Header Image


In Maths

January 27, 2020

Fact: Machines see details where humans don’t

Let's start with an example: a data science team at a mortgage brokerage company built an automated app for online loan applications. They input all available data from interactions with each mortgage calculator and build profiles that match applications with contracts written with digital IDs. Humans see nothing out of the ordinary with this data. However, machines see details that their human counterparts don’t. Using different methods to analyze the entire data set, some models emerge - one of which is remarkable.

The algorithm confidently anticipated that people who played more than 3.8 seconds with a specific slider on the Mortgage Calculator had a 80% chance of getting good credit risks and being accepted for the loan. The math worked in both directions. If a person plays with a calculator within 3.8 seconds, there is an 80% confidence level, and there is an 85% chance that the person will remain undiscovered. This interaction with advertising reduced human underwriter time by over 40% and increased ROAS effectiveness by over 11%.

You don’t need to know why humans who spend more than 4 seconds on a slide in a mortgage calculator have a 80 percent chance of getting approved for a loan, but you do need to know that the statistics are correct. By the way, when human marketers seek to optimize the calculator to extend the usage of this particular slider, the effectiveness of the application process is reduced. There is no causal relationship between the proposed slider usage and the expected effect. Humans are determined to tell a story about why and how - only the computer did the math. What worked? Getting more people in need of contracts to communicate with the calculator.


Yes, we are storytellers! That's what marketers do: they tell stories. We are engaged in the meta-domain of emotion and the unspoken language of music and color. I get it all. Do you doubt my sanity, I believe in a world-class story. But stories are not everything. Today, they are nowhere near enough. Most marketers spend a lot of time and energy describing the reason why a consumer should be reaching out and grabbing their product. The quest for a reason is by definition a grail quest. It is useless, since no one cares or believes in the Holy Grail.

If you work for the marketing department in the average mega-corporation, you design mood posters, and you target those posters with the names and details of a buyer persona. “John is a 24-year-old undergraduate student. He lives in ABC Town. He has a pit bull named Zap. He is a nester and wants to own his own home one day”, etc. Consumer travel is as inherent as the marketing department of movie studios. It’s all very old school, and it has its place, but it is not that important in the context of today’s marketing.

Correlation and causes

If you spend some time to analyze sales data, models will come out. They are not always obvious. That is why you need to lean more on your data science for help. If you identify patterns, they will often work in getting you actionable results. Actions are empirical. Ultimately, that means that the results can be improved. That's because value is always produced.

Case study

One of my favorite case studies is about a mortgage brokerage firm that wanted to create a mortgage calculator to inspire people to apply for a mortgage. They have built a simple and interactive mortgage calculator that fits in IAB standard ad units. They launched a digital advertising campaign that matched their design goal (John, from above, as the first home buyer). Applications are starting to come in very fast. Too fast to be handled by human underwriters. They needed an algorithm to analyze inbound and score them for their ability to be “good loans”. They decided to combine online interaction data with banking data. Sadly, none of the maths they have done with the 80-ish data points they collect are good for assessing the quality of the mortgage customer, and the bank’s human underwriters, which are associated with the bank’s current black box credit score information, could not make it work. But in an interconnected world, “everything can be counted, but not everything counts.” Your 80-ish online data points tell only a small fraction of the story. It is interesting to learn about abandoned loan applications and to optimize the copy and colors used in advertising to attract more communication with your persona's look-alike. To make it all work together, it is by no mean sufficient to solve the number issues and to speed up bad applications. None of it changes reality.

Money is being printed

If you find a slot machine that pays 5.1 cents for every nickel you put in, you will hack your house and try to find every nickel you can to feed it. This is the difference between searching for reason (narrative) and finding a mathematical correlation that predicts a task with high assurance. It’s not sexy, it’s not sentimental, it’s not even an entertaining thing to do. But today, you have all the tools available to find and use correlations in your marketing statistics. The more capable the data is in the presence of better content, the more visually appealing the visual corporation can systematically generate the most value.

The Mailman

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

MailMinds.com is a service provided by Codecide, a company located in Chicago, IL USA.
linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram