (Branding? Promotion? Retention? Happiness?)
- Type of product
- Audience
- Time of day
- Conversations
In the end, marketing analytics is more useful than straight-up attribution modeling, because it allows you to view your marketing efforts holistically. When you view individual customer touches as part of a larger whole instead of siloed by medium, you can take a longer and more customer-driven view of your marketing efforts.
Sorry to spoil the party here. But i have to disagree with this post. First of all, we already have a well known term called 'Digital Analytics' which encompasses everything the new shiny term called 'marketing analytics' has to offer. Secondly Attribution modelling is used to find the most profitable marketing channels for investment and not to solve the puzzle of which channels should get the credit for conversions. In a world of multi channel marketing, different channels work together to create a sales or conversion.So it is not really about figuring who should and shouldn't get the credit. It is more about who should get the most credit, the second most credit and so on.
The biggest problem with attribution modelling is not that its roots lies in web analytics tools like 'Google Analytics'. It is infact, not being able to successfully track across multiple devices/channels and the privacy issues. Universal analytics is an attempt to resolve these tracking issues. However privacy issues will remain a big problem.
It is true that not all conversions are created equal. But it is also true that not all conversions (particularly micro conversions like customers loyalty etc) impact the business bottomline esp. if their is a negative correlation between conversions and revenue or positive correlation between conversions and acquisition cost. The foundation of all great analysis is to ensure constant rise in sales and profit. This can be achieved only when your marketing efforts are profit centric and not campaign centric or channel centric or conversions centric.
To begin with, KPI is a metric which can be quantified. If it can't be quantified, it can't be measured and hence it can't be a KPI. So positive feelings, customers happiness can not be a KPI. Which leaves me with the question which attribution model doesn't live up to your expectation in terms of these non-quantifiable metrics?
Sorry but i also have to discount all the advanced metrics you proposed for attribution modelling and the "marketing Analytics". The reason for this is that when we talk about attribution modelling we talk about first, middle and last touches. So we wont determine 'top referrers' here. We would determine top referrers which initiated a sales, top referrers which assisted the sales and top referrers which completed the sales. When you just talk about 'top referrers', you are talking about the default last click attribution model which defeats the whole purpose of looking beyond the last click.
I would say that "getting credit for conversions" and "which channels to invest in" are two very closely related items - but you're correct, talking about which channels it's worth spending time and money on would have been a more precise way to discuss this point.
Customer happiness is absolutely something you can measure - not in and of itself, but by assigning KPIs to it. You can do this using things like sentiment analysis tools and regular opportunities for customers to give feedback.
I would agree that customer happiness can be a KPI, who is to say at a basic level a regular satisfaction survey cannot be used as a KPI? It can be specific, measurable, achievable, relevant and can have a timeline. All things important to a good kpi.
Beyond something specific like a survey, as RuthBurr mentiones, using something like a sentiment analysis tool can give you trending, which many of the folks in the analytics world would argue as something worthy of looking at anyways.
If you are talking about 'task completion rate' to quantify satisfaction then it makes some sense. Other than that you can't quantify happiness through satisfaction surveys. Just because i am satisfied doesn't mean i am happy. Happiness is not even a metric let alone KPI.
Case in point: How does this post you've written have 555 tweets, but only 266 visits, according to your own Post Analytics?
People will tweet (and retweet) based on a title without actually viewing the content :)
Yes Dawn Wentzell you are absolutely right People will retweet based on title and one most important think is that they will see who has tweeted.
There are Twitter accounts that subscribe to our RSS feed and tweet our content without ever visiting. We get dozens of tweets each time we publish, even if we have a mistake and the content isn't available or there's a bad URL..there will still be tweets from bots that have never visited the page.
This suggests that many of the tweets are worthless. #notalltweetsareequal
First: thank you for writing about attribution. This has been, and will continue to be, a key topic and challenge for digital marketers... I enjoy every article and conversation on the subject.
I like that you hit on channel versus campaign. It speaks to a general need to organize campaigns and plan measurement in such a way that it is possible to attribute any given social media campaign activity (e.g. tweet) back to the campaign that caused it to happen.
I'm not sure that marketing analytics is new so much as there is a wealth of digital data now being poured into the marketing analytics stew. To speak to the semantic argument, and perhaps perpetuate it a bit... I think of digital analytics as a necessary component of marketing analytics for any business in the digital realm ... I consider almost everything that I do to be digital analytics, customer analytics AND marketing analytics -- all combined. I have a hard time calling it any one of those things, because each leaves out a nuance that the others bring to the table. That said, I believe marketing analytics is the most inclusive term.
Sometime I wonder if we shouldn't just throw Attribution Modeling in the arena of Branding. Here is my take on the subject. For businesses there is Branding and then there is marketing. I think Attribution Modeling kind of sits in between. For online marketers I think it is a great reason to provide an increased number of marketing services for your clients.That way if you control all of the marketing channels included in the campaign, then you know that the conversion can be attributed to your efforts. Avinash Kaushik does a great job of covering this topic in his book Web Analytics 2.0.
I agree with the idea that marketing analytics is about campaigns. I think if you’re going to measure marketing campaigns, you need to strongly consider cohort analysis because of the length of time it may take to convert. For example, a mortgage loan process from application to conversion could take months. Cohort analysis will give you can opportunity to marketing campaign effectiveness of days, weeks, or months and get credit in the long-run. Check out Kissmetrics or Google Analytics for cohort analysis.
As for the offline marketing tactics, I would recommend using call tracking, vanity urls, tagging QR codes, and review your inventory of marketing materials. These items should be for specific campaigns. You’ll need information on sales, leads, etc. from your business or client. Look at implementing a measurement plan (details on data collection) and get the necessary sign-off.
Very good article, I have been looking for more information on this topic and in this blog I have been able to find the most relevant information. Thank you very much
Thank you very much Ruth for this article. This article is very educational and easy to read. The topic of the article the attribution models is the most hot topic right now in online industry since it’s kind of new, still evolves and not everyone know what it is, how it’s being applied and what you can learn from it. Many companies don’t even know from where to start and don't understand the importance of the attribution models. I’m sure after reading your article many people started better understand what its all about. I myself wrote couple articles about attribution models, and would love to hear from experts (your opinion).
I believe marketing analytics is one of the most important and not only its important but it also helps in understanding the hidden and potential areas..Good write up
I agree the Google analytics is making great strides in the attribution analytics/modeling world. It was mentioned previously, that custom groups could be used, which i think is vital as we look at campaign success. In addition, looking at how specific actions are impacting end goals is important. As you mentioned, we all want to look at the revenue, but many marketers focus on additional elements, such as newsletter sign ups, blog comments, ect. However, At the end of the day, we want to see the impact these pieces have on bottom line revenue.
Does someone who comments once a month hold on to their SEOmoz account longer for example? Perhaps we don't look at every site visit( using custom variables with GA, we can filter the visits we want to see), but instead we look at visits that have key actions taken.
There are plenty of off site activities that should be considered, and unfortunately they just are not as friendly with data sharing as having on site data. However, I foresee that with Google analytics shifting to universal analytics, the user level, logged in users, will become more common within analysis and reporting. (we are already seeing the impact of logged in users in not provided reporting, lets turn those legged in users into user level analytics!)
At the end of your post, you break down advanced metrics for attribution modeling. As i see it, these are simply segments of your overall data set. What I feel is important to note is that these segments may not apply to all people, and that it is important to differentiate metrics from dimensions. What you have listed are more along the lines of dimensions, when planning your analytics, having these dimensions, alongside the metrics that you need is helpful in defining your analytics framework and strategy.
That's a good point and actually I was walking to the bus this morning and was suddenly like "Dang, I should have said 'dimensions' instead of 'metrics'!" #publishersremorse
As we analyze everything, let's not forget to leave time for actual marketing.
I think this relates to the "correlation versus causation" dilemma that a lot of people face. You're right - it is difficult to pinpoint the real origin of a conversion.
There are some CRM systems (and CRM plug-ins) that combine user tracking with campaign tracking so you can get a clear picture of each user's path from first contact to conversion. Some of them even have social tracking. This topic may warrant a discussion about which tool is best for the job.
There is a ton of usable information here, thank you Ruth. In regards to the shortcomings of Google Analytics; are any of the other analytics programs better for attribution modeling? For many people it helps that Google Analytics is free, but that doesn't a;ways equate to the best product. Some SEO's suggested SiteCatalyst. I am interested to hear other opinions.
Actually I think that Google Analytics is doing the best out of the available analytics tools. SiteCatalyst can be really good but I find it works much better for paid channels (used alongside SearchCenter) than for free channels. I think the answer right now is still Google Analytics + data from additional tools (like SEOmoz tools, social media analysis tools like Radian6, etc).
Thanks for this post Ruth, it's great to see people starting to think about more advanced measurement topics.
I completely agree that in order to create the most accurate business measurement plan we need to look at many different data sources. That's one of the reasons that Google Analytics introduced external data via the Social Data Hub and Universal Analytics.
Business measurement is going to rapidly evolve for a lot of people. We started with web analytics, moved to digital analytics, then customer analytics and now we're just going to focus on Analytics. This evolution is driven by data sources. As a business you're now able to combine many data sources easily to measure acquisition and retention activities.
Honestly, a lot of businesses have been doing for a while. The reason for the evolution is that the technology is now getting pushed to the mass market at an affordable price.
All of this data will help us get back to basic principles of measurement: recency, frequency and monitization. This will then drive better measurement of lifetime value. It's old-school measurement that will help us better measure paid, earned and owned media.
But that's coming in the future. What about now?
I think attribution modeling and multi-channel measurement is getting very, very good. I agree, it's better with more data, but what we have now works very well.
One metric that I think you should include in your list is Assisted Conversions. This is an absolutely critical metrics for all marketers. If you are ignoring this metric then you're not doing your job as a marketer or analyst.
I also want to mention that there is a lot of ways that you can customize your attribution and multi-channel funnel reports in Google Analytics. These modifications make it easy to mitigate many of the issues you describe. For example:
--You can use the channel groupings feature in attribution models and multi-channel funnels to build custom groups of channels that match your marketing objective. And yes, it's easy to exclude tweets from Rand.
--You can segment users (using Conversion Segments) out of multi-channel funnels. This makes it easy to exclude daily "users" that may not be valid in your model.
--You can use the models to measure both macro and micro conversions.
Overall we're all moving towards a time of user-based analytics. But until we get there I think everyone should take a look at closer look the multi-channel funnels and attribution models.
Thanks, Justin. I agree that GA's gotten much better at attribution mapping in the last couple of years - I think Mike's MozCon presentation has some outstanding tips on how to do that.
Assisted Conversions is certainly important to measure. I didn't include it because I assumed it was such a basic part of attribution modeling, it would be too basic for the "advanced" metrics I was outlining in this post - but it's worth calling out.
Justin, Where do you see the value of assisted conversions vs. using the attribution modeling tools and looking at things like assigned value with the various models?
Hi RuthBurr!
Its really informative post in perspective marketing analytics.
I really wan say thanks for the topic where you have constituted about tracking brand identity and how exactly campaign of branding is performing at search engines through marketing analytics. And another the list of advanced matrices of attribute modeling are really excellent to measure marketing analytics.
Thanks & Regards