Data-Driven Design Day 2017 with Helsinki Design Week took place on Tuesday 12th September. Sponsored by SC5 and OP Group, the event attracted over 230 people to Valkoinen Sali in the heart of Helsinki and had close to 600 remote viewers. Attendees home and abroad enjoyed learning experiences most focused on data-driven development of ecommerce and retail. Here’s how I see the day’s highlights.
Preface: What is data-driven design?
I have had several shots at defining data-driven design over the years. I have always started from evidence-based design, maintaining that design decisions must be rooted in objective data instead of gut feeling of the highest paid person in the room. In 2016, I compared data-driven design to growth hacking and conversion optimization. This contrasting highlighted that data-driven design is differentiated by method, goals, and ethics from its peers. Often times methods across three approaches are shared, but mindsets rarely so.
In my ACM Interactions article published in 2017, I redefined data-driven design in the light of emerging new domains of design which also heavily rely on data. These new approaches are known as algorithmic or generative design and involve variations of assistive or agentive design technologies (instances of artificial intelligence).
My latest proposal states that data-driven design is about scalable, automated data acquisition and analysis for design research. This makes two important claims. First, design in this fashion is human-agents-first approach, AI does not play a significant role in making design proposal. Second, decision making is still gated by human, not by silicon brains. This is the reflection of reality.
I do think that this state of data-driven design, which you may call “Data-Driven Design 1.0” will gradually fade with first decision making being automated. This is because it is easier of the two and because it is already happening. For instance, Ronny Kohavi of Microsoft has publicly recorded many details of how big companies run their experimental platforms that include automated decision rules to help to weed out experiment ideas that perform below expectations.
Professor Antti Oulasvirta from Aalto University talked about the future of generative design. He thinks that evaluation and generation are the biggest bottlenecks in present day design process. And he also believes that self-designing interfaces will eventually design themselves. To succeed, computational design will need to solve three major challenges: inference, search and prediction. Computational “design” thinking means that computers will learn to find their own solution from the design space and predict their value using models of human thinking and behaviour. This will lead up to “Data-driven design 2.0” if you will.
What happened at DDDD2017?
The theme of 2017 event was retail and ecommerce. My intention was to cater a good overview to this theme by inviting leading Finnish companies that help others do ecommerce (Frosmo and Nosto) as well as a new agency focusing solely on ecommerce (Columbia Road) to the event. Naturally the biggest of their kind from ecommerce (Veikkaus) and retail (Kesko) were also displayed. To keep the program interesting to people outside sales-transactions oriented world, two keynotes discussed the future of design and data (Antti Oulasvirta), and how data helps to find problems and verify problems in a digital company with several product lines (Mail.Ru).
The video recordings of the talks are now available with the exception of Antti Oulasvirta’s talk. I will describe in brief the major themes next. If you are interested in generative design, user interfaces that will design themselves, please contact me or Oulasvirta by email to receive more information!
— Lassi A Liikkanen (@lassial) September 12, 2017
Across the presentations I believe the main repeating theme was that of getting to know the user through data and acting upon it. This leads to two directions.
First, the picture of customers and users gets more detailed, finer grained and realistic. Designers can benefit from this at every instance. Less guesswork, fewer prototype or bullshit personas, more information. However, more of something is not just automatically better. Jussi Mantere illustrated how Kesko had worked hard to make loads of information more accessible to store managers by creating visualization of several kinds.
Second, technical solutions afford personalizing experiences in electronic commerce. This needs human design to begin with, but the solutions displayed by Veikkaus and Nosto all show that once the wheels are set in motion, the digital solution can effectively serve customers personally. Let’s dig a bit deeper into this.
Ecommerce is now personal
Personalization was one of the repeating themes across several presentations. This should not surprise anyone, as personalization is one of the emerging trends in utilizing machine learning in interaction systems. However, the implementations of personal services were increasingly variable, as they occur in several forms.
And why not, large behavioural data that internet merchants have an access to is ideally suited for customization. It only takes some learning solutions that can accommodate variable user needs and behaviours without much intervention of the user.
When I look at the Finnish landscape of digital services and different customers for instance which SC5 serves, the data maturity of organizations is greatly variable. Even in terms of only knowing what their customers and users are up to and ranging up to companies such as Veikkaus or Supercell, which master user data and shape their products around it.
I warmly recommend everyone to watch Eetu Paloheimo’s talk which describes about five years of efforts at Veikkaus leading up to a tailored personalization solution. The path included several technologies, for example, Conductrics. Today their personalized AI solution is responsible for over a third of sales at Veikkaus.fi. I think this is a great case study of how you develop new requirements and can observe new benefits by experimenting with relatively simple solutions at first.
50% of decisions could be made by Conductrics
In post scriptum, you may ask, doesn’t all this personalization go against the definition of DDD mentioned earlier, algorithms defining your experience? Not exactly. Someone still has to outline what the personalized service will be like, the bounds of variation. That someone is a human designer who together with various software tinkerers sets the bounds for what will happen. Personalization currently still happens within pretty strict constraints of content.
These were just few of the highlights of the Data-Driven Design Day 2017, go and watch the recorded streams (note attached PDF’s when available, mentioned in the comments) and learn for yourself!
Find the history of all past Data-Driven Design Day events here: datadrivendesignday.wordpress.com
You can also find the playlist here: https://www.youtube.com/playlist?list=PLNqx9rM5Kv1IS6uznb-gkTf7Niv-SXDHL
Text: Lassi A. Liikkanen, Data-Driven Design Specialist at SC5, @lassial
Featured photo: Mikko Rajala, SC5.