From the list of 2012 shameless buzzwords, attribution analysis is the one that really interests me and is a hot topic amongst most senior ecommerce professionals I know.
Because there isn’t a right answer for how use attribution analysis. During the attribution modelling sessions I moderated on for Econsultancy’s Digital Cream event, a constant theme was uncertainty about how to start using attribution and how to apply it to the business.
For many there was scepticism about the reliability and validity of the data.
I’m hoping this blog kick starts a discussion……
There is a common acceptance that attribution analysis is important and should be invested in. People don’t behave in a linear, uniform pattern. Knowing how each thing you do contributes to task completion is really, really helpful.
It can mean the difference between cutting activity that improves KPIs and focusing your budget where it delivers results.
I think there is a lack of clarity about which challenges attribution analysis can be used to tackle. I don’t mean the end goal, which surely is to drive conversion and revenue, but the questions that play on your mind in the darkest hours of the ecommerce day.
You know the ones I mean, the ones that feel easy enough to answer but the process of doing it will be more complicated that you really want, so they sit on the post-it note, unloved.
At Think Vis, Russell McAthy, a digital marketing consultant doing some interesting things with attribution modelling, asked the audience three questions and got really interesting feedback:
- How many people are currently using attribution/or have a client who does? 40%
- How many people understand how the models work? 20%
- How many people are using the attribution to actually change the way the business runs? 0%
According to Russell:
Attribution has become a buzzword that people use to explain a misunderstanding of the influence outside the world of last click. The biggest benefit you can gain using the methodology of attribution is understanding the best combination of activity when managing incremental or reduction of marketing spend.”
I recommend tapping in to Russell’s knowledge stream by following him on Twitter @therustybear (which means something very different in Soho).
So having spent some time thinking about this in the context of retail ecommerce, I thought I’d get scribbling and share my take on what sorts of questions ecommerce teams should be thinking about when planning how to use attribution analysis to help them make investment/optimisation decisions.
Please feel free to questions and add your take via the comments below. Please note this isn’t an exhaustive list (we’d be here all day), it’s more a starting point for discussion.
A definition (indulge me)
Let’s start with my definition – attribution analysis is the process of attributing outcomes to the component activities that contributed to their realisation.
This doesn’t just mean revenue, nor does it just relate to marketing campaigns, it’s about understanding all activities and how they contribute to all financial and non-financial goals.
For example, it includes analysing how content assets on a website influence goal or task completion e.g. my banner version X on page Y is increasing newsletter sign-up completion by Z%.
Set one: general
Which marketing channels have the greatest influence on my total sales?
Using data to identify the top-level transactions and revenue contribution to:
- Last click conversions.
- Assisted conversions (where the channel was part of the sales path but not the last click).
Piecing the two together gives you a fuller picture e.g. social creates a lot of non-converting traffic but we’re seeing return path conversion from organic brand search.
Below is a Google Analytics screenshot for the multi-channel funnel report filtered to show conversion paths where email is an interaction.
What’s the conversion lag for different marketing channels?
You can look at the time difference, and number of steps, between first and last click to determine the tail of campaigns. For example, you might find that visits from social media drive more long tail activity than email campaigns.
The number of steps is interesting. The more steps between first and last click, the greater the indication that the initiating visit has sparked a research cycle.
You can drill down in to what campaigns and content initiated these longer paths, then look at the intermediate stages, to understand what content people are using to help them make a purchase decision.
How can I use this data?
It can help you rethink internal reporting. Instead of basing performance reports on the first X days, cast the net wide to pick up the full contribution.
I worked with a retailer where we changed the end of campaign reports to include a seven day and 28 day view and for some campaigns the numbers were vastly different [accepting that cookies mean the data doesn’t reflect 100% of real campaign impact].
How does each marketing channel create brand awareness without generating visible sales?
With cookies, tracking is imperfect. There may be return visits that fall outside the cookie tracking window, so first and last click aren’t associated and last click falls into a new visit.
So, to build a fuller picture of channel value, look at new visitor data from each channel where there was no goal completion. You should be looking for:
- Total number of visits and how this is trending over time.
- Bounce rate. If people are bouncing, you might not be attracting the right people (to get even funkier, you could look at single visit bounces – visits that bounce and then don’t come back).
- Landing page. Some pages (e.g. blogs) naturally have a high bounce rate, so think about the context of the page when looking at engagement – if you’re getting lots of new visitors to read blog content that could be beneficial.
- Direct traffic. Are you seeing increases in direct following spikes in campaign traffic? Direct is a good indicator of brand interest.
Please note this isn’t a ‘perfect’ picture of new visitors but it does permit trend analysis over time to see where you have growth points for each marketing channel.
What types of campaign are helping with task completion?
First, define your tasks. For example, on a retail ecommerce site a task could be downloading a digital catalogue that has been promoted via offline/online marketing.
By setting up this task as an event, you can track the visits to the page the event occurs on and the number of visits triggering the event, then break this down by traffic source.
How can I use this data?
It will help you tailor messaging in marketing campaigns, learning which channels are effective in driving which actions.
You can of course go more granular in attribution analysis and segment the data (recommended). For example, you could segment by device to look at task completion on desktop vs. tablet vs. mobile or by geographical region. The more granular you go, the greater the insight that will help tailor marketing activity.
What types of campaign content on third party domains help drive goal completion?
Let’s look at mobile. I tend to agree with the view that mobile is about task completion, not conversion. Yes, people do convert but on mobile devices conversion is generally much lower and browsing habits geared towards other tasks, for example finding stock in a local store.
So you can segment analytics reports to show mobile traffic from specific campaigns and/or domains. You can then drill down to look at goal completion data e.g. mobile banner on domain.com drove Y newsletter registrations with a goal value of £Z.
This is useful when deciding what types of content to place on each domain – you might find for some domains video ads are the best currency and on others video has no impact on click through or conversion.
To enable this level of analysis, you’ll need to use campaign tracking so you can tell which asset from which campaign on which domain has contributed to your goals.
Set two: digital marketing
Should I carry on spending on paid search keywords that aren’t converting?
This is a discussion I’ve had with many senior managers who look at the data, see lots of keywords with negative ROAS and understandably think, “I’m spending money and getting no ROI, I should spend it elsewhere”.
Another common complaint about paid search is that brand bidding is redundant and wasteful. If you’re nailing one position for SEO, you don’t need paid search.
I’m not saying either is definitely wrong, I just know from experience that most times it is. I’ve seen many keywords or ad groups that have a negative ROI when you factor in media cost, product margin and tax. However, when I look at the assisted conversions these keywords support, the ROI picture changes. I’ve even seen keywords where the value of assisted conversions is far higher than the direct revenue they are generating.
Then there is the lifetime value consideration – if these keywords are driving new visitor acquisition and you know the average spend for customers based on the media channel they were acquired through, you can start to understand the long-term benefit.
When you factor in these contributions, the investment decision can change.
Which email segments are contributing the most to my social efforts?
Your email opt-in list should represent one of your most engaged customer segments.
Visitors using social bookmarking and voting buttons are completing tasks and adding value by generating social proof. Content shared socially has positive brand resonance and can lead to visits to the website. This can be tracked, though not perfectly, by adding tracking parameters to the URLs that are shared.
Thinking laterally, information like this can help inform customer loyalty programs. Loyalty can and should be about more than rewarding buying, it should be encouraging people to complete tasks that have a benefit to your business.
How does offline PR drive online activity?
PR is generally considered brand awareness but it’s important to try and quantify the impact on your website.
There are ways to do this, including using QR codes on printed materials to encourage mobile customers to access your website. Where there is a digital trigger in offline marketing, visits can and should be tracked.
For example, if you embed campaign tracking in a URL that sits behind a QR code, you can use analytics reports to track visits and goal completion.
This is harder to quantify due to limitations on tracking offline to online activity.
What content is contributing to my KPIs?
Content is important for influencing visitor behaviour, so it’s useful to know how each content asset and different types of content asset affect KPIs. For example, you write a buying guide and see a lot of visits but little direct conversion. However, digging in to the data shows that this page attracts a high percentage of new visitors who then return to the website via other media and convert.
The analysis reveals a link between content that, at face value, contributes no revenue, and conversion in other channels. You can start to build a picture of its true value to your website.
First step is to ensure that your content is being tracked. There are three common options:
- URL parameters in links shared externally to content pages e.g. blog posted on Facebook
- URL parameters in links on social sharing buttons on content pages e.g. AddThis
- Events added to individual content assets on webpages e.g. product video, pdf download.
If you know when someone has ‘consumed’ a content asset, you can look at website visits that include this consumption and see what outcomes are affected. For example, a retailer I know tested product videos in their blog – they found there was significant click to product page and conversion uplift. Video became a higher content priority.
How can I use this data?
Blog optimisation is a good example. Learn what types of post lead to different outcomes and then refine your content plan to drive specific actions. For example, a retailer may find that a shopping guide with product insertions drives more online activity than other blog types.
What do you do with all this information?
Don’t make assumptions! To validate your hypotheses and ensure there is a proper correlation between activities, you need to embrace testing and modelling.
For example, a Client had a strong correlation between first click exact brand paid search and conversion via non-brand generic and paid search. However, the CEO didn’t think they should spend any money on brand paid search.
We ran a test – benchmark data (last click & attributed sales), then paused one of the key PPC campaigns and used negative match on other phrase match ads to block out the keywords.
We found, unsurprisingly, that the loss in transactions, revenue and margin outweighed the cost savings. Plus, there was a minor downturn in direct traffic, so the full impact was probably more significant.
I wrote-up a fuller case study previously.
A word of caution – just because it looks like A causes B, don’t ignore the potential for C to also occur which could be a negative impact. Take the blog example, you focus on blog style A that has an uplift on orders and revenue. However, by reducing other types of blog, you end up attracting less total traffic and you find that your overall direct traffic starts to shrink
I think the best advice I’ve heard came from Lewis Jenssen at DC Storm: accept errors because the data will never be 100% perfect and focus on a few important questions that, if answered, would help your business make better decisions.
Then build and model the data, implement changes and learn. From this sensible starting point you can then increase the complexity in the model. For example, you might start by focusing on your main paid media channel (often PPC) and use attribution to help reduce ad cost and improve conversion.
What are the questions you are using attribution to answer?
I’d be interested to hear from other people and learn what questions you try to tackle with attribution analysis
If you don’t agree with something I’ve written, or you’re an analyst who can provide a more detailed explanation for any the questions discussed, please let me know – I’m always open to suggestions and advice.
I leave the best way to track and measure this data up to the specialists – ask any of the analytics and optimisations legends, people like @danbarker @optimiseordie @peter_o’neill @feiner @pritesh_Patel @fastbloke @timlb who are well worth Twitter-stalking.
OK, over to you now….