Who Needs Marketing Attribution and Why
Long gone are the days when it was acceptable to guess which half of your marketing was working. Today, you have depth and variety of data, optimized funnels, automation, and personalization. You have cross-device tracking, event tracking, user profiles, and integrated marketing stacks. You leverage machine learning and AI. You pay for analytics suites and multi-touch marketing attribution tools.
Yet all this technology and these best practices don’t always make your job easier. You still run into dead ends when trying to scale return on ad spend (ROAS). Often, your campaign optimization and budget allocation decisions still lack actionable customer journey data.
Can multi-touch attribution modeling solve these challenges? Can decisions become clear, even if it requires extensive work upfront? What concepts are important to understand first? What multi-touch models are viable? How do you decide if an attribution tool is a good fit for you?
This article will help you find clear answers to the questions above. We’ve rolled out attribution tools for large clients, cross-examined engineers who build them, compared the benefits of attribution to alternative marketing practices and tested the impact on ROAS. Read on to learn what we’ve learned.
Key Terms to Understand for Marketing Attribution
Before we dive deep into multi-touch attribution tools, let’s review a few foundational definitions.
Multi-Touch Attribution, Touchpoints, Conversion Values
Multi-touch attribution (MTA) is the practice of identifying a set of touchpoints that contribute to a conversion and assigning a value to each of the touchpoints. By doing so, MTA helps you understand the combinations of touchpoints, and their chronological order needed to get customers to take the desired action. Multi-touch attribution and marketing attribution are interchangeable terms.
In tracking and reporting, customer interaction events are often what define the touchpoints. But it is important to note that a touchpoint doesn’t need to happen as an intentional action. An ad impression in a display campaign can also be a touchpoint.
Attributing conversion value to a set of touchpoints leads to a better understanding of the customer journey, or the path to conversion. It can also lead to a much better analysis of the marketing mix, including the offline channels. Insights stemming from such an analysis are typically used to optimize campaigns and to level up the return on ad spend (ROAS).
What Are Lookback Windows?
The lookback window is the period of time before the conversion that you are willing to credit for the conversion.
The lookback window answers the question:
“Where did my leads, accumulated as of today, come from?”
If you’ve used Facebook Ads or Google Ads, the term will probably be familiar to you.
Most attribution tools (as reviewed in a section below) allow you set the lookback window, but not all do. This is critical because the lookback window is one of the biggest factors in how the multi-touch attribution model will work. If a vendor limits the way you can do this compared to how their date picker works, it could be a recipe for disaster.
We worked on an attribution clean-up project where the client purchased the attribution tool on their own, could not figure it out, and brought us in to help. To keep the story simple, we wound up having to fire the vendor because they did not explain their model to the client before the purchase, and the model would not work with the client’s business model. It was sad because the sales rep painted a rosy picture made up of pure, 100% therapeutic-grade snake essential oil, and the product was a complete mismatch for our client’s business. This is a key area where we drive impact for our clients. We are hired by top brands to select and vet tools that are proven to drive revenue for their business model and get results.
It is very important to understand attribution models and their windows because this is where the sausage is made. It is critical to how you will report on your data.
What are Look-Forward Windows?
When you build an attribution model, you need to look back in history to be able to tally the touchpoints and create a linear model to distribute the credit of the conversion. Well, to make this even more fun, some attribution tools give you the ability to also take into consideration the revenues that have come in after the first conversion.
The look-forward window answers the question:
“What did I get for my spend?”
To put this into context, let’s say you are an e-commerce site or subscription business. You may spend $90 to acquire a customer when you know their first transaction with you is only going to be $25. Because you know your lifetime value (LTV) metrics and repeat purchase rate, you know it will take 4 months to become whole on your spend. Since your first transaction is only $25, your ROAS model in Google Ads and other tools appears to be upside down. That said, any smart marketer knows you just use your average LTV multiplied by conversion amount to get your true ROAS.
The problem with that model is that not all advertising campaigns are average, and none of them have the same LTV. It is just the nature of the beast. While this makes it hard to track ROAS in some tools, companies like Attribution have a perpetual look-forward model that will take into consideration the LTV of a user. This model is a truly helpful differentiator for subscription businesses such as SaaS companies.
How Do UTM Parameters Work With Marketing Attribution?
Proper use of UTM parameters is essential to marketing attribution. UTMs enrich session data in a way that helps you track the user across the multiple touches. For the purpose of this article, we assume your knowledge of using UTMs is intermediate or higher. If you’re interested in additional resources related to UTMs or to get a tune-up on this concept, check out our popular UTM guide. For an effective way to streamline the UTM workflow of your team, check out the best UTM generation and automation tool out there, UTM.io.
The Different Attribution Models
Now that we have some basic definitions covered, let’s move on to attribution models. There are multiple ways you can structure your attribution models, and the one you choose will depend on factors such as your industry, content strategy, or funnel design. We’ll review some of the main attribution models out there, giving you the basic overview along with pros and cons for each.
First Touch Attribution Model
With first touch attribution, you attribute the cost and conversion to the source of the first visit that occurred within a look back window. As a quick refresh, a look back window is the period of time before the conversion that you are willing to credit for the conversion.
The conversion can happen at any point during or after your selected look back window. Importantly, it is very common to see conversions happen on a date that is outside the look back window.
Last Touch Attribution Model
When looking at last touch attribution, you attribute the cost and conversion to the source of the last visit in the look back window right before the conversion event.
Without marketing attribution models in place, last touch is usually the default way to assign conversions to touchpoints in ad platforms. Just like any model that attributes all conversion value to just one touchpoint, this model is often why marketers feel the need for more sophisticated attribution.
Linear Attribution Model
When using the Linear Attribution model, you attribute the cost and conversion equally to each visit in the look back window that subsequently resulted in a conversion.
Time Decay Attribution Models
When viewing the time decay attribution model, you attribute credit based on the number of days the visit was before the conversion.
For example, at McGaw.io, we use this model :
y = 2^( -x / 7 )
The x is the number of days prior to the conversion that the referral happened. The 2 in the equation is the half-life. A touchpoint 7 days before a different touchpoint will receive half the credit.
The result is that the last touchpoint will get the most credit and the first touch will get the least credit.
Position-Based Attribution Models
This model is best known from Google Ads and similar keyword bidding platforms. It gives 40% credit to both the first and last touchpoint, with the remaining 20% is spread out across the other touchpoints on the user journey.
Data-Driven, Shapley, Bayesian, and Markov Attribution Models
Beyond the attribution models above, you can also tap into models that are more dynamic and flexible.
Data-driven attribution models are common to Google Ads accounts with large data sets. They use machine learning to give credit to touchpoints based on how impactful they were on the path to conversion. Data-driven models, as coined by Google, do not represent just one specific model. They’re applied on multi-channel funnels in a variety of ways, including the Shapley distribution model:
A Bayesian model is a data-driven approach to attribution models used by vendors such as C3 Metrics. The concept is straightforward — by updating your initial beliefs with objective new information, you get a new and improved belief. The adaptive model doesn’t toss the existing equation for multi-touch attribution — it updates it when new patterns are identified.
The Markov model is an application of the Bayesian approach. (Interestingly, the lowest-cost tool on our list is the one that uses Markov prominently.) The approach looks at the touchpoints to conversions in chronological order and then simulates the impact of the removal of each touchpoint. Windsor.ai illustrates the Markov approach well:
How Do Custom Attribution Models work?
When reading into the data-driven models, you may wonder if machine learning really is good enough to determine how impactful the individual touchpoints are. That question has been on our minds, too, and we’ve discussed it with clients. Machines are more objective than humans, but they lack common sense and they don’t always understand the full picture. To reap the benefits of tailored weights for different touchpoints, you’ll want to consider building your own custom attribution model.
This is where things start to get really hairy, though. When building a custom model, the first issue you have is that everyone has an opinion on how the model should be built, and many times that opinion comes from a defensive standpoint when the model does not show the outcome they expected or hoped for. For example, maybe the model shows that a certain campaign wasn’t that effective in driving conversions. The person who advocated for spending on that campaign would likely argue that there’s a problem with the model, not their decision.
While custom attribution models are a normal thing, the big warning we have (one that is shared by all the vendors we spoke to) is that you should only be changing the model in the beginning and then try not to mess with it too much. Leaving the model alone will prevent people from creating self-fulfilling prophecies. This is super important.
You always need to use analytics to help guide your decisions. You’ll be throwing money down the drain if you use analytics to prove you’re right.
For an example of a custom attribution model, imagine a linear approach to the model where you distribute the touchpoints over the journey of a customer and add weight to certain channels. The concept is fairly simple. The hard part may lie in deciding which touchpoints to include and what weights to assign to them.
Let’s take an example of direct mailers that are a huge driver of applications for a financial service. While this will be a channel that sends a lot of applications, it is just the starting point. Many people from direct mail come to the site and don’t fill out an application, so then they get slammed with retargeting and possibly Google Ads. To make sure the direct mail’s impact does not get lost in the shuffle compared to the digital ads, it should get a higher weight in the model compared to other channels (because it is what drove people to those other channels in the first place).
We’ve done a lot of work to decide that off-the-shelf attribution tools and their standard models are more for the optimization of your campaigns, and they are most likely to be used by the people who manage ad spend. Meanwhile, if you are looking for historical reporting, which should be used by the executive leadership team, rolling out your own custom model can be better and give you a much higher level of control.
Get an amazing marketing audit, then have us roll out your attribution for you. Talk to us.
Use Common Multi-touch Attribution Models for Advertising Optimization Only
Attribution models are flawed in nature. They should be used for advertising optimization, not for historical reporting. The easiest way to think about attribution modeling is that it is directional, not definitive. If you go with a vendor compared to building your own model, you should be using it mainly to optimize your campaigns and not to look back at data from three quarters ago.
If you want historical reporting, you should probably build your own model using your warehouse and tools like Looker.
Why You Should Probably Build Your Own Attribution Model Using Your Warehouse
We feel attribution models should work by storing the cost data with the individual user. The problem is that if a user does not convert and the lookback window is over, you need to redistribute the cost associated with that user to the other users who did convert. While this is not rocket science, it is difficult to create a foolproof system to manage the migration of those costs from user to user.
The way all of the attribution tools work is by looking at periods of time and then tallying the amount of money spent in that period, and also the number of conversions that happen inside that same window. Some vendors do provide an algorithmic approach to conversions to alleviate the time-period flaw, but it is still not perfect.
The key issue with looking at a time period is that if you look at June 1st to June 30th, you can see your total spend for that period. When looking at this same window, you are also able to see the conversions that happen in this period. Since most conversion windows can take 30+ days from the first touchpoint for a user to convert, most of the conversions for the time period of June 1st to June 30th will actually happen outside of that time window.
Simply put, you will mainly be looking at conversions from the spend in the period before this time period. Even with lookback windows and look-forward windows, you are misappropriating the spend to the conversion, and the ROAS number is flawed.
While bringing this up to many of the vendors, only one of them had a really good answer for it, and that was Visual IQ (now rebranded to Nielsen Marketing Effectiveness). They were clear on the fact that you need to have the correct look-forward window and correct lookback window and only look at the attribution ROAS after your average conversion window would have closed. So if your average conversion window is 45 days, then you cannot look at the ROAS for a period shorter than 45 days ago, because the data would not be accurate yet. Moreover, the data should only be used for optimizing campaigns, and not historical reporting.
Which Attribution Tools Are Best for Marketing Attribution?
Our research into and testing of the attribution tools was not quick and easy. Just like we did with the rest of this article, we put an amazing amount of effort into the vendor review section. We had to get aggressive when demanding answers from vendors. We scared the crap out of some of the sales reps, and in all cases, we were escalated to the leadership team for accurate answers. We had multiple solutions engineers tell us they did not know, and they would need to get back to us with details. Many of them were not able to provide concrete documentation, and their workflow would not be able to stand the test of time with our data team.
We’ve compiled the following list of attribution tools to consider. We encourage you to try the tools yourself, but we also understand it’s hard to do so with more than a few. Please use the reviews below as the next best thing.
Full disclosure: it is also important for us to state that we work with most of the vendors reviewed. When you talk to us and we help you sign with a vendor, we may get referral fees. At the time of this writing, the vendors listed below were willing to provide information, or are vendors we’ve chosen to work with for our own clients.
- $$ – Moderately Priced
- Interesting solution that enables you to generate ROAS numbers
- Aesthetically pleasing reports
- Fast-growing company
- Feel-good platform focused on graphs
- Flexibility in attribution models types
- Robust feature set
Cons & considerations:
- Reporting designed more for the executive user or the marketer: Users who need graphs to feel good about a report
- Reporting lacks nitty-gritty data
- Poor support: Based on our experience
- $$ – Moderately Priced
- Simple dashboard
- Focus on showing your ROAS: Does not distract you with graphs made to look good more so than to help you make decisions
- The most native integrations out of the other vendors: And they can pull in cost data from all kinds of places
- Some of the strongest integrations for conversions as well: They are one of the few vendors that have native Salesforce, Segment, and MetaRouter integrations
- Attribution model suitable for subscription-based and e-commerce companies
- Presence of the P-value: This tells you the statistical significance of the model results. In other words, the P-value helps you avoid making a conclusion based on the dart hitting the dartboard, when the dart is actually outside the last ring.
- Support for a broad variety of attribution models: Including data-driven, Shapley, Bayesian, Markov, random forest, and custom
- Data auditability, model auditability, and data exportability: The tool caters to these needs
- Ability to go down to the keyword level for Google Ads ROI attribution
Cons & considerations:
- Focus on look-forward windows: Custom lookback windows are available, but the tool’s focus on look-forward windows is prevalent and may be hard to work with for some marketing teams. We think look-forward windows are important because the logic behind them is sound. But look-forward windows will be something you need to consider carefully before you commit, because they may not be ideal for lead generation
- Lifetime value (LTV): The look-forward window aggregates all revenue from after the period you are looking at. While this gives you LTV on your spend and gives LTV in the ROAS, it also makes it so your reports are constantly changing. If you look back at January in the month March, it will show one number, and then if you look again in April you would see a new, higher number because more revenue for those users would have come in.
Pros and price point:
This space intentionally left blank.
Cons & considerations:
- Marketo acquisition, then Adobe acquisition: Much of the team that knows what is going on has left. They are having a tough time getting integrated into the multiple mergers.
- Horror stories of no support, no customer success, and more from our friends who use them.
- Data typically based on Marketo and Salesforce: This is not bad, but often not a good fit
Nielsen Attribution (formerly VisualIQ)
A note on naming: this tool used to be called VisualIQ. The rebrand came about after the acquisition by Nielsen. Their attribution tool now falls under Nielsen Marketing Effectiveness, a division of Nielsen.
- $$$$ – Expensive
- Marketing mix modeling
- Willingness to talk about the entire attribution space: And where they felt the flaws were
- TV attribution: As you’d expect of a Nielsen company
- Rigid support: One of the more rigid vendors in the mix in general
- Lack of key integrations: The tool doesn’t integrate with customer data platform (CDP) or customer data infrastructure (CDI) tools such as Segment
- Traditional marketing stack: Used by marketers who rely predominantly on the traditional marketing stack, e.g. Adobe or Tableau. We prefer CDPs and CDIs.
- $$$ – Pricey
- Cool UI: They provide some advanced features when it comes to segmenting revenue from different conversions
- B2C space: This tool is built for B2C
- Impressions tracking
- Great team to work with
- Honest communication about custom models: The team clearly conveyed that if you want to dive deep into custom models, you aren’t going to want something pre-packaged and out of the box. They seemed forthcoming about providing tools to make customizations, instead of trying to convince you to use a solution that already exists in their tool.
- Honest communication about the flaws of attribution: The team was very clear about that attribution itself is flawed and that companies need to use multi-touch models to be directional and optimize. The models aren’t meant to tell you the secrets of your customers and their journey.
- Ability to track everything: Even TV
- Cohorting of revenue and LTV tracking
- High level of granularity and accuracy in ROAS: Thanks to customization of how ad data and link data are collected
- Interactive model editor: You can make changes without asking for support help
Cons & considerations:
- Customization of ad and link data collection: This necessary workflow can be hard to maintain
- B2B: The tool is not perfectly equipped to work with B2B companies right now. We are unsure if they will expand to the B2B market, either.
- Salesforce integration: This is lacking and may not be built anytime soon
- Customization fees: If you don’t have your own team with the expertise to customize the models, their help comes with a steep fee
- $$$$ – Expensive
- Great communication
- Experienced team: One of the early players in this market
- Custom models
- Machine learning: Helpful when refining the models
- Custom requests: They have the ability to accomplish other requests, such as cohorting of conversions, custom lookback windows, and impressions tracking
- Bayesian model: We appreciate the use of advanced and less common attribution models
- Locking the look-forward period window: The ability to do this makes it so that the numbers will not change if you look back in time
- Good integrations with advertising networks: Better than most vendors
- Customer data platforms: The tool integrates well with CDP’s
Cons & considerations:
- Cost: The investment only pays off for truly large ad spends, sometimes only after counting in offline media such as direct mail or TV ads
- $ – Affordable
- Bayesian model, Markov chain: We appreciate the use of advanced and less common attribution models
- Lightweight tool: There is little additional tagging or tracking required
- CRM integration and data-driven models: This helps track offline activity
- Integration of costs: You can pull in data through API or email/SFTP for channels such as TV or radio
- Custom attribution models: The team has the ability to create them for you
- Young company
- Product still evolving
- Less polished user experience
- Limited documentation
Price point: $$$ – Pricey
- Easy setup: Minimal additional tagging required
- Native integrations: Google Analytics, Google Ads, Google Analytics, and DoubleClick Search
- First, last, linear, time-decay, and position based attribution models
- Data-driven/machine-learning based attribution
- Custom attribution models
- Excellent offline media attribution
- Unlimited touchpoints tracked within the conversion window
- Intuitive use
- Automatic BigQuery data warehousing
- Tracking outside of Google ecosystem: This need can be difficult to configure
- Limited export or access to raw data outside of BigQuery
- ROAS calculations limited
- Support not great
- Powerful free attribution platform
- Additional tagging not required
- Integration into with Google Analytics (GA)
- Native integration with Google Ads, Google Analytics, and DoubleClick Search
- Data-driven/machine-learning based attribution
- Unlimited touchpoints tracked within the conversion window
- Intuitive to use
- Great solution for small businesses: Good fit if you use GA and are just getting started with attribution
- Free version limited to Google advertising
- Offline attribution: Not present
- View-through tracking for impressions as conversion factors: Not present
- Export or access to raw data: Not present
- ROAS calculation: Not present
- Custom attribution models: Not present
- Limited support: Only in community forums
Final Thoughts on Attribution Tools
Your approach to the tools is more important than the tools themselves. Also, this space is still really young. When talking to the vendors, they all admitted that the way date ranges in the model work compared to when the spend and conversion happened leaves a lot of room for error. As with most analytics tools and MarTech mistakes, the difference between directional and definitive is the key thing here.
It is important to re-emphasize that all of the tools above will be best used for the optimization of paid channels. The only tool we felt which would also help with historical reporting of the ad spend would be C3 Metrics.
For some of our larger clients, an attribution vendor would not provide what they’re looking for, because they want something more custom. The alternative is to build your own model on top of a data warehouse and a customer data platform.
Alternatives to Marketing Attribution
We highly recommend the pragmatic episode about multi-touch attribution on the Analytics Power Hour podcast. Priscilla Cheung, Moe Kiss, Tim Wilson, and Michael Helbling took a critical approach to whether MTA is or isn’t worth it. The thoughts we found to be the most constructive were the ones about marketing tactics that can be used as viable alternatives to attribution. Diving deep into them is beyond the scope of this piece, but we do think there’s value in briefly listing a handful of the alternatives mentioned by the four experts:
- Lift tests (potentially combined with attribution models)
- Customer journey mapping
- Journey analytics
- ROAS tests that look at varying ad spends
The common theme is that instead of relying on attribution modeling that may not be a good fit for you, you might get better results by:
- Setting a specific goal
- Working toward the goal by applying insights and tactics that are under your full control
Conclusion: Do Multi-Touch Attribution with a Wide Open Mind and Full Buy-in, or Don’t Do It At All
We’ve seen cases where a marketing leader refused to believe the attribution models even though the data was correct. This is a huge issue in attribution projects. If you have the expectation that attribution is going to prove you right, and it proves you wrong, you’ll find yourself blaming the model and not the campaigns. This can easily lead to losing the buy-in of you and your team.
Another common scenario for a very sophisticated attribution implementation to go sideways is being blocked by developers over and over again. Not having a full commitment to the complex project can quickly lead to wasted money, time, and opportunity.
Ryan Koonce, the executive chairman of Attribution, observed yet another tendency:
What we have seen is that media buyers (usually agencies) who compare to GA or Facebook Insights are frustrated with MTA because they can only count the conversion once, and their net numbers, therefore, don’t look as good. Internal media buyers, on the other hand, who are not incentivized to increase spend, but rather optimize ROAS, tend to get a lot of value from MTA.
What you will get as an outcome is never what you expected, and that is the point.
Even when the data is imperfect and you don’t have a complete picture, you can still do a ton with that data. You can, for instance, gain a profound understanding of the customer journey, and use the insight to set up effective personalization.
With that said, marketing attribution is a lot like product analytics and product management. When building a product, you often create a feature to be used in a certain way, and then your users use the feature differently than you ever anticipated. Just like product data, marketing touchpoint data helps you understand the customer journey. It can tell you where your customers truly came from, what their pain points are, what their experience is throughout the steps, what the role of the product or offer is in the journey, and what the conversion factors are beyond your current campaigns
This is why you want to use analytics to tell you what is happening, not confirm what you thought would happen.
Improve your ROAS by understanding the value of your individual touchpoints along the customer journey. Schedule your free consultation and let’s get started.