Data Enabled Design

Eva Deckers at UXDX Europe. Video: https://youtu.be/W88vmBhB5Vs

Readable transcript: edited from the recording's captions for readability (fillers and false starts removed, punctuation and section headings added). Wording is the speaker's own. Timestamps are positions in the video. Names marked [?] could not be verified against the audio.

Design with, by, for and through data

[00:00:09] All right. Thank you. Thank you for having me. Good to be here and speak to you all about how do we actually share and discuss about how do we bring UX into what we are doing as a business, and how we keep it there, and how we also evolve that moving forward. I'll talk a little bit about the team that I'm leading, the data-enabled design team.

[00:00:36] Let's say something about data to start with. Design with, by, for, through data. We all know the examples, and I don't have to inform this audience about our peers in the digitally born companies where A/B testing is daily reality and where we do data-driven design practices, and where we very well understand how you validate and understand the impact or the effectiveness of an incremental change in your digital solution.

[00:01:13] On the other end of it, we also see first examples of where artificial intelligence is used to make design decisions. Can an AI actually pick the color of your product? Or can the AI, and that's the example here, actually decide what your homepage looks like every day again?

[00:01:33] We also see examples more leaning into the marketing side of things, where we use growth hacking, more digitally understanding where are we growing our business. Or more in building personal profiles to actually do targeted advertisement. At the same time, we also know that, for example, Procter & Gamble reduced the investments in profiling like that tremendously, because so far they hadn't seen the impact of it, and advertisement didn't necessarily end up at the right place at the right time. So much for personalization.

[00:02:09] Another anecdote is that with the huge forest fires in the US during the summer, the navigation software was actually leading people into the fire, because, oh, it was so nice and quiet a drive over there, no traffic at all. So let's see. There is something to say still about that and what we think about that as designers.

[00:02:41] Because probably some of you actually master this A/B testing part. Most designers still go into sort of a defense mode when you start talking about AI and AI taking over design decisions. That can never be as good as the intuition of the designer, or the experience, or the skill of the designer.

[00:03:01] If you look at the tagline of this conference track: continuous delivery, validating assumptions and nonstop iteration. Getting these right means excellent execution. Well, I couldn't actually agree more. However, in my daily reality, we need, I think, a bit of a different understanding of how you use data in your design process than most of our digitally born peers that solely bring a digital solution. Not saying that that is better or worse, but it's just not cutting it for us.

Philips and why we built a data-enabled design team

[00:03:39] I'm at Philips, and Philips is a health tech company. It reduced actually to that essence recently. We house a portfolio of businesses ranging from precision diagnostics into our healthcare, from mother and child care into image guided therapy. All these businesses are asked to more and more deliver solutions that go across these different silos. So we are asked to move from touch points to solutions, from single products to systems. And with that, of course, also connect these different offerings that we already have in our portfolio.

[00:04:25] At Philips I have nothing to complain about, the user not being in the center of what we do. The mission and the vision and the strategy of Philips as a whole, user is in the middle of that. And we are delivering solutions that are part of the so-called health continuum, and we're also delivering solutions that go across. But of course a few challenges with that. How do you connect them up in a meaningful manner? How do you actually move from your heritage and mainly hardware products into this connected reality of systems?

[00:05:00] From a design point of view, we thought we had to give an answer to this by bringing a team together, like a year and a half ago, that's called data-enabled design. That's our answer to how do you bring data into the design process. We are driven by two main things.

[00:05:19] We set out to understand the value of data to users, and with data maybe also going into artificial intelligence or related developments. So how do we ensure that propositions that encompass data are actually meaningful to our users? And in the healthcare setting, moving forward, most of the propositions we bring, part of them are driven by data, or maybe it's even at the core of it.

[00:05:50] Another element is that if you want to do that, I think you should also use data as a creative material. You should be able to both quantitatively and qualitatively synthesize from the data to understand truly what's going on. The design community, in my perspective, doesn't do that enough yet, and we try to treat data as creative material.

[00:06:14] Obviously it's not about the data itself in the end. Data in itself is not really something. It's about what you learn from it, the insights that you gain from it. So it's more about the questions: how do you design for personalized systems, for ecosystems, or how does a design perspective on the artificial intelligence of Internet of Things look like?

Three reasons designers struggle with data

[00:06:39] I'm not the only one who wonders that. I got this from some design research in the academic environment, and it did some interviews with quite a large group of UX designers on their encounter with machine learning, and I freely translated that to data and intelligence. And there were three learnings.

[00:06:59] So, it's actually quite difficult to envision what data and intelligence can and cannot do, and I think that applies to our design community. Either we are naive: when we have the data we can do anything. Whereas whoever knows something about data, it can be really ugly, and it's the question if what you think is in there is in there.

[00:07:22] The opposite thing is also true. We also, because we lack the knowledge, miss opportunities that are in there. And we're often also not so well placed to challenge our business or research and development stakeholders or colleagues on if what they are doing with the data or with the intelligence is truly meaningful to our end users or to our customers. I can also turn it around: if you look more at my computer science, data science, developer colleagues, they might actually know these kinds of elements, but then not make the translation to why, what, when it's relevant to our customer.

[00:08:05] So designers in general, and that doesn't go for everyone, but I think for most, designers are not trained and lack the tools to work with data and intelligence. It's not part of the regular design curriculum. And it's actually also quite a difficult material to grasp. It's not tangible. You have to somehow be able to code a program or have some kind of tool that gives you access to it.

[00:08:31] And then the third point is that designers are not enough involved in the developments around intelligence and Internet of Things. So what is now an experience-driven viewpoint on these technology developments? Of course the ethical viewpoint is there, but that's gaining quite a bit of traction, of course. But what if we argue from the experience point of view? What if we argue from the behavioral, from the experiential and contextual point of view?

Five ways of working

[00:09:01] As a team we actually set five goals, or ways of working, that we want to address these challenges with. These are the five, I will go through them one by one.

[00:09:12] The first one is that everyone in my team is actually gathering and working with data in the projects that they do. This means both quantitative and qualitative data. And this might sound easier than it is. Because although we might have a lot of devices in our portfolio, it is not said that we just have readily access to the data, or that the data actually contains these behavioral, experiential or contextual data points that we as designers normally use in our qualitative research.

[00:09:46] So, what is actually really in there in this data log about our user? If it's just about the workflow, when did someone hit a button, does that really tell us about how stressed someone is, maybe in an emergency situation? Does it tell us something about that patient and what he or she is going through in that care path?

[00:10:09] So what we do is we build and hack our own assets, and we use third-party devices to actually get an insight into this behavior, into this experience and into that context. We also build manual trackers, in which we get people to think about it a little bit differently, and try to think about what they are tracking and why they are tracking it. And the third example that you see is actually a first exploration more on the big data side, where we try to have a different perspective on population health data. I will show a little bit more about that later.

[00:10:45] The second thing we do is we're using existing tools, and we're building new tools, to move from data to insights. Said before, data is one thing, but what you want, of course, in the design process, is to move to insights and to knowledge, and be able to synthesize on that.

[00:11:02] So we have the more digital tools like growth hacking. We of course also in our studio have our experience flows and our service blueprints. They will be there as well. But what my team is especially working on is, how do we actually get insight into the data from a user experience research point of view? How do we build a dashboard or tools that are able to bring this quantitative and qualitative view together, and really make us understand how do we get from this to a design intervention that makes sense?

[00:11:38] An important element, I think, is the integration of data design and service design. Service designers are inherently the designers that oversee the bigger service flow, that understand how touch points belong to one another and what happens if you change one of these touch points. So you need this service thinking, this system thinking, to combine with the people that actually understand the data flows in these kinds of services or systems. And that together, I think, is the way to build meaningful propositions, differentiating propositions, around data and intelligence.

Data visualization and making the tools accessible

[00:12:16] We already for a few years were investing in data visualization. But we have some problems really kicking it off in our organization, not really landing it outside of the innovation program into the business programs.

[00:12:32] There are multiple reasons for that. One of them is that data visualization, the impact and the difference it makes, often is only understood once you really experience it and when you're working with it. Another thing is that the business choices for underlying platforms sometimes just make it impossible to design for data visualizations that make sense. If your business decides to work with a BI tool like Tableau, we are done. We can work within the Tableau boxes, which might work fine for the direct needs of the business, but from a data visualization point of view it's very limited.

[00:13:14] So we have to think about how do we advise and work together with our business to make choices from a financial point of view, from a functional point of view, from a technology point of view, also from a user experience point of view. And this is done both about how do you use visualization to get insight in the data, but also how you use data visualization as a proposition in itself.

[00:13:40] The last part is that we are building the tools to make exploring with data and intelligence in the design process more accessible to more designers in our organization. Here you see that we build this tool where you can build chatbot conversations on the fly and combine that with sensor and data input that we are getting from a real-life setting.

[00:14:06] The middle one is where you see that we try to annotate data from an experience point of view. Instead of listing events, we're actually looking at, this part of the data actually showed a good experience around brushing your teeth. So how could we use that kind of insight and find it back throughout the data points that we have? And there you see also the flow of this tool, Motion AI or Flow AI, where we can really start to build these flows, use some automation to see if we can also let a system act without actually wizarding everything that we do.

Example one: population health

[00:14:52] Let me show you two examples of what our team works on. The first one is on population health. Population health is all about understanding, for a care provider, healthcare provider: how is my population, my patients, how are they doing, and how do I understand the bigger picture of health in the area that I'm responsible for as a care provider?

[00:15:19] One of the insights was that the things that make us healthy are not necessarily the things that we spend most money on. There's a lot of behavioral and sociocultural elements that make up how healthy we are, but it's not so much used in our healthcare delivery. So how can we actually give insight into this and make that useful for our customers?

[00:15:48] I'm not showing you static images, but this is done in a prototype, and it goes near to industrialization, but it's a prototype where we, together with our research colleagues and the business, think about how could we use these socioeconomic determinants of health, and how do we make them insightful to our customer?

[00:16:11] What you can see here is, up to zip code level, you can see what kind of social and cultural elements influence the social health index. You see areas that have a lower score in the index, and regions that have a higher score in the index. We are then involved because, what is this index? How do you calculate that? That's a machine learning algorithm that tells you something about which attributes actually say something about how healthy people in a zip code area are.

[00:16:47] What you see is that we have different elements. You can see your areas, you can see the attributes. You also have an idea of how accurate your algorithm is based on the filters you choose. And as a healthcare provider, you can use this to get more insights. You can have a look at, for the people you are caring for, what kind of care are they actually consuming, and what would be maybe the care that they need. Or where are they actually getting their care? If I'm in one zip code, where do I go to actually get the care that I need?

[00:17:33] In this way we build a differentiating proposition by three elements. We help our customers to think about market expansion. Which care, and that sounds very commercial, but in the end it's really about, which care do I deliver to which area? So who needs what? Not all zip codes need a non-smoking campaign, or some might.

[00:18:03] The second thing is that we do care optimization. This relates to the same thing: how do you dedicate the care to the area that you're providing it to? And the third one is that you normalize the quality. Sometimes it appears that some care organizations in one place are performing less than the other. If you look at the social determinants of health, it might actually be that one care facility is in a little bit more difficult circumstances than the other. If you can normalize, you can actually see what is the quality, what is the difference they are making, which is valuable information to our customers.

Example two: parent-tracked baby data

[00:18:51] Another showcase I want to show you is around patient-tracked data in healthcare. What we notice is that there's a lot to do about this, the patient getting more and more in charge of their own health. We've just seen an interesting talk about that, of course. But we also see on the healthcare professional side that there are doubts or worries, the worries a little bit more prevalent than the opportunities, in what to do with all of this data, and if the data that is tracked by the patient is actually accurate and valid and if you could base decisions on that.

[00:19:26] This first started off with a project that we did around infant feeding, and we just wanted to learn more about the experience of infant feeding by actually hooking up some sensors into the bottle and understanding what is happening. This was to accelerate on a single connected device, bottle feeding.

[00:19:47] The interesting thing is that it gave insights to our experts in the company, that literally know everything about this, that they had never thought of before, never seen before. And why is that? We focus on bottle feeding, so we focus on the bottle, we focus on the feed, and we think about the drinking. Whereas bottle feeding, or infant feeding, has everything to do with, where did I come from? Was I in a rush? Is there a lot of noise? Am I at home? There are a lot of different elements that determine, let's say, the success of a feed.

[00:20:25] This led us also to think about how can this parent-tracked baby data be relevant to healthcare professionals? And of course this is in the context of a business that we have and a proposition that we are actually offering.

[00:20:43] What we started to do is we really put a... it's more a probe, not really a prototype, into the families' homes, so they can track the data, and they have contact with the healthcare professional via that kit as well. So we do it in the actual context, we use the data as creative material, and I'll show you through a few steps in the process how we've done that.

[00:21:12] We're really thinking qualitatively and quantitatively: what are we learning about the situation, about the behavior again, about the experience and the context, and how does that translate into a design intervention? And that comes through the explorative and continuous part. This is not a pre-built prototype that we put with a customer and let them test it and then take the learnings back. No, we actually put it out there and let it evolve while it's out there. So it's not prototype, test, prototype, test. No, get the insights and make sure that we evolve on the go.

[00:21:52] I'll show you how we've done that. We have the parent toolkit that gives the parents the ability to track some data freely. There are some buttons and they can themselves indicate what I'm using them for. Am I tracking the sleep? Am I tracking crying intensity? Am I tracking the feedings? There's different things one can track. There's a healthcare professional dashboard where the professional gets access to the data and where they can both communicate. And then on our side, we have a research dashboard where we can follow the things that happen, and then we can also from that part intervene with new design concepts.

[00:22:35] As I said, it's in context, so we have a set of professionals that recruit their own patients to participate in a small-scale study like this one. So there are also babies that have a care question. There is something around these parents and babies that there's a question around the care. Just to give you an idea of how small this test actually is. But it gives a lot of insights.

[00:23:04] And then we go in. This is what it looks like in the home. I try to convince my business partners very often about the fact that we make beautiful pictures of what something looks like, but that's not what the home looks like. So here you see it, and that's what we also value.

What we learned from the probe

[00:23:23] People start tracking data. We very quickly notice that they want complete data overviews. Because they are tracking for their healthcare professionals, they find it very important that it's complete and accurate. That might seem like a small thing, but it's actually a big thing for the parents, because they want to ensure that they have the right dialogue with the healthcare professional. So we quickly make sure that there is a way to adjust data points in the system, and that you can also annotate the data points.

[00:23:59] At the same time, the healthcare professional, in this case a preventative nurse, gets a data overload. There's too much going on and she doesn't know where to look. So we built a way to actually filter the data so she can more browse through it. That still is a bit difficult, because she says, "Yeah, there are a lot of data points, but I still don't really get the context. I miss the conversation with the parents explaining what's going on."

[00:24:32] What she says is, normally I work on a question base. A parent has a question or a problem and then I try to anticipate that. So what we do is we make sure that they can have a richer dialogue about what the data means. By changing some of the elements of the design, we put more emphasis on what the trackers are actually used for. So the preventative nurse and the parent have a discussion about, are you tracking, is it actually valuable to track the duration of the sleep of the baby? Or do you want to track how quickly the baby falls asleep? They get more nuance into what it is that they are after.

[00:25:21] Which is a complete difference from all the baby apps that are out there now, where you have a standard set of trackers that tell you what to track. Whereas for every parent there might be a slight nuance to the problem or the challenge they encounter.

[00:25:36] What we also do is that we give them an opportunity to start adding videos to a data point. If they track something about crying, they can then, to that exact data point, add the video of the crying. So the nurse has an understanding of what the parents think is crying your lungs out, to have a better idea of the experience and the context.

[00:26:04] And this then results in a concept in which we saw that the question was about expressing milk, which is quite a serious topic for a young mom. She was not expressing enough breast milk for the baby and she wanted to increase that. She got advice, or she asked the question, and she said, "These are the data points I think are relevant for this question." And then the nurse could see that particular data. After a week it was not going better, so the advice didn't help and they had to revise that. They both agreed that if that didn't happen at that moment, she would have stopped expressing milk already. So you make an impact in someone's life.

[00:26:56] For the higher care, the second or third level care professionals, we see a more controlled approach, where the care professional says, "These are the things that I want you to measure up front of a first consult." We have this case of reflux, where the baby actually spits up. We see that the healthcare professional gets this data about when was the baby getting a bottle, when was it crying, and when was it spitting up. And the interesting thing is, this is all an average. If this child had reflux, they would be spitting in the first half an hour after each bottle. So in one go, the healthcare professional would say, "This is not reflux. It cannot be." It would have taken this healthcare professional one, two, three consults to actually find out that it was not that problem.

What's next

[00:27:49] So this is where we are. Gave you an idea, showed you the challenge. This is what we are after. And there are a few things that are next on the agenda.

[00:28:01] I need to strengthen the bridge to data science and development teams. My designers are very hands-on. They mine, structure, build, prototype, program with data themselves. But of course, if we want to scale up and bring more business impact, we have to build that bridge. Another important point that I made is, how do we influence, from an experience point of view, the choices that are made around platforms and software?

[00:28:36] I can talk quite a bit more about the points that are in there, and one that I'm very passionate about is, how do we portray a designerly point of view towards artificial intelligence? How do you experience attributes in your AI? And also, how, as a design function organization, do you take accountability for your design? How do you go from designing it to actually understanding the impact? And with impact I mean in this case on health outcomes, on sustainability outcomes. Not per se on the conversion rates, but how can we go one step further? How can we actually show what our design interventions mean to the people we design for?