Keeping things human in the age of machine learning

You’ve probably read ad nauseam recently about algorithms, AI, and machine learning. These technologies have enormously impacted engineers and product experiences but often overlooked are those who have the job title Product Designer.

How do you make your product design a coherent and quality experience, but flexible enough to evolve and personalize? Content, ordering, and layouts may change per customer and evolve even more over time. What people see when they interact with a highly personalized product is tremendously influential. It can inspire loyalty or it can erode trust and credibility.

As a designer, you are creating a playground for algorithms and machine learning to play around in. There are different types of people and objects that can move around and do different things, but at the same time there are boundaries and rules. Like real playgrounds, both good and bad things can happen within them. It’s your job as a designer to work with your team to help set the rules and constraints, and also make it a safe place.

Communicating design intent is a critical skill for product designers. The following are some tips I’ve learned along the way while designing experiences fueled by algorithms and machine learning.

Join the conversation and become a partner

Spend time learning the vocabulary of your development and product management teams. Don’t be afraid to ask stupid questions! At many companies, it’s probably not the cultural norm for designers to be part of this team.

Here’s my simple argument for including product designers:
An algorithm is a set of rules a computer uses to perform a task. Machine learning requires additional complexity and constraints for how the rules can change over time. People make these rules and constraints, not the computer. Product designers represent the voice of the customer and bring research insights, design strategy and content intent to the product so it makes sense that designers should be part of the team defining the rules.

Do your homework

The quality of personalized content not only affects the overall experience, but can also dramatically affect the perception of trust and quality from the customer, and impact your A/B tests and experiments. Pay as much attention to the details in personalized content as you would to visual or interaction details.

  • What do the existing algorithms do? How frequently are they updated?
  • Where does the content come from? What are the potential sources?
  • What are the options to nuance the experience?
  • Are the results too narrow or too broad?
  • How do all the pieces work in combination? How are they ordered, weighted or prioritized? (Tip: if you hear the terms Multi-armed Bandit or Contextual Bandit ask about its rules and constraints.)
  • Do layouts change under different circumstances? What is the framework? Does it work well for all common customer stages and scenarios of use?
  • Does the team have an agreement on your ethics standards to avoid bias?

Design for real content and real situations

Use real content. I hope all digital designers have learned by now to not use lorem ipsum in designs. Tools like Sketch and Framer allow you to pull in real content and reduce manual labor for your mockups and prototypes.

  • Take your content strategy game to the next level and communicate the content intent to your team. Explain how you see it adapting in a few key situations through mockups, specs or prototypes. Think through the different customer stages and edge cases with your team.
  • If you have the resources available, partner with a developer to make a functional prototype or proof of concept using real data.
  • Look at different scenarios of real content in product. Your QA team may have accounts that represent customers in different stages and types of personalization. Your engineering team may be able to simulate variations. Regardless of the method you use, don’t skip this step and if you don’t have much support, maintain your own separate accounts for testing purposes.

Invite the humans into the experience

Algorithms look backwards to historical actions of a customer. If little is known about the customer, is their experience sparsely populated or back-filled with other content? Algorithms tend to be slower to grasp new intents and interests of the customer and the content can feel stale or irrelevant. How might you introduce fresher or brand-new content into the mix? Machine learning may not have the right constraints in place and may evolve in unintended ways.

Explore ways to mix customization into your personalization:

  • Make your personalization more transparent and invite people to give feedback to stop showing irrelevant info, or inversely to receive more of something they particularly like.
  • Invite people to give input at any time for their priorities, interests or intentions so they don’t have to wait for relevant content to appear or to organize in ways that make sense to them.
  • Speak up if you feel the experience is feeling too robotic and show your examples. Add or take away parts of the experience to keep it feeling human.

Define what “best” means

The goal of personalized experiences is to surface the best content for each customer, so in the end what this means is that your team creates a definition for what “best” means to your customers, to the best of your combined abilities. No one company or product has this figured out. Create an open culture of discussion and incremental improvement. And most of all, bring a sense of adventure to your design because the one thing you can count on is this: it will always be changing.

Author: Jenny Kolcun

Collect by: