Designing for Automation: Begin with Artisan Processes

Automation is not magic. 

It starts with a complete understanding of the path humans follow and the ways a site can react.

When you set out to automate a personalized ecommerce experience, no system will know what a particular user or audience needs without the crucial factor of human insight. 

A marketing specialist who knows their customer will have to decide what message or part of a message they think will resonate—and then test it. They can then handcraft the message and try a variation, just as an artisan uses knowledge of their craft to create something that will engage with another human. ‘

The marketer will then try other variations and learn what other items might work and which ones do not.

Once you understand what drives messaging and design unique engagement components for it, you can experiment. 

Here is where the metrics and feedback allow for optimization—handcrafted optimization. 

This is not practical or scalable. But is the beginning of understanding how to vary the design elements that can be recombined by a system and process and, ultimately, an algorithm.

One organization had 20 components that could be varied on a landing page: different images, offers, calls to action, phrasing, and headlines that could be rearranged and reassembled like Lego blocks. 

Suppose these elements could be arranged in five slots on the page, with four possible elements in each slot. 

The number of possible combinations of elements that could make up that page is four to the fifth power (4 × 4 ×4 × 4 × 4) or 1,024—over 1,000 possible variations of that page from those Lego blocks.

After breaking the messages up into reusable building blocks, the blocks need to be structured and organized through—as you probably guessedthe ontology. 

The ontology has the instructions for how the pieces fit together by virtue of its models of the content. 

Think of how tiles on a mosaic floor might fit together. 

The pieces have regular patterns that allow them to fit together within predetermined templates that have slots for each of the pieces.

What’s the best way to assemble these hundreds or thousands of possible combinations? 

This is where machine learning algorithms come in. 

The input is the messaging architecture and a collection of components for each design element. The output is a random assembly of those components, which is then tested with audiences. 

The algorithm analyzes the data, picks the winning elements for that audience and journey stage, and then continues to experiment across other audiences, selecting winners in a Darwinian process. 

Over time, the system can generate hundreds or thousands of personalized experiences. 

Other algorithms can determine which products are presented once the customer clicks through on an offering. What that customer sees will depend on past purchase analysis, behaviors of similar audiences, and recent activity signals. 

You can do the same testing with ad copy, promotional offers, and sequences of messages.

This analysis is based on large-scale experimentation. 

It uses a form of AI called a neural network—a computing structure that “learns” patterns by observing the results of experiments. 

The neural network tests various weights to components and audience attributes until they produce optimal outputs. 

The network has a target of maximizing revenue and will experiment until it reaches the best performance possible with those components.

__________________________________________________________________________

Would you like to read the rest of Chapter 5 of The AI Powered Enterprise? Click here