What We’ve Learned From Building Attribution Models for the Top Multichannel Retailers

Mariia is a Business Development Executive at OWOX BI  shares findings from building attribution models for top multichannel retailers and walk you through the main considerations in tackling attribution.

Covering the basics

What is attribution modeling? It’s a set of rules that help marketers understand how channels convert customers at different stages of the sales funnel and what the best way is to allocate credit among those stages.

THE NEW RULES OF (MULTI-CHANNEL) CUSTOMER ENGAGEMENT: HANDBOOK FOR GROWTH MARKETERS

Winning CX will come from Brands who can balance relevance, consistency and convenience to drive engagement. The kind of engagement that drives optimal customer lifetime value and real business impact.

Download

We all know that customers go through different stages in the buying cycle. The first is interest or awareness, when customers don’t yet have a strong intention to buy and might not be familiar with your company. At this stage, customers are just becoming aware of your brand and are starting to consider you as an option.

The next stage is consideration. Once targeted, customers become aware of your products and services and gain interest in them.

Conversion is the next stage in the customer journey. Most marketers think that at this point the battle is won; they’ve got the client.

But in reality, there’s another important stage in the funnel – retention. You should be using it to boost your business performance, since it’s always cheaper to retain a customer than to acquire a new one.

For each stage of the funnel, there are different campaigns. Campaigns focused on the top of the funnel usually target a broad audience and are not really attributable. On the other hand, campaigns focused on retention (the bottom of the funnel) are precisely targeted and highly attributable since you can identify existing customers by email, phone number, your own customer ID, and other means.

Campaigns targeted at one stage in the funnel can still influence customers at other stages, however. For instance, you might run a campaign that targets the top of the funnel, but at the end of the day it influences your offline sales. This indirect influence can be tricky to measure.

John Wanamaker, a 19th-century American merchant, once said: “Half the money I spend on advertising is wasted; the trouble is I don’t know which half.” And this problem is still relevant to modern marketing specialists.

Also Read: Fact Checking Three Common Multi-Touch Attribution Claims

Getting started

According to Gartner research, spending on digital advertising surpasses spending on offline advertising: CMOs invest two-thirds of advertising budgets in digital channels. In order to make the best use of this investment, you need to properly measure the results of your ad campaigns and adjust your advertising strategy accordingly.

Let’s take a closer look at what attribution models are available and how they can be applied to your business.

The following models are available in Google Analytics for free, but all of them have one big disadvantage: they don’t consider customer progress through the funnel and assign credit to traffic sources based only on their position in the sequence.

• First interaction attribution – When all credit is attributed to the channel that drove a customer to your site for the very first time, it overemphasizes the top of the marketing funnel. Still, this is an easy way to know what attracts new customers (new names) to your brand.
• Last interaction attribution – This model gives all credit to the final touchpoint leading to the conversion.
• Last non-direct attribution – Google Analytics reports, including audience and acquisition reports, use this model by default. If the source of the last session was direct, the algorithm will work backwards and attribute all value to the last non-direct source.
• Linear – Credit is evenly divided across all touchpoints that led to a conversion.
• Time decay – This model gives more credit to the final or most recent touchpoint and assigns less credit to touchpoints further back in time.
• Position-based – This model gives 40% of the credit to the first and 40% to the last touchpoint; the remaining 20% is evenly split among all sessions in between.
   

Right off the bat, we can say that first click and last click attribution are a no-go; if you go to the path length report in Google Analytics, you’ll most likely see that 80% of your orders are made after two or more sessions. That means that if you’re using single click attribution, you’re not evaluating about 80% of sessions that happen on the way to a purchase.

Few respondents to the 2017 Ad Roll State of Performance Marketing Report considered standard position-based attribution models "very effective." On the other hand, nearly all respondents considered custom and algorithmic models "very effective" or at least "somewhat effective."  Even so, 44% of marketers still use last click attribution, and only 18% use algorithmic attribution (according to statistics from the same Ad Roll report).

• Data-driven – This model is available in Google Analytics 360. On the one hand, it leverages your account data and Google’s machine learning capabilities to determine which user touchpoints are most influential and learns and adjusts over time based on performance. On the other hand, it’s a black box model that you just have to blindly trust. The main downside is that although data-driven attribution works for all conversions, including those imported from Google Analytics, it doesn’t apply to other conversions, such as calls and app downloads.
• Markov chains – Using this model allows you to switch from heuristic models to probabilistic ones. You can represent every customer journey (sequence of channels/touchpoints) as a chain in a directed Markov graph where each vertex is a possible state (channel/touchpoint) and the edges represent the probability of a transition between the states (including conversion). By computing the model and estimating the transition probabilities, you can attribute credit to every channel/touchpoint. The main obstacle to using this model is that you need to know how to work with R and be a bit geekier than most marketers today.
• The Shapley value is a solution concept in cooperative game theory. It assumes that a number of people are cooperating and obtain some gain from that cooperation. However, some people contribute more than others. The Shapley value provides a way to measure how important each player is and what payoff they can reasonably expect.
• Models available on ad platforms. Google Ads, Facebook, and other platforms have their own attribution models that show great results if you’re using just that one channel. But if you want to consider the mutual impact of multiple channels, you might want to try other approaches.
• Funnel-based attribution. This model considers how customers pass through the sales funnel and attributes value to a session based on customer progress during that particular session. It credits those sources and campaigns that bring you the most valuable customers.
• Custom attribution models. You can always build your own attribution model based on complete data on your customer journey and business model. For example, Hoff – one of the e-commerce leaders in Russia – decided to build their own attribution model that corresponds to their business objectives based on the raw data they obtained in Google BigQuery. The logic was the following:
   

1. Determine the channel that introduced the customer to the brand and assign it 20% of the attributed value.
2. Determine the channel that converted the customer (Last Click) and assign it 30% of the attributed value.
3. All other sessions should get a share of the remaining 50%.The more pages are viewed within a session and the closer that session is to the purchase, the more value the session receives. This helped Hoff increase their PPC advertising ROI by 17%.
    

            In my opinion, you can consider building a custom attribution model if

1. Your marketing spending is at least $100k/month
2. You have multiple (5+) campaigns running in parallel with significant spending in each
3. You have micro-conversion events that are tied to significant economic value

Building your own attribution model can be overwhelming, but using a model that doesn’t consider your funnel – or which is solely based on online data – is like driving a car blind.

Why is attribution difficult?

Everybody understands that attribution is important. So why do so many companies underuse it?

For one thing, attribution is difficult. In order to do it right, you need to educate your team and collect, analyze, optimize, and act upon data collected online, offline, from ad platforms, etc.

Based on our experience building attribution models for top multichannel retailers, we’ve observed several reasons why companies underuse attribution:

1. Lack of knowledge and data literacy. To me, this seems like the main obstacle blocking companies from implementing marketing attribution. If people don’t speak the same language and don’t understand where numbers come from and what they mean, it will be super difficult to get the ball rolling.
2. Data silos. What’s the point of spending time on advanced attribution models if data is scattered, meaning there’s no single source of facts. I’m sure you’re familiar with situations when the CFO looks at the marketing numbers and doesn’t agree with them since they don’t match the revenue he sees in his reports. Also, bringing data together for multichannel retailers who have both online and offline stores can be tricky. But it’s absolutely necessary if you want to consider all touchpoints along your customer journey. Usually, companies are afraid that it will take them too long to bring together all the necessary data. And with good reason: according to recent research from McKinsey, only 8% of data lakes have moved from the proof of concept stage to production. From our experience, in order to get things moving, you should define data sources, prioritize them in terms of their importance for reporting, and pull in data starting with the highest priority sources.
3. No trust in data. According to KPMG, only 35% of 2,190 global senior executives have a high level of trust in their organization’s use of data analytics.
4. No ownership of attribution or analytics. Usually, there’s no dedicated marketing analyst or attribution manager who will take responsibility for implementing attribution and applying it across different channels. Everyone is busy with their own thing – PPC, SMM, SEO, etc. – and mainly care for their particular channel.
5. Lack of understanding of the potential impact of attribution. If somebody told you that with proper attribution you could increase ROI by 40% without changing the ad budget, would it make you reconsider your current approach?
6. Requirements can change over time, so your attribution model needs to be flexible. This requirement comes from an increasing desire to customize models based on the characteristics that differentiate individual organizations and their media mixes. If your model isn’t flexible, the cost of changes might be too high.
7. Change management and governance are other frequently missing key components of successful attribution that should be a focus from the beginning of the project. Analytics, media, and management processes should be documented and clearly communicated to the team. In addition, decision makers and stakeholders should be clearly identified.
8. Missing internal processes. Any change will fail if you don’t have processes that support it. And no attribution model will benefit your business unless you act on the results. Each business has to focus on developing and implementing revenue-generating processes first and foremost, and attribution is one of them.
    

Considering all these reasons why companies underuse attribution – and the complexity of measuring campaigns – the results of the State of Marketing Attribution 2017 report by Ad Roll should come as no surprise. According to this report, 72.4% of marketers

1. don’t know why they chose their model
2. selected the easiest attribution option available to them.
    

This means that most companies fly blind and don’t properly distribute credit to all touchpoints along the customer journey.

Also Read: What’s Holding Marketers Back from a Holistic Attribution Strategy?

How to get buy-in from stakeholders

It’s not easy to get buy-in from executives, but doing these things should help you find the right selling points:

1. Map the implementation of attribution to your company goals. Marketers who link marketing metrics to business results are two times more likely to exceed their business goals than those who don’t, and are also two times more likely to play a role in corporate strategy discussions (according to Google and Forrester). Let’s say your goal is to expand the customer base; in this case, your main goal for attribution will be to find channels that bring you customers who aren’t familiar with your brand. On the other hand, if your goal is to increase revenue from existing customers, you might want to focus on channels that retain them and bring conversions. You might want to consider the framework from Mower.com, which allows you to link your marketing metrics to business results.

2. Estimate outcomes. One way to do this is to look at your current model – let’s say it’s Last Non-Direct Click – and the one you want to switch to – say it’s the Funnel-Based model. Your current model ignores the share of visits in the sequence that led to purchasing because only the latest source receives the full value. To estimate the extent of the problem, it’s necessary to measure the proportion of sessions that remain without evaluation using the Last Click model. You can estimate the share of these sessions by making two segments in Google Analytics. The first will include all the sessions in which an order was made and the second will include all sessions of users who made an order. In the first segment, all sessions with conversions are counted. And in the second are all sessions of users who have placed an order, including sessions without a conversion. You need to divide the first by the second. On average, this value equals ~17–20%, meaning that only about 20% of sessions get a value from this attribution approach. But what about the rest? How will your campaign assessment change if we consider the impact of those undervalued campaigns?
3. Finally, talk to people. Understand what they get promoted and fired for. If you understand their motives (both business and personal), it’s way easier to help them.

The payoff
 

If everything is done properly, you can reap the following benefits:

1. Decreased costs thanks to giving up inefficient channels.
2. Increased revenue thanks to reinvesting spare funds in channels that bring valuable customers.
3. A single source of facts for the whole team. If everyone is on the same page with numbers, you don’t have to sweat to justify your efforts or prove your point of view.
4. You can finally evaluate the influence your online campaigns have on your offline sales, or the so-called ROPO effect. For example, Chanelle – a top French retailer in the women’s lingerie market – has a huge offline presence, and its team was trying to evaluate the efficiency of their e-commerce division. Initially, they thought that only 5% of their annual revenue was generated thanks to online sales. But once the data was brought together and the ROPO effect was calculated, it turned out that online was responsible for approximately 40% of overall revenue. This helped them to justify an e-commerce budget increase for the next year.
5. Automated integrations. According to statistics, around 70% of marketing specialists have difficulties applying attribution results. If you trust your attribution results and don’t have to verify them manually, you can automate bidding. This will help you save resources, speed up adjustments to the bidding strategy, and eliminate human error.

Takeaways
 

Here are the lessons learned from other companies who have successfully measured their ad campaigns:

1. Be clear on your objectives from the very beginning and share them throughout the company, along with key performance indicators (KPIs) when appropriate. Think of the key stakeholders and other teams that need to contribute, and ensure the strategy is communicated to and supported by all those developing insightful attribution models.
2. Get internal buy-in for attribution. The impact can be far-reaching, affecting workflows, commissions, and bonuses. Failure to get buy-in can lead to failure to act on insights. Backing by management for attribution must be sought so that all departments are sold on the benefits while also being clear on the business goals and methodology. This will ensure that certain teams don’t suddenly question the validity of models when they don’t like the recommended outcomes.
3. Educate the team. Create an educational plan that involves C-levels, stakeholders, and operational personnel. Make sure you’re on the same page and that objectives are clear to everyone.
4. Data governance and processes. Ensure that these processes are in place and support the upcoming changes.
5. Improve communications, as they’re the key to successful change. We recently interviewed Simo Ahava on the importance of communication. One of my favorite quotes from this interview is: “Communication problems create rusty pipelines and rot the data flow within an organization. It’s absolutely imperative to fix those.”
6. Defining the online customer journey is the most significant barrier to using attribution effectively. Ensure that you take a holistic view of the touchpoints along the way. Although customer paths are becoming less linear and funnel-like, it’s still possible to build a picture of triggers and typical pathways. Make sure you consider your business model, the vertical you’re operating in, the length of the decision-making process, etc.
7. Focus on physical as well as digital touchpoints. Every company will benefit from a more connected approach. Unifying data is a clear starting point for developing insightful attribution models. ROI assessment should include both online and offline data.
8. Make sure that data sets are as clean and accurate as possible. Attribution models are only as strong as the weakest link in the chain. For multi-channel attribution modeling to work, all your marketing campaigns must be 100% tagged with campaign tracking parameters. Also, make sure your data collection is set up properly. Your analytics tracking doesn’t have to be a 100% perfect piece of art, but it has to be reliable at least in terms of main events. Otherwise, you might be misled by analytics.
9. Invest in technology that gives you the required flexibility. Choose a platform that allows for changes in patterns of behavior and adjustments to your models so you can test new hypotheses and continually refine your approach. According to Gartner research, marketing technology is the single largest area of investment when it comes to marketing resources and programs and takes about one-third of the CMO’s budget.
10. Experimenting with different attribution models and methods allows you to determine what works best for your data and which processes are most effective. Companies can benefit from an agile approach rooted in a commitment to test and learn. Consolidate your data first to understand which channels deliver the results you expect given the assigned budgets. After that, move into modelling the data, making small changes each time to move closer to your goal. Testing needs to occur before confidence can be put into any attribution model, which is where many companies stumble with the implementation process. Testing against a forecasting tool can instil confidence that the correct balance is being achieved.
11. Act on the results. Numbers themselves aren’t worth anything. Spend time turning insights into action and revenue for your company.