Applying Machine Learning to Multi-Touch Attribution Models
Your new media marketing planner may well incorporate machine learning to optimize multi-touch attribution models used in PPC advertising in 2017.
Tomorrow’s successful digital marketing agencies more likely will be those that can leverage data and apply insights from machine learning to optimize multi-touch attribution models.
Exploration of new innovations in Machine Intelligence at Google helped to position the company as the second more influential brand worldwide as we head into 2017. The search giant is actively exploring advancing aspects of machine learning from more classical algorithms to cutting-edge techniques such as deep learning. Given that businesses that advertise using Google AdWords continue to gain new wins in paid search, grasping how machine learning and multi-touch attribution models work together is helpful.
This involves work on language identification, speech, translation, visual processing, ranking, and predictive capabilities that take advantage of Machine Intelligence. Copious volumes of big data that lend evidence of direct or indirect relationships of interest are used to develop learning approaches to understand and quickly produce more accurate search results.
Advances in machine learning are forcing digital marketers to evolve – whether SEOs are aware of it or not. Or whether you prepare your website for it or not. This article will reveal why we prefer using position based modeling in multi-touch attribution.
The major change is this: media planning and buying companies know that they have emerge with a fresh take on advertising strategies to be successful in increasing the volume of online sales.
Understanding Multi-Touch Attribution Models: Rules-Based vs. Algorithmic
Accurately determining your brand’s attribution model can be lumped into two categories of models: rules-based and algorithmic machine learning. A rules-based model is humanly deducted, and therefore could be labeled as fairly subjective. These more traditional attribution models are based on commonly understood assumptions about first and last touch, equal touch, time decay attribution, and other scenarios. For example, it was the last exposure that drove the person’s purchase decision or the first and last touches are given equal weight.
o improve on this subjective process, in case perceptions are misaligned, multi-touch attribution modeling that relies on the analytics maturity scale is more descriptive and involves a lower consideration process while leaning on machine learning. By replacing a single touchpoint attribution — last interaction and first interaction — to multi-touchpoint models, it is easier to assign credit to various touchpoints based on existing rules.
Given the advancements in machine learning, a data-driven algorithmic approach has proven to be more dependable. This means that attribution outputs are established based on data and the modeling of that data. Partial attribution is assigned by percentage depending on one touch’s value relative to whole. To be accurate, though, algorithmic machine learning attribution depends much on the richness of the inputted data. When a trusted source is used, a comprehensive snapshot is gained, but if not, results could be brutally flawed.
Even though we are relying on machine learning, this model still is dependent on a high degree of human interaction. The contextual insights gained from a human analyst help reduce the potential flawed inputs. A seasoned digital marketer is more likely to quickly recognize a mishap before it taints the outputs.
The multi-touch attribution model gives marketers a comprehensive look at the impact every user interaction has on the end goal of better understanding what’s working, what’s not, and what effects what in this giant, cross-channel puzzle. These insights vastly product the potential for marketers and brands to make smarter marketing decisions.
Eric Enge says that “Machine Learning can be used from SEO planning to winning purchases online when one correctly attributes user engagement”. The Feb 4, 2016 Moz article titled The Machine Learning Revolution: How it Works and its Impact on SEO, goes into depth on how the investment made by Google and other search engines in algorithm-based technology can be used. Check it out.
Should you use Multi-Touch Model Attribution for Your Business?
Machine Learning can be used from SEO planning to winning purchases online when one correctly attributes user engagement.
One of the best ways to benefit from data mining is to segment your customers and then send targeted messages using AdWords. And it’s a fairly straightforward process. From your Search Console and Google Analytics data alone, it is possible to break down your audience into meaningful segments like age, interests, income, profession, or gender. And this is effective whether you are running email marketing campaigns, paid search, or refreshing your SEO strategies.
Another benefit of data segmentation is the ability to better understand your competitive landscape. This insight alone will help you identify that the usual markets you quickly identify as well as ones targeting the same potential buyers that your business wants to serve.
If you are like the typical business that when asked to list their competitors, they come back with a list of individuals. Think and plan bigger. The search engines have to choose from a worldwide pool of potential matches for user search queries. Most businesses need to expand their circle of competitive website out two or three times if they really want to compete and win. Data mining and understanding more about how machine learning works will help you do that.
E Marketers article titled Google’s Journey into Machine Learning: What Marketers Need to Know talks about the diverse methods that exist with a top-funnel approach. This involves a strategic plan as to getting new and qualified leads coming into your website.
“The sessions and the visits to a business’ website are not all of equal value and a company shouldn’t be paying the same amount for all leads,” Justin Cutroni states.
8 Commanding Advantages of a Machine-Learning Model to Improve Conversions:
Make better advertising decisions starts with having accurate data and the ability to utilize machine-learning model updating with every conversion.
1. Easily understand how deep your site visitor’s conversion funnel is and how user’s make decisions.
2. Know when to optimize by seeing how long it typically takes users to convert based upon where they start and end by making a purchase.
3. Better able to determine where to spend ad dollars to increase ROI and business revenue next quarter.
4. Decipher the number of conversions based on the device used and the effect of strategic device targeting in distinctive parts of the funnel.
5. Better ability to see how your upper funnel tactics and use of video content plays into each conversion.
6. More accurate ability to understand the entire customer lifecycle for both new and returning business.
7. One advantage to algorithmic-attribution models is that there is less guessing early on.
8. It reduces the need to re-check your data in hindsight.
Use Machine Learning Time Decay Attribution to Allocate Market Spend
“Basically, the last campaign or impression that the lead interacted with gets all of the conversion credit. If you’re only running one or two campaigns, the simplicity and out-of-the-box accessibility of this model is appealing. However, in cases where you have multiple marketing channels, this sort of attribution is misleading and won’t help you better allocate your marketing spend,” states Segment Analytics Academy.
When using the Time Decay Attribution Model, the touchpoints closest in time to the consumer’s purchase or conversion are attributed the largest share of the credit. If for example, both an AdWords campaign click and Email channels are recognized as the most recent customer interactions within a few hours before the sale, another channel would receive less credit than either the Ad or Email channels. Since the Paid Search interaction occurred close to the time of sale, this channel receives significantly more credit than an earlier social engagement with an agent.
Using Google’s Data-driven Attribution Modeling
Google is relying more on machine learning which influences your user behavior data and arithmetic savviness to find an attribution model that’s best to maximize total conversion volume. It is a gold mine dream to locate a data set that would increase your AdWords conversion volume more than any competitors.
A challenge with in-AdWords data-driven attribution is that it best offers a single-channel interpretation of user engagement. To successfully target new customers searching for your product online, it’s hard to identify conversions that may have come through other non-search marketing work such as email or social networks.
Google suggests having at least a minimum of 20,000 clicks and 800 conversions taken within a one month time period. Many businesses starting out with AdWords have more modest numbers and do best with position-based attribution at the start. Finding the best attribution model for PPC for your business depends on many factors and is revealed over time and testing.
Challenges of Selecting the Best Multi-channel Attribution Model
Not many aspects of analytics are more complex than discovering the best multi-channel attribution model to use for your business.
How can it be that last-click is wrong after it being the default PPC attribution model for years? And then to learn that first-click attribution is erroneous as well may be hard to take. While we are moving toward position-based models even over time-decay, no data-driven attribution technique seems perfect. Your marketing PPC spend in the coming year may be increased with better results and more return for your investment in paid advertising.
The last touch attribution model is simple and often what many marketers fall back on. However, it completely disregards the stimuli of all ad impressions, bar the last one. Alternatively, the newer idea of the multi touch attribution (MTA) model has been adopted by many. This means that more than one touch point can each have a fraction of the credit based on a truer user influence at each touch point leading up to the user’s decision to make a purchase.
Due to new machine learning technologies, don’t be put off by former conceptions about it. Born from pattern recognition of search and site visitor behavior, artificial intelligence research work lets your business learn from data and your user touch points. It is possible to better visualize and credit conversion data. The iterative aspect of machine learning is important because as models are exposed to new data, they are able to independently adapt. They learn from previous computations to produce reliable, repeatable decisions and results. This is an old science, but one that is vastly improved and gaining fresh buy-in from savvy digital marketers.
While many machine learning algorithms have existed for years, the power to automatically apply complex mathematical calculations to your Google Analytics data at a faster clip and much more reliable is a recent development.
How to Use the Multi-Channel Funnels Analytics Top Conversion Paths Report
Examine the Path Length report in the Multi-Channel Funnels standard report in your Google Analytics (or similar platform). Often a high number of site conversions have greater than one path length, making it challenging to attribute the conversion correctly. Google Analytics lets marketers combine insights about the user’s conversion path that with its multi-channel conversion visualize found in the Overview section.
Review the Assisted Conversions report in Google Analytics. Focus on the last column which shows all Assisted/Last Click or Direct Conversions:
• For values under one, that channel has a higher probability of driving last click conversions.
• For values over one, that channel has a likelihood of being available earlier in the conversion phase. These channels may be missing appropriate credit when using a last click attribution strategy.
It is possible to decipher what areas may be undervalued across your marketing channels.
Then adjust a percentage of the budget allocation and measure fresh outcomes. Entire marketing teams benefit from knowing which channel comes first (the one that introduces your brand to the consumer”), which channel comes next (“nurtures our potential buyer decisions”), which channel comes after that and so forth.
Next, locate insights in Analytics Top Conversion Paths report.
Data Mining for Better Attribution Models
Segmenting and your database and testing how your insights gained shape Ad copy can increase business conversion rates and thereby profit margins. By making your PPC promotions off of a tight, highly-interested marketing plan, it can, empower your company to customize products and promotions to better satisfy the needs of unique audience that a more generic, broad promotion possibly can.
Smarter top-funnel approaches on how to gain new and qualified leads to your website may involve more mid-funnel focus. Profit-driven marketing includes predicting what causes churn, user engagement, and better retention rates, it is possible to reduce organizational silos and see businesses integrate their data sources. We have the ability to engage machine learning to improve data integration that can be used to drive new customer acquisition, reduce landing page bounce rates, and build return purchases from existing customers.
Many are attributing machine learning, customer analytics, and how big data can work together to get past reaching a certain type of buyer with a specific message on one type of device. After determining a trigger behavior that a site visitor performs, marketers gain more qualified insights on what to change in messaging, offers, frequency, time of day, and other factors to make paid advertising more efficient.
Position Based Modeling in PPC Attribution
Every business wants to be as thrifty as possible to gain maximum results from their advertising spend. By keenly examining to the bottom of your conversion funnel, last-click attribution merits most of many marketers’ efforts. While it is easy to understand why, when the goal is PPC efficiency and outward growth is less demanding of attention, a time-decay model is my preference.
However when the business focus is on growth and increasing new customer acquisition, closer to the top of the funnel is a priority for AdWords optimization. Both first click attribution and last click only offer tiny pieces of the pie. Google in AdWords and Google Analytics data may someday provide a reverse iteration of time decay, but since it’s not offered today, a position-based or u-shaped model takes more work and skilled interpretation of available data reports.
By focusing on predictive data and relevant purchasing information, it is possible to increase the relationships that consumers have with your products and services.
The U-Shaped position based attribution model makes it easier to identify key user touch points while simultaneously attributing a percentage of credit to earlier marketing efforts. When it business has a lengthy purchasing cycle, middle touch-points are often given less weight. Attribution modeling is completely arbitrary. Years of practice and an analytical mind help you to assign credit to the metrics and channels that are best aligned with to your business goals.
Cross-Device Tracking Study offers Insights for Measurement and Attribution
Improved cross-device tracking offers a more complete view into a user’s behavior and can be valuable for a range of purposes, including ad targeting, multi-touch attribution, research, and conversion attribution. However, site users are often unaware or often skip reading cookie notices posted at the bottom of pages on how and how often their behavior is tracked across different devices. Therefore a study by OTech gleans information about cross-device tracking and offers observations from the perspective of the end user.
According to a January 5th email communication from Aaron Alva at the Technology Division of the FTC, “Internet advertising and analytics technology companies are increasingly trying to ﬁnd ways to link behavior across the various devices consumers own. This cross-device tracking can provide a more complete view into a consumer’s behavior and can be valuable for a range of purposes, including ad targeting, research, and conversion attribution.”
For more information, read the newly released report from the FTC called Cross-Device Tracking: Measurement and Disclosure by The Office of Technology Research and Investigation (OTech )
“In the Position Based attribution model, 40% credit is assigned to each the first and last interaction, and the remaining 20% credit is distributed evenly to the middle interactions. In this example, the Paid Search and Direct channels would each receive 40% credit, while the Social Network and Email channels would each receive 10% credit.” – Google Analytics Support
NOTE: Google for filed its application on Feb. 15, 2013 for a method and apparatus for data-driven multi-touch attribution determination in multichannel advertising campaigns. Otherwise known as US 20140236705 A1.