How AI is driving user experience

Remember the time you needed a new outfit and would simply go to the tailor to get it fitted for you? If not, it’s probably because you weren’t born before the Industrial Revolution or don’t usually indulge in what is now mostly reserved for high-end garments. Instead, we have grown accustomed to choosing among standard sizes or even trying to change ourselves to fit the clothing. 

The same goes for most products of the standardisation era. Historically, designers have gathered data (market surveys, product feedback and viewership statistics) to try and identify relevant content for audiences to consume. Based on the quality of data available these decisions would range from informed to educated guesses.

This ‘one-size-fits-most’ model was inherited from the publishing and diffusion age, where the aim once again was to find that one piece of content that would have the maximum impact. Think TV audiences or newspaper readership.

But on the users’ end, the quality of experience matters more than the quantity. Following this principle, the personalisation trend that started about a decade ago (popular and trending content) has become the new norm. Companies can now capture value from users in the form of the data they provide by using their services. That value is then redistributed in turn, to users in the shape of a customised (and in most cases superior) experience.

Techniques such as collaborative filtering used every day by large companies suggest personalised content to their users, effectively transferring the choice of content from designers and publishers to ‘users that are similar to you’. Rather than assuming what is the best experience for most cheap data and dynamic content delivery lets companies deliver a good experience for each user.

UNDERSTANDING IS CARING

Moving forward, we see machine learning will further strengthen user experiences across industries in several ways:

Experience designers can get stronger insight into the market 

The 21st Century has brought focus on customer behaviour to understand how to create better value for them. Data generated by users can be analysed with clustering algorithms to identify underlying trends that could escape the notice of researchers. This can then help generate databased personas that represent customer segments that were previously untapped.

Relevant content delivered to users

a) Content curated by companies is more relevant to a user

Human-centred machine learning captures users’ decisions and preferences over time, helping designers understand what creates value or not. Content managers are empowered to verify on a larger scale the hypotheses that they have about people engaging with certain things over others. That is, arguably, one of the key advantages machine learning brings to the table: helping people who try to engage you as much as possible make objective decisions, of course, specific applications can be a little scary to a user, such as hyper-targeted ads. 

b) Content ‘chosen’ by peers

A canonical example of content curation by peers is the Spotify music discovery algorithm. In order to suggest relevant music to their customers, Spotify uses, amongst other techniques, collaborative filtering. Simply put, their algorithm finds out who, in their database of users, listens to similar music, as people are more likely to appreciate songs from users with similar tastes. As a result, Spotify looks for songs that you don’t know from users with similar tastes that have already liked them.

This technique alone is not enough to determine the best content for you and Spotify combines it with other algorithms but it already shows how the power is gradually shifting from content managers to users themselves.

ANTICIPATING USER NEEDS HELPS REDUCE FRICTION AND INCREASE CONVERSIONS

Modelling human intelligence

When you enter a fancy store, the sales staff automatically assess you to help infer what products they should be offering you.

By profiling in this way, you are essentially spotting and removing entry barriers and friction points in your conversion path, not as a general rule but as a tailored experience.

How do you replicate this intelligence with modern-day analytics and dynamic content? Analysts, as we’ve traditionally known them, were used to observe user behaviour and understand retrospectively the parts of a digital experience that worked and those that didn’t. 

Machine learning can now help spot trends in data that can help you inform and predict future behaviours.

Asking questions such as “how do people who buy services from me typically behave?” can now lead to computed answers, giving a different weight to each of the potential determining factors. In other words, machine learning can observe your historical transaction path data, formulate hypotheses about what it takes to purchase, and what factors contribute to purchasing, and build models that predict whether that new customer on your site is going to buy or not (and how much attention you should give him).

Predicting relevant next steps in user journeys

Machine learning can conversely spot which events are likelier to cause a user drop off, and intervene before it’s too late. Imagine you notice that users who follow a specific navigation path on your site tend to drop out a lot more than others. When your algorithm spots that a user is following that path, it can push dynamic changes in the design seen by this user to influence them out of this path – typically by reorganising the hierarchy of the content on the page.

For example, you’re selling clothes and have one costly dress to sell among the many, and because it’s expensive people get discouraged from your site. However, those who do stay and purchase that dress generate more revenue at once for you. In old-style content management, you would compute the potential revenue lost with your many discouraged users (opportunity cost) versus the revenue generated by that dress, and decide to either keep the item or de-list it.

In this scenario, you can discriminate the users who bought that dress versus the ones who bounced off and build an algorithm to determine what factors influenced that decision (e.g. their demographic, the previous pages they visited, the last purchased products and how they entered the site).

Equipped with this knowledge, you can spot the users that would probably bounce off the dress page and take the initiative to show them another, maybe less expensive dress instead.

SHOULD YOU USE AI IN 2018?

Consider the costs

Computed solutions are very powerful, but they come with an implementation cost. Simply put, it takes time and skill to build machine learning algorithms, and for now, many companies might feel that they don’t have the resources to implement it.

From a commercial standpoint, you will want to weigh the resource cost of implementing these algorithms against the potential gains. The downside, however, is that it’s hard to know exactly how much you can gain from implementing these algorithms since its dependent on the actual opportunities that will be uncovered.

Heuristics still matter

Your decision on whether to implement machine-learning algorithms in your experience design process should be based on 3 factors:

1. How expensive in resources the implementation will be, e.g. talent acquisition, time and money

2. The volume of transactions or traffic you have that feeds the algorithm (it often takes larger volumes to make algorithms really good at predicting what experience would work best)

3. What low-hanging fruits are still available? (usability audit)

If you’re not feeling adventurous, you can stick to conversion heuristics that – spoiler alert – are actually usability heuristics. If people can’t intuitively navigate your platform, the chances are you’re losing some customers.

It is a good place to start your journey. In most cases it will be way cheaper to hire usability and conversion experts for an audit rather than implementing algorithms to reveal you could be doing things a lot of better.

Flow Fournier is senior UX manager at RBBi