We don’t want our marketing machinery to run idle. More specifically, we want our website to generate as many leads and customers as possible. This is where Conversion Ratio Optimization (CRO) comes in.
It helps increase the number of people who take the next step towards a purchase. This can be any number of things. For example, we may want to get as many people as possible to register with a form in return for a download of some sort, thus turning anonymous visitors into leads.
This downloadable incentive used to get the contact information is commonly called a lead magnet. Similarly, we want as many leads as possible to turn into paying customers by clicking the purchase button on a sales page, or the checkout button in the web shop.
An important activity in CRO is testing different alternatives of important design objects. This may involve changing copy, imagery, button size, color or positioning, and more.
The process of testing the best version of forms, landing pages, and sales pages is called A/B testing when two versions are tested against each other, and multivariate testing when many alternative versions compete with each other at the same time.
It is rumored that Google tried fifty-two different variations of blue for the hypertext links on their website, to test which color produced most clicks. Perhaps it is true, but it may be apocryphal.
I suspect they test far more versions today, probably using genetic algorithms and machine learning. With machine learning, we can do automated A/B testing (or rather, massively large-scale multivariate testing) automatically. It becomes easy to test what combinations of colors, font, imagery, and positioning works best with your audience.
Using genetic algorithms, new design permutations to test can be generated automatically. Machine learning algorithms will determine which changes work best over time, and use that knowledge for further design permutations.
Algorithms based on reinforcement learning can be useful here. As the tool proposes and tests new design variations using live website traffic, the algorithm is rewarded or punished for new design permutations dependent on if they improved the conversion ratio or not.
In effect, reinforcement algorithms learn by trial and error. AI can thus be used to evolve winning website designs for automated CRO.
Some tools generate and test thousands of different design changes to work out which combination improves the conversion ratio the most. These tools may not only test how design changes on one page affects the conversion rate, but can also measure how a combination of design changes on several pages will produce best the result.
Effectively, each visitor helps teach the system what works and what doesn’t. This enables self-learning AI-driven designs that adapt to your audience and their changed behaviors automatically. This is what a product called Sentient Ascend does.
I had a conversation with Jeremy Miller, Vice President of Marketing at Sentient Technologies, to discuss AI-driven CRO. He gave an example of how this works. Their client, Cosabella, thought that ‘free shipping’ would perform better than ‘family owned’ at getting people to sign up to their newsletter.
They were in fact wrong. It turned out their site visitors shared their same high regard for family values as they did. In justthe first month and half of testing 160 designs, they already saw a 38% improvement in conversions.Such a large increase means a lot of money for some companies.
To do this, Ascend uses evolutionary computation that works similar to natural evolution. Different changes are introduced, and the ones producing better results survive. They are then used as the basis for a further generation of offspring that tries even better changes, and so on.
Translated into Internet marketing and conversion ratio optimization, each web page is represented as a genome. Simulated genetic operators such as crossover and mutation are then performed.
If the parent genomes are chosen among those that convert well, then some of their offspring genomes are likely to perform well too—perhaps even better than their parents.
Each page permutation needs to be tested only to the extent that it is possible to decide whether it is promising (whether it should serve as a parent for the next generation) or should be discarded. I find this type of technology remarkable.
On the subject of automatic generation of webpage design, let’s look at other options like Firedrop. This chatbot-based virtual web designer discusses the website project with you first in a chat window, and then auto-generates a complete website design based on your conversation.
Other tools for AI-driven website design include The Grid, Wix Advanced Design Intelligence, and Bookmark.
Naser Alubaidi, Head of Marketing at Bookmark, explains his tool like this: “Bookmark uses genetic algorithms, machine learning and some human-assisted elements to provide every user with a website that is unique to them, their business and their industry. This is done by first getting some information about a user’s business and then combining that with data gathered previously. The AI software will then make predictions on what sections, elements, images and pages this website should have based on the data gathered.”
Bookmark’s system gets smarter with every new website by learning from each user’s choices and design decisions. Alubaidi says their mission is to make Bookmark the most experienced website designer in the world. They hope it will allow users to save at least 90% of their time and energy when creating and designing a website online. If they or someone else succeeds in doing this, then the market for web designers might plummet.
So far, we haven’t seen many commercial solutions where AI is used to actually generate the creative, such as ad imagery or entire website designs.
I think we will start to see more tools that generate this content automatically in the future, and Sentient and Bookmark are good examples of what is to come. Expect AI to assist in the design of ad creative as well in the near future.