As a marketer working for a SaaS marketing technology company, I am lucky enough to be surrounded by data scientists. We talk a lot about marketing attribution and, even though my undergraduate degree was in math, there were some terms that were new to my lingo.
The explosion of data and digital marketing has made far more sophisticated attribution possible. Digital channels, such as paid/organic search, display, and email, can be tracked with much greater precision and granularity than traditional, offline ads.
A is for Algorithmic
Algorithmic attribution is the process of assigning a portion of credit for a conversion to each touchpoint based on effectiveness. The key differentiator of algorithmic attribution is its use of advanced statistical modeling and inferences to determine an optimal, custom model that continually refines itself based on your data – put more simply, human assisted machine learning.
B is for Bias
The models you use for attribution can introduce bias. For example, last click is biased in favor of channels that appear later in the buy cycle, such as coupon sites that often attract customers right before they buy, though they likely would have bought anyway.
C is for Channel
Attribution is about examining all the various channels that are part of the customer journey – both online and offline. Online channels include search, social, display, affiliate, email and so much more. Offline channels like print, television, radio and outdoor are equally as important in an omnichannel customer journey. The nuance here is that the offline channels must be addressable, i.e. they can be traced back to an online visitor in order for the offline channel to appear in the journey. At the aggregate level, both offline addressable and non-addressable explain overall customer response to marketing stimuli.
D is for Data
When it comes to data I believe in garbage in, garbage out. The data used in attribution modeling needs to be harmonized and cleaned to a common level of granularity so that it is useful. Utilizing data from various sources guarantees disparate data and finding a way to correlate it is critical.
E is for Even Model
Even, or linear, model applies an even percentage breakout of revenue across all touchpoints in the customer journey to purchase. Optimizing marketing channels based on an even model means that the advertiser is rewarding frequency alone but not any external factors such as seasonal or macro-economic factors. An issue with this model is that diminishing returns and relative channel effectiveness are not accounted for as all channels and path positions are credited equally so more spend leads to linearly more conversion.
F is for Forecast
Attribution is no longer about just looking back to see what led to the desired action, it’s about being able to forecast how shifts in spending will ultimately affect your revenue. Forecasting, or marketing mix modeling, is a great tool to help marketers determine optimal media investments.
G is for Granularity
User-level customer journey data provides a level of granularity that isn’t part of marketing mix models (MMM). The ability to construct the exact sequence of touchpoints leading to a conversion provides a level of insight that can identify correlations between channels and make it possible to optimize your integrated marketing strategy.
H is for Homogeneous
Disparate data is the root of all evil when it comes to attribution. Mapping data from various sources into a single source of the truth is necessary to establish a homogeneous data set for modeling. Don’t start modeling until your data is homogeneous.
I is for Integration
With so much data available these days, the challenge is to consolidate it all and extract clear, actionable insights. Finding a platform that can systematically integrate data from various sources will help to tame your big data madness.
J is for Journey
The value of attribution is to examine the journey that led to the desired action – this includes cross channel (online, offline) and cross device (desktop, mobile, tablet). 79% of users own three or more devices. Recent studies show that users switch between devices up to 27 times per hour.
K is for KPIs
A KPI, or Key Performance Indicator, is a measurement that will directly affect your marketing objectives. They can be identified by examining the correlation between touchpoints and actions. Every business has unique KPIs so be sure you are measuring the most meaningful metrics to make more educated marketing decisions. One of my favorite KPIs is ROAS (return on ad spend). ROAS is a measurement that evaluates gross revenue generated for every dollar spent. The math is simple if you have the tracking data you need. ROAS = revenue from ad campaign, minus cost of ad campaign, divided by the cost of the ad campaign.
L is for Last Click
Last Click attribution assigns 100% credit to the final touchpoint (i.e. clicks) that immediately precedes a sale or conversion. While last click is important in identifying the closer, marketers should be sure to also examine the introducer (first click) and influencers (middle touches) as well.
M is for Modeling
The value of attribution modeling is comparing various models to determine which is the best fit for your goals and gain a more holistic view of each media’s contribution. There is no one perfect model, an organization should continuously update their models and examine their ability to predict future performance.
N is for Non-Marketing
A strong attribution model will take into account non-marketing elements such as seasonality, major holiday events, macroeconomic factors and competitive activities which can also greatly influence sales.
O is for Omnichannel
Omnichannel is defined as a multi-channel sales approach that focuses on an integrated shopping experience across all channels. Customers may encounter many touchpoints and move between online and offline channels, such as ordering online for in-store pickup. Each channel’s role is considered in relation to others and the customer experience is designed to be seamless and consistent.
P is for Position-Based Model
The Position-Based model attributes 40% credit each to the first and last interactions and the remaining 20% is distributed evenly to all the interactions in the middle.
Q is for Query
Most people think of search when it comes to a query – the search query is the word or the string of words entered in the search box of a search engine to access some information on the web. The study of search query trends is key to search engine marketing (SEM) optimization.
R is for Rules-Based
There are various rules-based attribution models such as last touch, first touch, first and last, position-based, time decay and linear. The difference between rules-based and algorithmic models is that rules-based applies the same rules to each and every conversion while algorithmic learns from your data to continually refine a custom data-driven model. Algorithmic takes into account correlations between your media and even external factors like economic conditions and seasonality.
S is for Statistical Significance
In the simplest of terms, when a statistic is significant, it simply means that you are very sure that the statistic is reliable.There is a lot of math behind this calculation but what you need to understand is the “p-value” which represents the probability that random chance could explain the result. In general, a 5% or lower p-value is considered to be statistically significant.
T is for Time Decay
Time decay models reward the media touchpoint closest to conversion most and the others receive less credit the earlier they are in the path. Time decay models reward path position regardless of the relative effectiveness of each of the channels in the customer journey. Just like a linear model, they ignore external factors.
U is for Unique Revenue
Sales revenue driven by a channel that you would not have received if it was not for that channel’s sole contribution is what we call unique revenue. With the complex channel overlap of today’s marketing environment, you want to measure each media’s unique revenue so you know where you’re getting the most bang for your buck.
V is for Validation
The statistical model used to generate attribution findings should be validated with in-sample as well as holdout sample, or “control” (a sample of data not used in fitting a model) – the holdout sample is used to assess the performance of the models.
W is for Well-Documented
Your attribution model should be well-documented and validated using best statistical modeling practices; you must be able to explain how your model works to your stakeholders and get buy-in. Be wary of mysterious black boxes.
X is for X Marks The Spot
Admittedly, X was a tough one, but one of the most important things with attribution modeling is understanding the problem you are trying to solve. Your models should be very focused on estimating the correlation between activities and desired outcomes. Are you trying to maximize sales, minimize costs or streamline the customer journey? Make sure you know the goal of your model.
Y is for Yield
It’s all about the results. It can take a lot of time and effort to build a statistically-significant, accurate attribution model but don’t lose sight of the ROI. Keep your investment balanced with the potential savings or increase in revenue. You wouldn’t spend $100,000 to save $10,000 now would you? But if you have a large revenue stream, a 1% improvement in close rate can easily pay for your investment in attribution.
Z is for Zealot
If you are passionate about marketing attribution, consider yourself a zealot. You’re not alone. Great minds think alike, and I am a true believer that marketing attribution is essential in today’s complex omnichannel marketing world.
Now that you know your ABCs, you can get back to the fun stuff. Be grateful that we have a way to truly show the value of our marketing efforts when in days past it was more of a guessing game.back to all blogs