How do i approach attribution?

 

Using the most effective attribution model for your organization is important, as results will differ depending on which model you use. Determining an approach to attribution entails experimenting with both channel-based and user-based interaction models. Channel-based interactions tend to be more predictive in that they use historical data to help organizations project an effective media mix for the future, while user-based models use current data to help inform decisions.

Marketing Mix Modeling (MMM)


Rather than look at each individual touchpoint and interaction, media mix modeling looks at the data from a higher marketing channel level. It attempts to determine at this high level how changes in expenditure across certain marketing channels will affect other channels and overall results. This model is mostly used by brands that focus on offline channels, especially CPG; businesses highly affected by external factors and branding; and organizations that don't have user-level data.

Benefits

  • Insights provide support for strategic scenario planning and annual budget setting
  • Insights can ultimately help optimize marketing mix at a higher level

Limitations

  • The risk of predictors being highly correlated with each other, which makes it difficult to measure media at the tactical level
  • The crude measurement of dynamic and fast-changing digital and direct marketing channels
  • The inability to measure the performance of new products and campaigns, which lack historical data   

 

Complexity: High Cost: $$$ (100K+ a year) Commonly used by: B2C

 

Algorithmic

Algorithmic attribution is the process of assigning a portion of credit for conversion to each touchpoint based on effectiveness. This method uses advanced statistical modeling and inferences to continually optimize and customize your model based on results.

Benefits

  • Truly data-driven
  • Eliminates the challenge of multiple data sources and provide a single, reliable view of the entire buyer’s journey
  • Incorporates data from online and offline marketing channels and works across all devices
  • Can adjust for external factors

Limitations

  • May be difficult to normalize data previously drawn from different sources
  • Complexity of model may make it difficult to understand
  • Critics may question the methodology
Complexity: High Cost: $$$ (100K+ a year) Commonly used by: B2C, Academia

 

USER-BASED INTERACTION

SINGLE TOUCH

Single-touch attribution models are best for smaller companies with a short sales cycle and simpler marketing and sales systems.

First Click


Assigns all the credit to the first marketing event encountered before a conversion.

Benefits

  • Easy model to put in place
  • Data should be available and easy to access
  • Necessary for conversion to happen

Limitations

  • Only rewards one channel and ignores the contribution of others
  • Limiting when seeking to optimize or demonstrate the value of their efforts
Click Credit
First Click  
Complexity: Low Cost: $ (free-10K a year) Commonly used by: B2B; B2C; Non-Profit

 

Last Click


Assigns all credit to the last marketing event encountered before a conversion.

Benefits

  • Simplistic model that is easy to track and implement
  • Easy to access information about, as it is the most common method
  • Most common method

Limitations

  • Limiting when seeking to optimize or demonstrate the value of non-traditional partners
  • Does not consider conversion path patterns or external factors
Click Credit
Last Click  
Complexity: Low Cost: $ (free-10K a year) Commonly used by: B2B; B2C; Non-Profit

 

USER-BASED INTERACTION

Last Non-Direct


Gives all credit to the last non-direct click and excludes all other traffic.

Benefits

  • Excludes direct clicks which aren’t actionable

Limitations

  • Does not track all clicks
  • Shifts more credit to other clicks
Click Credit
Last "campaign" prior to the conversion  
Complexity: Low Cost: $ (free-10K a year) Commonly used by: B2B; B2C; Non-Profit

 

MULTI TOUCH

Multi-touch attribution models are more effective for companies that leverage three or more marketing channels, generally have a longer sales cycle, or have a bigger marketing budget.

Time Decay


Rewards the marketing event closest to the conversion with the majority of the credit and the events prior with a diminishing amount of credit.

Benefits

  • Appeals to businesses with short sales cycles
  • Rewards key players for their impact on a quick decision and others for the diminishing value they provide

Limitations

  • Applied subjectively and does not consider the relative effectiveness of each channel in the customer journey
  • Key players’ contributions might not have occurred without the others
  • Ignores event frequency and external factors
Click Credit
All Clicks  
Complexity: Average Cost: $$ (11K-99K a year); Commonly used by: B2B; B2C; Non-Profit, Academia

 

Even (Linear)


Rewards every marketing event in the buyer’s or prospect’s journey equally for the conversion.

Benefits

  • Good choice for environments where every touchpoint is equally valued during the consideration process.
  • Every participant is rewarded so it is easy to track and manage
  • Teamwork approach

Limitations

  • Does not adjust for high volume media (like retargeting), diminishing returns or the relative effectiveness of individual channel partners
  • Does not consider seasonal or macroeconomic contributions
Click Credit
All Clicks  
Complexity: Low Cost: $ (free-10K a year) Commonly used by: B2B; B2C; Non-Profit, Academia

 

USER-BASED INTERACTION

Position-based


Applies 40% credit each to last and first touch. It then distributes the remaining 20% across each touch in between.

Benefits

  • Suitable for those who highly value introducers and closers, as it rewards those two channels equally, while still giving incentive to contributions throughout the buyer’s journey
  • Model can be adjusted

Limitations

  • Ignores patterns in customer behavior and instead assigns an arbitrary value to each marketing event
  • It puts some marketers at risk of over or underpaying partners because it does not account for changes in their level of contribution
Click Credit
All Clicks  
Complexity: Average Cost: $$(11K-99K a year) Commonly used by: B2B; B2C; Non-Profit

 

Controlled Experiments

Conducting controlled experiments to test and validate your attribution approach can help move you closer to a more accurate reflection of the effectiveness of your marketing channels. These experiments – such as hold-out testing, A/B testing, and multi-variate testing – generally compare the results obtained from an experimental campaign except for one aspect whose effect is being tested. The goal is to create two or more similar conditions, change the desired variable, and then measure impact to tease out cause and effect.

Benefits to conducting controlled experiments include easing difficult decision-making, answering fundamental business questions, and revealing the impact of cross-channel and multichannel campaigns. Factors such as seasonality, shifting consumer behavior, competitive landscape changes, disruptive product introductions, and new technologies pose challenges to testing, ultimately affecting test results. Additionally, superior level commitment is needed to obtain resources (finances, technology and skilled staff) for gathering extensive data results.

There are several types of controlled experiments, including hold-out testing, A/B testing, and multivariate testing. Each is described in further detail below:

Hold-out testing

Hold-out testing involves “holding out” a campaign for a control group- a group of customers or prospects who are excluded from a marketing campaign. Measure the purchase behavior of the hold out group after a certain length of time, typically 90 days. You ultimately compare their value as customers or prospects to the value of that group that received the marketing. The purpose of this test is to determine how much lift (incremental increase in revenue that is generated or not generated by sending a marketing campaign) is obtained and whether customers or prospects would make a purchase without the campaign.

Benefits

  • Great for optimizing processes
  • Answers fundamental questions not answered by A/B testing
  • Good starting point to conducting tests

Limitations

  • Difficult to measure and requires careful customer segmentation and bookkeeping
  • Lack of sufficient data collection tooling as measurement requires building substantial infrastructure to conduct accurate tracking

A/B Testing

A/B Testing is a technique for testing two or more versions of a marketing campaign, such as a webpage, catalogs, and emails. Each version of the marketing campaign is unique and can be visually differentiated from the control without too much effort. The goal is to try a few versions of the marketing campaign and identify which version delivers the desired outcome. To accomplish the test, each version is randomly shown to a predetermined percentage of people, usually half. Once you have the results, hold a brainstorming exercise for ideas to improve the marketing piece, merge the most promising aspects, and get your designers and developers to create one or two versions with the new ideas and measure.

Benefits

  • Least expensive testing approach
  • Existing company resources can be used
  • Minimal effort needed

Limitations

  • All or nothing approach as results will show which version of the test succeeded, but it is difficult to determine what factors contributed the most or not at all
  • Testing is incremental, so overreliance on this method can mean you are making changes too slowly

Multivariate testing

Multivariate testing is a technique for testing changes to many different elements, such as formats, images headlines, and offers all at the same time. The goal is to determine which combination of elements performs the best out of all the possible combinations. Variations of the different elements are combined to create multiple versions of the content, which are tested concurrently to find the winning combination. Once you have the results you can then determine which variables were able to improve your results.

Benefits

  • Testing tools create rich data, which allow you to easily understand the outcomes
  • Tests can be designed as simple or as complex as needed

Limitations

  • Requires significant effort and commitment from across the organization, including marketers, analysts, IT team, and senior executives
  • Many resources (time, financial investment, skilled staff and technology) are needed to complete the testing process
  • Results may conclude that one or more variables did not affect the conversion goal

 

Sources:

Web Analytics 2.0: The Art of Online Accountability and Science of Customer Centricity by Avinash Kaushik

http://engineering.simondata.com/holdout-testing-and-conspiracy-theories https://www.optimizely.com/resources/multivariate-testing/

http://www.mccarthyandking.com/direct-mail-campaigns-2/direct-mail-testing

https://www.kaushik.net/avinash/controlled-experiments-measuring-incrementality/