Data-driven Attribution In Your Area: Strategies for your Industry | PART 4 of 4.

White Paper | Authors: Catherine Candano Google, June Cheung Oracle Advertising and CX, Rupesh Kumar Carat Media, Tara Crosby Twitch, Vidyarth Eluppai Srivatsan The Coca-Cola Company

Playbook for E-commerce brands for Effective attribution with multi-touch attribution (MTA)

We’ve come to the final article in our four part series on demystifying attribution, MTA (Multi-Touch Attribution) and MMM (Marketing Mix Modelling). In this final article, we will deep dive into an eCommerce Brand’s MTA strategy.

Today, Jessica, Marketing Director of eCommerce brand, will share her playbook on how she utilises MTA to identify the optimised digital media mix to drive increased sales.

Jessica: –  Multi-touch attribution collects individual or user-level data for addressable media and conversion/events to determine the impact each media action/event has on a customers’ path to conversion. It determines the value of every touchpoint on the way to a conversion rather than giving all the credit to one touch point, such as First touch, Last touch. MTA gives credit to every advertising channel interacted with on the customer journey – similar to our BLACKPINK band where all the credit, or attribution, is not given just to lead vocalist, Rose. Applause from the crowd is for the entire BLACKPINK band including back-up musicians.

Josh: Jessica, this is very helpful and I can totally relate to the BLACKPINk band. 

I have a question: does MTA help in measuring the impact of like print, radio or traditional TV?

MTA mainly utilises digital media touchpoints because it requires tracking and connecting all media at the user level. However, it does not account for non-addressable media, like print, radio, and traditional (linear) TV, which cannot be tracked to individual users.

Josh: That makes sense Jessica, but does this not impact your decision for media investment as MTA does not capture non-addressable media like print or radio?

Great question, but as you know we are an eCommerce organisation and 100% of our sales come through digital platforms like our website or app. Because of this our investments are also skewed towards digital media channels (like Social, Display, OTT, Google Search) and we don’t invest on non-addressable or ATL (above the line) media channels such as radio, print or linear TV. In future, if we plan to invest in non-addressable media channels then we do need to utilise marketing mix modelling in conjunction with MTA and your MMM playbook can come handy for us.

Josh: In that case surely you should only use MMM because it can also provide insights on your marketing investment at media channel level, like how much on online media vs display media? I don’t see much benefit for MTA.

MTA provides insights not only at media channel level but it can provide insights and recommendations at different levels including tactical, creative or publisher. Plus it also it captures paid media and owned media attribution.

With MTA we are able to get insights or true conversions at a granular level and optimize the media campaign accordingly or take informed decisions on investment at tactic level.

Josh: Got it. Do you consider any caveats as you brought up MTA at e-Commerce?

Yes. there are few considerations before we use MTA:

1. Scale: For MTA the thresholds are lower since it can be set up on a campaign level instead of a multi-year modelling; however MTA should only be used if investments on digital channels are also varied by channels and tactics. For example, if a brand is using only 2-3 online media channels and 2-3 tactics then it doesn’t make sense to utilise MTA as the difference in results would not be that significant.
2. No consideration for external factors: MTA works well for online touchpoints but does not account for other external factors that affect or influence sales and marketing effectiveness, such as pricing, promotions, economy, demand etc which are captured in MMM.
3. MTA does not account for offline media to online media effects: Multi-touch attribution only considers online media touchpoints and does not take any consideration of any offline media impact on the user journey. For example, if a user sees an Out of Home (OOH) ad and then engages with a social media ad and converts via organic search, then MTA will attribute the conversion to only online media touchpoints which are the social media ad and an organic touchpoint, so nothing will be attributed to the OOH ad.

Josh: That is insightful Jessica. Did you have any other considerations as you deployed MTA at e-Commerce?

MTA deployment is a time-consuming process and requires a six-step process –

1. Identify the key actions
2. Choose the right partner for MTA deployment
3. Data collection and consolidation
4. Taxonomy
5. Testing
6. Optimisation

Identify the key actions: First thing is to identify key actions to be considered as conversions on the landing page/app which are important for the business to drive success. For an eCommerce brand –  view products, add to cart and purchase are the three key action points which are important, especially purchase.

Choosing the partner: It is critical to research and identify the right partner which can provide MTA services for your brand. They should provide customised solutions, a good user experience, transparency along with actionable insights.

Data Collection and consolidation: Once the key actions and MTA partner is identified it is important to tag all the landing pages or app pages. Your MTA partner will provide pixel or tag to be implemented on the landing page and more importantly on key actions (view product, add to cart and purchase) with event tracking or goal tracking. Now the data of each visit to your site is consolidated at a data attribution tool/warehouse. (If key actions can occur from point of sale system outside of the site/app, it is possible also to securely hash mobile/emails from customer relationship management (CRM system) and append this unique identifier’s conversion data via API to provide combined information on conversions onsite with happenings offsite.)

Taxonomy: Once data has started consolidating in a single platform, it is important to define your campaign, creative, placement name with the right taxonomy or naming convention. Correct naming conventions helps in classification and enables us to analyze the performance basis of these naming conventions. For example, Social = SOC, Search=SEM, India=IN. Once these naming conventions are defined, start using them in the campaign set up.

Testing: Once the data collection has started and taxonomy is in place, your MTA partner can start testing different attribution models. MTA models can be rule-based, algorithm-based  or customised.

Rule based models: Time decay, position-based or linear model can be manually adjusted by the marketer as a fixed rule to prioritise which channel generally is assigned to receive credit in advance.

Algorithm based models: Data-driven attribution (DDA) – a predictive algorithm to analyse the data to determine which channels, campaigns, creatives and keywords have the biggest impact on conversions. A statistical model uses the first-party data that marketer has to show the patterns of interdependent channels to show the  path to  conversions. It uses the marketer’s own historical data available to see how each different touchpoint delivers conversions (or may not directly deliver conversion but influences it interdependently), and uses machine learning to learn with statistical certainty the relationship between channels towards paths to conversion.

Because this automated approach inclusively considers all data available from channel touch points as data input to the model and learns patterns based on actual consumer journey, it is responsive and dynamic to real time shifts in consumer touchpoints (as it will adjust as consumer behaviours change in their interactions with campaigns) without the need for manual adjustments by the marketer. Marketers can adjust bids in campaigns automatically in response to real-time signals on the most impactful channels from the statistical model, instead of adjusting bids manually.

Customised Model: A custom model created for your brand gives you precise control over how you distribute credit for conversions. A customised model is used to create your own attribution model which is not based on any rule or statistical/algorithm. For example, a custom model allows a marketer to assign more value to social touchpoints or assign a lower value to display touchpoints if they feel social media might have more impact. With rule and algorithm models marketers cannot make these adjustments.

MTA requires at least two to six months for deployment (depending on the data sets), including data collection, consolidation and building effective models. Once you finalise the model, start implementing it on one of the campaigns rather than implementing it on all the campaigns as this gives a set of results and allows testing and learning about the system that has been set up.

Optimise: Determine which platform, tactic, creative, keyword, or audience are impacting performance in a positive way which enables on-the-fly optimisations and also helps in better forecasting and planning. For example, the algorithm model might suggest investing more in Facebook when prospecting audiences, and a customised model can recommend investing more in Programmatic PMP buys. Test these recommendations and optimise your campaign on the fly to get the best results from your campaign. Consider also how bidding strategies on these channels could work better together with multi-touch attribution because incorporating  automated bidding strategies into MTA can enable more readily instant, real-time optimisation.

Our eCommerce brand has optimised campaigns through algorithm and customised models and realised that algorithm models delivered slightly better results; hence we are using the Data-Driven Attribution model for all of our campaigns.

Josh: Thank you, Jessica, for sharing your playbook. I have the last question: How would you see the future of attribution, or any other trends you foresee?

One of the near-term challenges for Multi-touch attribution would be a world without 3rd party cookies since most browsers already have or are moving away from 3rd party cookies. Multi-touch attribution relies on a single customer view and how a customer is engaging with a brand on different ad platforms, with varying ad placements and different creative across more complex shopping journeys; however, with the removal of 3rd party cookies, brands need to move away from one-to-one tracking.

We have already started working on first-party data strategies to better understand our consumers’ journey before the removal of 3rd party cookies. Our focus is to get insights about the consumer conversion patterns that demonstrate channel’s/tactic’s value, with first-party data that we have. We’ve reviewed the basics of our data collection to ensure it is precise and comprehensive — from tagging campaigns properly with first party tags, as well as connecting conversions of  consumer touchpoints offline with digital  to ensure there is robust first-party data foundation in our company.  First-party data offers opportunities to map users via owned media and can also work as identifier across platforms (like mobile number and email IDs) to understand the consumer journey on their  path to purchase. The integrated view of a consumer’s journey allows the mapping of the online conversions and the key actions tied to user identifier. This allows us to better attribute what is the relationship between the channels and eventual purchases directly by customer at point of sale.

With our robust data, automation insights on these conversion journeys to a sale, we know our attribution playbooks are reliable guides. At the same time a good marketer will use their own contextual knowledge of  brand, market, audience and product knowledge to make any investment decision. For example, if MTA results suggest investing 50% on retargeting, which is not possible because of consumption capacity on the retargeting audience. Similarly, MMM can suggest heavily on digital; however, your past learnings show that without investing in ATL business, we would not be able to make an impact to drive a combination of both ATL and digital strategies, then make a business call based on your industry benchmark insights. And then review the data insights again!

This also reminds us that attribution has a lot in common with  BLACKPINK as success or conversion is an interdependent  team effort and not simply  reliant on one person or one channel. For example, a good band manager will make sure that the credit is given to every member of the team Jisso, Jennie, Lisa, and Rose (singers and rappers) and to the choreographer, make-up artist, and other people who are making this band a success around the world, while also taking into consideration the internal and external factors which may impact the BLACKPINK performance. Similarly, a good marketer would not just rely on last click or first click attribution. They would give credit to every interaction or path to conversion and consider the internal and external factors as those have a more significant impact on sales or conversions so the business can evolve with the time.

The Conclusion

1. The User Journey is not linear anymore, and consumers jump between channels and devices; MMM and MTA help marketers accurately understand the impact of each channel and tactic, and relationships of how each channel influences the path to purchase.
2. Decide on the KPIs for your campaign and for your business and use these as the basis that decide which MTA or MMM models are suitable for your brand.
3. Identify which data sources are available internally before deciding to go for MTA/MMM, or both, as implementing these requires time and investment to generate results.
4. Test attribution models using data available, and keep your focus on experimentation for effectiveness (Impact on investing on one channel vs. the second or a combination)
5. Invest in building your first-party data as this will support future-proofing attribution in the post-3rd party cookie era.

With this we end our 4-part series on attribution, You can discover more about attribution from industry leaders with their real-life challenges and learnings in the on-demand webinar about Data-Informed Attribution Strategies available here.