Data science lies at the intersection of a huge array of modern digital processes, products, and applications. Today, UX design is no longer dictated by arbitrary concepts and has become all the more data-driven. Of course, the importance of quality UX design remains as good as ever:
- Adobe found that users want to consume something visually beautiful rather than visually plain and that 38% will stop using a site if it’s unattractive.
- SmallBizGenius cite that 88% of shoppers don’t return to sites that provide a poor user experience and that 45% want content that displays responsively on smartphones and tablets.
- TechJury found that users are 74% more likely to return to a site that has quality mobile UX and that 79% of users will leave if they can’t find what they want quickly and easily.
Data science responds to these preferences and pain points. Designing a fully-functional UX that works with fluidity is just one half of the story – users also want that UX to fulfill a practical purpose.
Here is how data science can improve UX design.
Questions And Answers In Data
Fundamentally speaking, data in UX design provides answers to questions. Developing the problem space and working out how data can help involves the following:
Qualitatively assessing the website or product:
What is the goal of the site or product? Is it an eCommerce store looking to make more sales, a B2B agency looking to attract more warm leads, a SaaS platform looking to convert large-scale customers, or a blog looking to convert traffic into affiliate sales or referrals? Or perhaps the site is non-profit and is simply looking for a way to better engage users. What can be improved to achieve those goals?
Developing a hypothesis:
Once a rough goal is established, it’s time to envision the problem space and develop a hypothesis that data science can address. In many situations, UX design can help reduce bounce rates, increase conversions and increase device compatibility. A now-ubiquitous example of data-driven UX design is recommendation engines and personalized content, e.g. storefronts that present personalized products.
Developing experiments:
Next, it’s time to develop experiments to test the hypothesis and generate some ideas for action. Typical examples include AB testing, gathering interaction data from specific areas of the website, heat maps, etc.
Actioning findings:
Put findings into action and then monitor change. If the results are not positive, it’s time to reiterate until a winning formula is found.
Five Methods for Using Data in UX Design
Creative uses for data in UX design are emerging all the time. Perhaps the frontier, for now, is building powerful personalized content and recommendation systems that are both sensitive and accurate.
Major eCommerce sites, as well as TV and music apps like Netflix and Spotify, are already exceptionally good at personalization and recommended. In the future, AI and machine learning (ML) will likely power increasingly powerful recommendation engines that display effective personalized content for users based on a wide range of entity and event data.
Using Customer Data To Redesign The Site or Product
Fundamental to pretty much any and every website or product are its users. After all, websites and products are (currently) designed for human users. Whilst very large websites like Wikipedia and Amazon attract customers from virtually every demographic worldwide, this is unlikely to be the case for smaller websites.
UX also revolves around your users and their preferences – different demographics have different preferences. Using data from either your own customer data databases or Google Analytics (specifically the Users tab), you can gauge who your audience is and develop a strategy to fine-tune your site to that particular demographic.
These principles apply to products and apps too – data can inform the design of new product UX features.
For example:
- If your customer data and website users are signaling an over-60s audience then you might wish to make your website more easily navigable for that demographic, also potentially adding some extra accessibility features. Additionally, a complex website design may not be necessary here. Create test content (or even just a landing page) to experiment with fine-tuning for that audience and measure whether user engagement, retention, and sales increase, decrease or stay the same.
- If your customer data and website users signal a Gen-Z, millennial, or otherwise younger audience, they might appreciate a more design-intensive and modern website. Younger audiences could be more design-motivated and testing a sleeker, modern and more complex version of the site might be wise. Analyze findings to discover whether changes are conducive to more sales, etc. In both of these examples, experimenting with color usage, typography, and other visual design elements might also be useful.
Heat Mapping the User Journey
It’s near-enough impossible to know precisely how users interact with your site without analyzing heatmap data. Heat maps enable you to:
- Discover how far users scroll down your pages (scroll depth).
- Find out whether there are any particularly engaging elements that users tend to click on a lot.
- See where the user’s cursor is when they exit the page, providing clues as to why they click off.
- Analyze which CTAs or product placements generate the most clicks.
- Compare mobile to desktop to see if mobile UX features are as effective.
Heatmap data can be used to build new UX features, pages, and content. An example might be moving CTAs, subscription boxes, or product carousels. Analyze data from new iterations to see whether there’s any change in user behavior. Every sale counts and moving a box or CTA based on heatmap findings could increase sales dramatically over time.
Landing Page A/B or Multivariate Testing
Landing page testing is a classic application of data in UX design. The concept is very simple; you take two or more landing pages, collect click-through and conversion data from all variations, compare the results, and then roll-out changes based on what you find.
Professional data scientists and analysts will be able to determine whether the findings of such an experiment are empirically robust, or whether they’re the result of the novelty effect, Twyman’s law, or some other false positive. It’s crucial to not make hasty decisions based on incomplete, poor-quality, sparse, or otherwise flawed data.
- Testing different landing pages is a fairly simple and powerful means to use data in UX design.
- It’s usually advisable to build multiple variants rather than just two.
- Don’t make hasty decisions based on sparse data and beware of methodological issues and false positives such as the novelty effect and Twyman’s law.
Personalization and Recommendations
Many sophisticated websites today use some form of personalization or recommendations to further engage visitors and customers.
All major eCommerce sites use personalization and recommendations to engage users, including Amazon, eBay, Airbnb, Uber, etc. For example, Airbnb uses location and search history to determine where a user is and what they’re looking for. Amazon and eBay draw on a vast selection of customer searches and other data collected from first-party and third-party sources.
The algorithms behind recommendation systems take user data (e.g. location, demographic data, past searches, purchasing history, etc) and use that to display new products and personalized content that is relevant to the specific user.
Models used here are typically probabilistic and calculate the chance of X item being purchased based on Y interests and prior purchases. The end goal is to provide a personalized experience that is quicker and more accessible, also driving sales and conversions.
- Recommendations engines make decisions based on entity (customer) data and event data (e.g. purchases).
- Similarly, dynamic content can be displayed based on purchasing history (e.g. how-to guides associated with the user’s last purchase).
- Products are labeled to trigger their display in the UX for different customers.
Segmentation
Tied in closely with the above, it’s now possible to build adaptive segments that shape websites with dynamic content. This might mean that a site displays different content, products, and messaging for an over-50s visitor compared to an under-20s visitor, for example.
Segmentation data can be collected from the customer database or CRM, cookies, tracking pixels, or some other means.
It’s usually not necessary to go overboard with segmentation as it’s very difficult to understand everything about your users from typical customer or user datasets.
But, it’s still possible to infer that different segments of users might have different common interests that can guide UX design. For example, a beverage site that sells all sorts of teas and coffees could adapt to produce more tea-related content for serial tea buyers and more coffee-related content for serial coffee buyers.
- Segmenting users allows different UX iterations to be displayed for different groups of customers.
- The same data can be fed into recommendations engines, marketing campaigns, etc.
- Over-segmentation can be detrimental; measure your audience but avoid pigeonholing them too early in the UX design process.
Summary: How Can Data Science Improve UX Design
UX design is no longer a purely creative discipline that relies on assumptions and arbitrary choices. Data makes every UX decision measurable, allowing UX designers to dive into the nuances of how users interact with a site or product.
There are various ways to apply data science in UX design and UX designers, product teams, and data scientists alike are coming up with new ideas all the time. From powerful dynamic content and personalization to recommendation engines and segmented offerings, the entire concept of the UX is changing.
