Machine learning is a set of algorithms used to develop systems that can:

• Learn from existing and new data without being explicitly programmed.
• Automatically apply what they have learned to the new data.
• Draw inferences from data sets and make predictions about future outcomes/events.

These algorithms are widely used by big companies like Google, Amazon, Facebook and Netflix to increase user engagement and maximize conversions.

For example, both ‘Amazon’ and ‘Netflix’ use machine learning for its recommendation engines. Facebook uses machine learning to produce its news feed. Google Adwords use machine learning to execute automated bidding strategies. Google Analytics use machine learning to produce data driven attribution model.

Many tools based on behavioral/marketing automation like: Bounce exchange, Infer, giosg, landy etc use machine learning. You can use these machine learning based tools to optimize your website for conversions.

In this article I will show you the various applications of machine learning in conversion optimization which will help you in understanding, how machine learning can be used to maximize conversions. 

Recommendation Engine

Amazon recommendation engine uses machine learning algorithms for recommending products to their customers:

You are recommended products which you are most interested in and which you are most likely to purchase. These recommendations are based on your past, present and possible behavioural patterns and propensity to make a purchase.

The recommendations are personalized for each user/visitor of the website.

So different users are likely to see different products being recommended on the same landing page, at the same time and in the same geolocation. This is called real time personalization which has increased Amazon’s sales by 35%.

35% of Amazon.com sales is generated by its recommendation engine. (Source)

59% of shoppers who have experienced personalization believe it has noticeable influence on purchasing. (Source)

In case of a brand new user with no historical data, the machine learning algorithm make predictions based on similar users and their behavioural patterns.

Ranking Products in search results

According to a new study from a personalization platform company BloomReach:

55 percent of online shoppers in the US start their product searches on Amazon (source)

This makes ranking high on amazon very important for many retailers:

Now unlike Google, Amazon does not use backlinks to rank products. The products can be ranked differently for each user based on their interest, propensity to purchase and the current sales volume of the product itself.

Because of this reason, you will often see products with little to no reviews or lower rating, ranking higher, sometimes even higher than best sellers with hundreds or thousands of five star reviews on Amazon. 

Since Amazon has access to sales data for most people and also at the individual product level they can determine what is selling and what is not. They are in a very good position to optimize their search results and overall user experience for sales. Their main objective is to sell as much as possible.

Adjusting pricing and offers in real time to maximize conversions

All the offers you see on Amazon books, are most likely for the books not selling at the listed price, either for you or the users similar to you:

Book sellers can not set such offers. These offers are dynamically generated by algorithms, in order to push sales. You can set your book list price to whatever you like. But amazon will eventually set your book price to the price point where it is most likely to be bought by you or users similar to you, at a particular point in time.

So different people can see different different prices and promotions/offers for the same product, at the same time and in the same geo location.

#1 If your propensity of buying a product is zero, you may not see any offer as your buying behavior can not be modified by any treatment (sales, discount).

#2 If your propensity of buying a product is very high, you may not need any treatment (sales, discount) to guide you, towards purchase. In that case you may not see any offer. When you are already eager to buy then why offer any discount and reduce the average order value.

#3 If your propensity of buying a product is high but you are not completely sure, you are a candidate who may need a treatment (sales, discount) to guide you towards purchase. In that case you are most likely to see an offer.

#4 If your propensity of buying a product is high but you are not satisfied with the current offer, you may get even better offer (steeper discount).

All of this personalization of offers and prices by machine learning algorithms is done at the individual user level and in the real time.

Are you ‘worthy’? Eligibility criteria for using machine learning tools

In the Norse mythology, ‘Mjölnir’ is the hammer of Thor, a norse god associated with thunder. Only a ‘worthy’ can wield the hammer.

Just like this hammer, not every website/business can benefit from conversion optimization tools which are based on machine learning. This is because they are not ‘worthy’ aka ‘Data Mature’.

A high level of data maturity is required before you can use tools based on machine learning. Otherwise they could do more harm than good.

If a company has a high level of data maturity then it means it has immediate access to high volumes of high quality data from multiple data sources and it has people and processes in place to capitalize on actionable insight in a timely manner. In order to achieve and maintain high quality data, a company would need to eliminate or at least minimize all data collection and data integration issues across their organization. Not many are able to do that.

On the other hand if a company has a low level of data maturity then it means it is either using a high volume of low quality data or a low volume of data (small data sets). It can also mean that the company does not have the people and processes in place to capitalize on actionable insight in a timely manner.

Majority of companies big or small, do not have the level of data maturity required in order to benefit from machine learning.

Yet they use machine learning tools in a hope that by some magic it will improve their website conversions.

This could have dire consequences. Instead of improving sales their conversion rate and sales could tank. Let me give you couple of examples:

#1 Data driven attribution model – Many company use this attribution model tool provided by Google when they are not ‘worthy’ of it. Their conversion tracking is not setup correctly. They are not tracking all of the conversions from all of the data sources. They do not have high volume of high quality conversion data from multiple data sources and yet they are doing attribution modelling. Yet they are taking data driven attribution insight into consideration while taking business and marketing decisions.

#2 Adwords automated bid strategies – Google Adwords use machine learning. Whenever you use any of their automated bidding strategy (target CPA, enhanced cpc, target ROAS etc) adwords rely on their machine learning algorithms to adjust your bids and ad placements. Many advertisers use and rely on these automated bidding strategies to optimize their campaigns for conversions when they are not ‘worthy’ of it.  Their conversion tracking is not setup correctly. They may not have have big enough conversion history, yet they rely on the automated bidding strategies to maximize conversions.

Now vendors of the machine learning tools won’t tell you that you are not ‘worthy’ of their tools. They are in the business of making money. They just want to sell you their products. Whether or not it benefits you, is not really their concern. So Adwords won’t say to you “hold on a second, you do not have any conversion history. So how can I optimize your marketing campaigns to maximize conversions”.

Instead Adwords will just start optimizing your campaigns for conversions based on similar ads in your industry. This may not work for you as each business is different and you could start loosing lot of money as a result. 

Remember a machine learning based tool is only as good as the data you feed to it. If you feed it garbage, it will produce garbage. So you need to first fix your data collection and integration issues before you start using machine learning based tools for optimization.