B2B marketers use attribution to understand how their marketing impacts down-funnel sales. It enables them to see which marketing efforts successfully engaged and moved prospects through the funnel. Attribution data — the number of leads, opportunities, revenue, etc. that were generated by each channel, campaign, ad, etc. — then, is the foundation for measuring, analyzing, and mapping the customer journey.
To make sense of attribution data, marketers apply attribution models. Most models are built around a set of rules — in a first touch model, the first touch gets 100% of the credit; in a last touch, the last touch gets 100% of the credit; in a U-shaped model, the first and last touch split the majority of the credit; etc. (See all attribution models here.) This is where critics of attribution tend to jump in. Many believe attribution models built around a set of predetermined rules simplify the customer journey to the point where it is no longer valuable.
Of course, each model attempts to balance two factors: 1) accuracy and 2) ease of implementation and use. On one hand, single touch models are relatively easy to implement and understand, but they clearly don’t represent the full customer journey. On the other hand, it takes a sophisticated marketing organization and attribution solution to accurately track each touchpoint and make sense of a more-complicated, multi-touch attribution model. This brings us to algorithmic attribution – the most advanced and perhaps most complicated method for modeling attribution data of all.
What is algorithmic attribution?
Algorithmic attribution is one of, if not the most advanced ways to model attribution data in order to most accurately represent the customer journey. According to a report (gated) by AdRoll, 96% of respondents said that algorithmic attribution is at least somewhat effective, the highest of any model methodology. Of course, algorithms tend to be proprietary so what factors are considered in the algorithm and what weight each factor gets can vary by attribution provider.
However, the most accurate algorithmic attribution models should use machine learning to intake all of your data — all of the touchpoints, both historical and going forward, that went into closed-won deals, closed-lost deals, deals that fell apart at or before the opportunity stage, etc. — to create a model that is specific to your customer journey.
The algorithm then creates custom weights for each of your stages to represent how your prospects go through the funnel. It’s important to note that it should also use new data as you continue to engage prospects and close deals to refine and improve the model, which is the machine learning aspect.
For example, if your historical data shows that getting prospects to MQL is an important indicator of success, it may get 30% of the eventual revenue credit. If your historical data shows that getting prospects to MQL means very little to the eventual outcome, it may get 5% of the eventual revenue credit. Algorithmic attribution is tailored specific to your funnel and your customer journey.
Here’s an example of what an algorithmic attribution model could look like compared to a W-shaped model:
Why should B2B marketers care about algorithmic attribution?
Increasingly, B2B marketing success is dictated by who knows their customer the best. Again, it boils down to who can provide the right content at the right time to the right people. Accurate attribution data gives marketers a big leg up.
Algorithmic attribution gives marketers the most accurate picture of the customer journey as possible. While advanced models like W-Shaped and Full-Path get close to modeling a typical B2B customer journey, they are not going to be as accurate as an algorithmic model that’s customized using historical customer data.
With algorithmic attribution, marketers can see what stages in the journey are most critical to their prospects, as well as what channels and content are performing well at those stages. If the algorithmic model shows that the ‘demo scheduled’ stage is critical and the attribution data shows that Ebook A is great at driving the demo request, the marketing team knows to invest more resources there.
How do I know if algorithmic attribution will be significantly different from my multi-touch model?
There are two components to this:
- How different will an algorithmic model be from an advanced multi-touch model, say, a Full Path model?
- How significant does the difference need to be for the change to be worthwhile?
As to the first component, it depends on your company, of course. If you have a straightforward marketing and sales process, it’s less likely that an algorithmic model will turn up something surprising. If your marketing and sales funnel is long and complicated, the chances are greater that a rule-based model will be insufficient, even a Full Path model that takes into account four key touchpoints along the customer journey. Finally, if you want to see your marketing and sales teams’ performance at more specific stages of the funnel — e.g. lead, MQL, and SQL instead of just lead — an algorithmic model will allow you to do that.
Once you figure out the answer to the first part, you have to determine whether the incremental improvement of going from a rule-based multi-touch model to an algorithmic model is worth making the change. More than just checking a box that says you want to use a different model, you must consider the ramifications of changing how your attribution model represents your marketing performance.
Let’s say that the algorithm determines that your stage percentages should be adjusted by 5% — is it worth making the change? It depends, of course. It depends on how much that 5% adjustment to stages leads to changes in what percentage of credit goes to each channel. It depends on how much you’re spending on each channel. It depends on whether your team has bandwidth to make changes.
For some organizations, a 1% change in how the attribution model gives credit to funnel stages could mean shifting tens or hundreds of thousands of dollars from one channel to another. Then, of course, making the change would be worthwhile. For other organizations, even a 20% change in the attribution model may not be enough cause for action.
Our experience
At Bizible, using an algorithmic attribution model showed us that we need to account for and give credit to four additional stages in the funnel. It also decreased the weight given to two stages by over 10%, and gave over 20% to a stage that we hadn’t previously considered in our W-shaped and Full-Path models.
While these changes seem big, they didn’t have a profound impact on our strategy. The data showed that channels that we previously thought were generating a lot of value continued to generate a lot of value.
The impact that it did have, however, was that it moved credit that was going to Direct and Other to different marketing channels, which gives us more accurate and actionable data. One of the reasons for this change was that the algorithm moved weight from the Opportunity Created stage (successful demo) to the Demo Scheduled stage. Ultimately, switching to an algorithmic model gives us richer and more accurate data about how our prospects (and eventual customers) move through the funnel.
Takeaway
Attribution is intended to help marketers see what efforts are working and which aren’t by representing the customer journey. Through an attribution model, marketing efforts that move prospects through the funnel are given credit. Simple attribution models often fall short because they oversimplify the customer journey to the point that the data is unusable. Algorithmic attribution uses customer data and math to create an accurate model of the customer journey.
If you’re not using attribution because you don’t believe the models are accurate enough, algorithmic attribution may be the antidote.
At the other end, if you’re already using an advanced multi-touch model, algorithmic attribution may be the next logical step in getting even more accurate data.
