WHITEPAPER

How to Avoid “Last Mile Attribution Bias” in B2B Marketing Mix Modelling

The client is an ERP applications software marketer who sells the application to small and medium size businesses. The company maintains a database of over 500,000 businesses in the country with contact details of key personnel. The client uses their customer database to target customers through direct mail, e-mail and has a sales force to visit customers. The client is also has a substantial digital presence with a website, online advertising, search engines optimization and social media presence. Every quarter there are live events held in various parts of the country to select prospects. Prospective customers can download the trial version of the software for a limited period. Recently, on the advice of a leading consultant, the client released TV and radio ads to improve brand imagery and generate leads. The campaigns ran mostly concurrently for a period of 4 months.

RainMan • min read • August 19, 2026

Background

The client is an ERP applications software marketer who sells the application to small and medium size businesses. The company maintains a database of over 500,000 businesses in the country with contact details of key personnel. The client uses their customer database to target customers through direct mail, e-mail and has a sales force to visit customers. The client is also has a substantial digital presence with a website, online advertising, search engines optimization and social media presence. Every quarter there are live events held in various parts of the country to select prospects. Prospective customers can download the trial version of the software for a limited period. Recently, on the advice of a leading consultant, the client released TV and radio ads to improve brand imagery and generate leads. The campaigns ran mostly concurrently for a period of 4 months.

Task at hand

The client wanted to know the effect of the various media on total sales, with particular focus on the effect of TV and Radio and how it has contributed to sales and brand imagery.

Data at hand

  1. Vintage : Jan 2009 to July 2012
  2. GranularCty : Weekly
  3. Marketing Stimuli
  4. Mass - tv and Radio
  5. One to One - Email, Direct Mail
  6. Face to Face - Sales Visits, Events
  7. Online - Search impressions, Non search impressions
  8. Incentives and Promotions
  9. Competitions Activity ( 2 large competitors )
  10. Economic sentiment ( 2 Indicators )
  11. Sales revenue in Dollars
  12. Intermediate steps in the purchase process
  13. Number of Emails opened
  14. Web visitors
  15. Live event attendees
  16. Trail Downloads of software

Solution adopted

The following analysis was conducted on the data provided to understand the impact of all elements, especially TV and Radio.

  1. Exploratory Analysis
  2. `Counts, sums, averages, distributions
  3. Trends, seasonal Peaks and troughs
  4. Bayesian Networks
  5. Multivariate relationships
  6. Synergies
  7. Path to Purchase
  8. Vector Auto Regression ( VAR )
  9. Multiple success criteria ( sales + intermediate steps )
  10. Parameter Estimates
  11. Carry over effects

Results

It was found that e-mail, radio & search Impressions were the leading contributors to dollar sales. While TV & did not directly contribute to sales, they had an abnormally high impact on intermediate steps in the path to purchase – opening of emails sent, software downloads and website visits. VAR was able to discover that a substantial portion of downloads web visits and email opens lead to sales in successive months.

Learning

It was found that e-mail, radio & search Impressions were the leading contributors to dollar sales. While TV & did not directly contribute to sales, they had an abnormally high impact on intermediate steps in the path to purchase – opening of emails sent, software downloads and website visits. VAR was able to discover that a substantial portion of downloads web visits and email opens lead to sales in successive months.