Early in my career, I learned to gather requirements rather than take orders. It only took me a couple of times to adjust my approach. There’s nothing worse than working on a report to hear, “I can’t use this.” I wanted to yell, but this is what you asked for! Instead, I took a deep breath and asked, “What exactly do you want to answer?
And then, I asked the same question in a few different ways until I got the needed answer.
“Who’s asking for these numbers?”
“What problem do they think we have?”
“Where do you think the problem actually is?”
“What questions do you want to answer?”
The result was a report with several more metrics than they requested, and I delivered it with the context I gathered along the way. It was satisfying to finally hear, “This is what I needed.”
People don’t mislead us through their requests on purpose. The thing they are asking for may have solved their problem at past companies. They may know they have a problem and think they know how to confirm it, but they’re looking at the wrong data.
Asking more questions is a great way also to understand whether this is genuinely a priority or if they’ll likely forget about the request in 24 hours.
While operations professionals are probably nodding along to this article, it’s an important reminder. Many prospective buyers hesitate before answering, “What are you trying to achieve with attribution?” Particularly with a project fraught with potential missteps and pushback from executive stakeholders, it’s critical that operations professionals do a lot of legwork before running an attribution project.
“WHY” Should Drive Attribution Modeling
At CaliberMind, we offer multiple attribution models because they answer different questions. We’ve since learned that our discovery process needed to run deeper to avoid misalignment down the road.
It’s normal for customers to want to use attribution for multiple reasons, which drive customizations vital to fulfilling those use cases. It’s also typical for companies to need more than one model.
For example, we’ve found that executives are looking for a better way to determine who should get “credit” for pipeline and bookings. They’re moving from a single-touch (usually Last Touch) model and manual intervention that declares a single team responsible for sourcing an opportunity. Multi-touch attribution often seems more logical, but it only fills this need if you are working with finance, your sales team, and others to figure out which activities should be tracked to make the model palatable to everyone involved.
Another common use case for attribution is trying to calculate return on marketing investment (ROMI) to fulfill a request from finance. Hearing that this is a goal should create a cascade of subsequent tasks that involve discussing your plan with finance to figure out how they think about ROMI so your model isn’t thrown out the minute you present it.
As an aside, I’ve seen this happen all too often because a marketing leader insists on using pipeline to calculate ROMI instead of bookings, or they’ve neglected to consider sales activity and effort. Finance estimates Customer Acquisition Cost as a blend of sales and marketing expenses, so why would they settle for calculating return on a number that isn’t final (pipeline is aspirational, not realized) and entirely “credited” to marketing (instead of bookings split between departments)?
As a final example, sometimes marketers want to explain why their specialty is business-critical. Creating correlation models for content and email marketers that show which content resonates with the people who buy is an excellent start for supporting these organizations without bloating your marketing-credited attribution, thus protecting ROMI by not blowing out the denominator.
“Why” Prepares Your Team for Future Asks
I recommend asking your team to define attribution. You’ll get many answers. Some think it’s a complete footprint of your buyer’s digital journey. Some think it’s a stage-bound efficiency engine. Others think it’s a way to defend marketing spend.
Asking what your team wants to use attribution for will uncover many reasons why they want to use it. Some of those reasons won’t be a good fit for attribution and will necessitate some education sessions for the broader marketing team. For example, tracking digital touches is a land field filled with privacy laws, ad blockers, and client-side privacy first design. We can’t track many things through direct touch-to-person mapping, but we can draw correlations that are almost as useful.
Understanding what your executive thinks attribution is capable of also helps you prepare for the requests that will come into effect after your model has been developed. It’s also an opportunity to level-set on what is and is not possible and can kick off a conversation about shortcomings. Your executive will be better off in the board room and beyond if they understand that attribution is a good faith estimate, not an exact representation of your buyer journey.
Finally, you’ll also get some excellent material for building a business case for an additional headcount. Analytics tools like attribution often create more questions than answers, which means an analyst would be busy answering questions the current model can address and building out requirements for future models.
“Why” Should Be Used to Gate Machine Learning
If you’re tackling a DIY attribution modeling project and the words “machine learning” come up, it’s your reminder to ask, “WHY?”. What is the gap they want machine learning to fill? What is the winning population they want to reproduce? What questions are they trying to answer that they think machine learning will help them do better?
Machine learning is fabulous. We use the Markov Removal effect to determine whether discontinuing an activity will harm the selling cycle, which is helpful when evaluating just about anything we spend money on. However, machine learning requires high data volume, and you need to decide whether you throw in some recency bias.
Frequently, marketers see machine learning as a way to get easy answers we’d otherwise miss. What we see happen when marketers don’t get the answers they want is they distrust the model. People come into analysis with preconceived notions. They *think* something is happening, and they’re looking for a way to confirm that suspicion. If it doesn’t, machine learning can be thrown out as “a black box” they don’t understand.
We’ve found that digging into what we’re trying to avoid or solve and then focusing on incorporating the “correct” influential touchpoints is far more vital to adoption than advanced modeling. Advanced modeling can even create unnecessary friction in the adoption process.
Every model has frequency bias, and what works changes over time. Therefore, adapting the data is a necessary first step before incorporating machine learning.
“Why” Should Foster Cross-Functional Buy-In
I can’t remember when I asked a company to describe what they wanted to use attribution for, that I didn’t receive an answer that involved sharing those numbers with another department. Usually, they want to defend their investment, which requires finance to trust the model. The rest of the time, they want a more advanced way to think about which departments get credit for pipeline and bookings. Which requires buy-in from everyone.
I consistently fail to see a project framework that includes sitting down with cross-functional stakeholders. They’re rarely brought in until after the work of building a model is done. And if they don’t feel like they can give input, chances are high that they’ll feel the need to shoot down attribution.
Emotional buy-in is just as necessary as the thought and care put into building out attribution. Whether we like it or not, humans are emotional decision-makers. If your attribution work represents ROMI or the “share” of marketing’s impact on pipeline and bookings, start gathering stakeholder requirements, develop your model to meet those needs, and sell people outside of marketing on the idea of attribution.
The critical thing to remember in operations is that you know the data far better than anyone else in the company. You should know how the data flows through different systems, and it should be straightforward to confirm or deny that a data point is even collected.
It’s easy to assume someone more senior than yourself knows precisely what they want, and they may, but you will save yourself a lot of time and heartache if you add a layer of questions before moving forward with a project.
This approach isn’t always well received. But, as they say in marketing, it’s all about positioning. Communicating your goals and collaborating with people outside of marketing from the start will give you far better odds of succeeding.
