How Bayesian CLV predictions helped a firm optimize Google Adwords spend


Researchers at the Olin Business School evaluated a firm’s search advertising strategy and found that they were significantly undervaluing their customers and underbidding for ads. Here is a brief summary of their findings.

The researchers looked at a retailer of chemical lab supplies, and evaluated their marketing strategy over a three year period. They found that cost-per-click (CPC) has been rising sharply due to intense competition for keywords and an increase in the popularity of search advertising. The firm was wondering if it was worthwhile to continue their bidding strategy.

On average, the firm was spending about $190 per customer they acquired from Google. However, looking at the conversions from Google advertising, they realized the average gross margin was only $142. Furthermore, with CPC rising to $0.80, the story was getting worse.  Cost per acquisition was approaching $266. For Google to be an effective channel, they could only spend $0.37 per customer, so they were spending above their break even point.

The story changed when researchers helped the firm look beyond the initial conversion.  The researchers used predictive models to project the customer lifetime value of customers acquired from Google Adwords and found that customers were worth much more than the value of their first transaction. Customers often returned, and they ordered with increasing frequency the longer they had done business with the firm. The researchers found that the 1-year gross margin from customers from google was $562.  This raised the break-even price of customers to $1.79, which was well above the current CPC.

Cost Per Click Rises, surpassing the first transaction gross margin, but remains less than the lifetime gross margin.

Furthermore, they found that customers who came from paid search were significantly more valuable than those who came from other channels. The average cost 1-year gross margin of customers who came from google was nearly twice that of customers who came from other sources ($562 vs. $289).

By taking into account the direct results of advertisements, and looking at transaction data in aggregate, they would have significantly undervalued their customers and abandoned the CPC channel. But by using advanced analytics, they realized the true value of their customers, and continued to invest in paid search.

“Measuring the Lifetime Value of Customers Acquired from Google Search Advertising,” Marketing Science, Tat Y. Chan, Chunhua Wu and Ying Xie
A note about methodologies: Chan et al. used a probability model derived from the work of Schmittlein et al.. They use the same individual level likelihood function, but use a log-normal mixing distribution. The use of the log normal mixing distribution allows the introduction of correlation between the order frequency, gross margin, and attrition. Most of which they find to be uncorrelated. At Custora we use the original model developed by Schmittlein to predict customer lifetime value. For more information on different methods of predicting CLV, look at our other posts.


Infinite Content


2018 Marketing ‘Prerequisites for Success’

The Adobe 2018 Digital Trends report was recently released and it gave its readers some interesting insights into this year’s top marketing priorities around data, design, and technology. The report, which consisted of close to 13,000 participants from various industries including retail, e-commerce and the marketing agency sector, revealed the trends that distinguished the […]