As far as marketing attribution models are concerned it’s easy to fall into the trap of more is more. With so many ways to track, trace and justify your spend – it can be hard to understand which method of attribution will actually work for your business in the real world. Which explains why clients often come to us convinced that using data science is the answer to all of their attribution woes.
Don’t get us wrong, we understand the pain of today’s increasingly complex digital marketing funnel; having left businesses with non-linear multiple touchpoint customer journeys to monitor and cry about late at night. So it makes sense that businesses are hungry for a mysterious, artificially intelligent saviour to come to their rescue and make sense of this marketing madness.
But despite the obvious appeal of air tight, data deduced insights that will improve overall decision making – complex attribution models simply aren’t suitable for everyone. In fact, by our measures a startling 90% of companies would see no benefit at all. Because not only do these models require an exceptional amount of time, money and team spirit to implement…they are also unlikely to deliver the kind of clarity you seek at all!
Not quite convinced? Still think that a complex attribution model is the positive change your psychic predicted for 2021? Allow us to debunk such notions, starting with a breakdown of the two categories of attribution model you should be aware of:
- Rule-Based Attribution Models
Fondly dubbed “use your own logic attribution” because these attribute credit to marketing touchpoints using a simple predetermined rule, such as a customer’s Last Click before purchase. Within this framework you have the option of crediting that purchase to a single touchpoint OR distributing across multiple touchpoints throughout the customer’s journey.
Rules such as First Click are most useful for highlighting touch points that are effective in the awareness and nurturing stages of your marketing funnel. While Last Click focuses purely on what drove final conversions. Simples!
The shorter your sales cycle the more you’ll benefit from a Rule-based approach because it’s a straightforward representation of what really happened.
- Algorithmic Attribution Models
Often referred to as “complex attribution” due its use of data science and the blackbox nature of how it attributes credit. These models credit based on statistically significant customer behavioural trends, rather than the aforementioned simple rules. Meaning that it’s actually hard to know why a touchpoint has been credited at all!
If you are a massive organisation with a long sales cycle and data out the wazoo – then you may be in a position to reap the rewards of an algorithmic attribution model. Otherwise, probably not.
Why You Should Avoid Complex Attribution Models
Now that we’ve outlined the differences between rule based attribution models and algorithmic attribution models, let’s take a look at the reasons why a complex approach is most likely wrong for your business:
1. Huge volume of data
Reason to avoid: you don’t have enough data.
Complex attribution requires a huge amount of data to develop any kind statistical insight. That means you need to have enough granular historic data from across a broad range of sources to make it meaningful.
But why does it require so much data?
Without sufficient source material, it doesn’t know what “normal” customer purchase behaviour looks like for your business and will therefore be unable to provide an accurate prognosis for where credit is due. Not to mention that unlike with Single Touch attribution, these models cannot compute single user level data, or even single campaign level data. Full stop. Instead they calculate through data aggregation. Which means more is more.
That being said, it’s worth mentioning that these models also have computational issues with scaling. In fact, a complex attribution model is a bit like Goldilocks. Hungry, but picky. Too little data won’t fill it up enough to provide statistically meaningful insights, but too much data will have it bursting at the seams. Meaning that if you’re not already set up with the right infrastructure and team to handle this volume of data, then results will start to take longer to arrive, with insights and actions being ever further delayed!
2. Lack of seasonality
Reason to avoid: You will miss the bandwagon.
Due to the fact that algorithmic attribution models can only focus on and report on activity over the last 90 days, they won’t be useful for helping you to react quickly to any new changes that emerge. Meaning that if a competitor started doing something different in the last 3 days, you’d be none the wiser and far too late to jump on that trend or alter your marketing activity accordingly.
When it comes to interpreting data, context is everything. But unfortunately an algorithmic model can neither compute nor consider the wider context at play. Now imagine you wanted to place a bet on your favourite football team. They have performed impeccably all season long, but you’ve just heard that their star player is injured. This would likely affect the outcome of the game and you might make the call to change your bet as a result. This is what you do with a Single Touch model. Review performance and react. With an Algorithmic model you’d see the team’s last three months of stellar performances and have no visibility as to any other context – meaning your recommended bet would stay the same. Shame you lost it all!
3. Time and labour intensive
Reason to avoid: It’s not time well spent.
Rome wasn’t built in a day, and neither are complex attribution models. In fact, by the time you get to onboarding – 6/7 months of your life will have vanished. During which time your Data Scientists, IT team and Engineers will have been focused on the build and implementation of this super complex project – and not on making use of your data to hit marketing KPIs and drive the business forward!
Aside from the time commitment, these models will instigate a cascade of followup work. Because once you change the way you attribute outcomes, this will inevitably lead to changes in the way you allocate budget, hire team members and prioritise projects. Which means it’s more of an endeavour than you may realise. Making this switch something that most mid-sized companies won’t be able to achieve overnight.
The value from data is compounding. The more you get, the more you can use and ultimately the more you can then action. Meaning that the sooner you start paying attention to it, the quicker you’ll be able to derive value from it. Can you afford to wait half a year before you start tracking your marketing performance? Probably not!
4. No guaranteed results
Reason to avoid: You might walk away with nothing.
You would think that once a complex attribution model is up and running, it would be pretty darn sturdy and good to deliver on some powerful behavioural insights. After all, it requires a lot of time, resources and budget to get to that stage.
Sadly you would be wrong!
Because unfortunately you can’t know whether these models actually work until you put them into production and see if they break. Even if your team is data savvy and everything is clean, tidy and well implemented on your end, you will still have no visibility on any issues from the external data source itself. Something as seemingly unimportant as them changing a single data field’s naming convention might cause a mismatch that immediately starts to skew your results.
Brands operating both online and offline will also find themselves in a tricky situation when relying on an algorithmic attribution model. Because online journeys resulting in offline purchases will be categorised as non-converted paths, heavily distorting the model and feeding you inaccurate results that will push your team further away from the truth.
5. A problematic blackbox
Reason to avoid: You won’t know how you got those results.
You may not realise that a complex attribution model involves feeding your data into a blackbox, with ZERO visibility as to how it is processing your information.
At first that may seem like a long awaited relief. Finally, a data driven machine can make the calls for you! In practice however, while the model tells you which touch points were statistically effective in getting your customers to buy in – it is almost impossible to understand why a given channel has been attributed a transaction. Neither is it possible to measure whether an increase of spend on X channel resulted in a change in performance. Making this dark and mysterious model exceptionally hard to act on.
Despite its initial mystic-like appeal, your business will almost inevitably lose trust in a model it can’t truly understand and stop using it – resulting in a lot of wasted time and money. As for the insights produced by that blackbox – they will be long, complex conversion paths that non technical humans have little hope of deciphering. Leaving you with sage advice that is utterly in-actionable.
Where to start with Attribution
Perfect attribution doesn’t exist. We just don’t have the tools or data to capture the realities of each shopper’s journey from awareness to purchase. But if you understand the limitations of your approach, you can work with it instead of tearing it down for its inadequacies. After all, attribution models are flawed like the rest of us. So just start somewhere and then optimise as necessary.
The best way to do that? By starting simple. We’ve seen countless e-commerce companies successfully use simple attribution models such as last and first click. These models are simple to implement, simple to action and simple to understand for all stakeholders. Which gives them credibility across the business and therefore buy-in. Many businesses may choose to use different attribution models for different channels depending on the purpose of the campaigns within a channel. Our key criteria for choosing an attribution model is just as simple:
- It must be very understandable
- It must be appropriate for the business strategy
- It must be implemented consistently over time
- It must be highly actionable
With that being said, the process for starting is pretty straightforward:
- Start by tracking Last Click
- Enable session level data collection in your Google Analytics configurations
- Monitor and track this data for a period of time
- Introduce First click
- Monitor and track this data for a period of time
- Now reassess whether you are deriving value – if yes continue on as before. If not, you may need to step back and reassess your approach because a complex attribution model at this stage is likely to hinder rather than help your business.
At the end of the day the difference between using a Rules-Based attribution model vs. an Algorithmic attribution model, is trusting your own judgement vs. that of a machine. While the latter may sound appealing, even a machine will fail to produce satisfactory results when forced to operate outside the right set of circumstances.
Above all else, your attribution model needs to be actionable by your team, for your business. If it isn’t, excitement will soon wane and your next attempt to become data driven will likely be met with more roadblocks!
