Scott Parsons
02-August-2007 WSG Sydney meeting
PodCast Duration: 29:34 - File size: 7.07 MB
Ambient Personalization: Applying what we can learn about a user's (and groups of users) behaviour, to improving their, and by extension everyone's, experience of a site.
The following is a presentation recorded at the Australian Museum on 2 August 2007 for the Web Standards Group. The presenter was Scott Parsons speaking on Ambient Personalization: Applying what we can learn about a user's (and groups of users) behaviour, to improve their experience of a site. Scott Parsons is an experienced - user experience and web design professional who has worked in NZ, UK and Australia for many media and advertising companies. He specialises in usability and user centered design, while keeping up with future trends and web design techniques.
I've come here to talk to you tonight about ambient personalisation which, apart from being a real mouthful to say and I'm kind of regretting I called it that because I'm sick of saying ambient personalisation, is all about making a usability, user experience better. So essentially this is me. I guess it's like net vibes. It's a page where you can add your own feeds, got a feed for a twitter, all my friends flick photographs up there. Basically it's something which I take it quite serious and I put together my own stuff. I use browsers like Firefox and I install large numbers of extensions. I use large numbers of tabs. I basically - basically I personalise. But this is my MySpace page. Now, have you ever seen a MySpace page? This is the way it looks when you sign up but don't do anything. I initially signed up for MySpace about three years ago, I think. Not a lot's happened since. This is my WriteWith page. WriteWith page.
WriteWith is just another web application out there which I signed up for and pretty much that's it. So sometimes I don't personalise. Sometimes I can't be bothered. And this is what people say. Whenever I have a usability testing session or just talking to people in general I ask them whether they use personalisation features, whether they want to actually try and make a site work better for them, put in that little bit of effort to build their Yahoo page or something like that. And the experience I get is that people say it's too hard. They say I already have everything I need. But most often, I simply can't be bothered. But one thing I've discovered is in the immortal words of one of my favourite TV personalities - everybody lies. Well, everybody lies. If personalisation is about getting what you want, you just have to get people to want something. The idea of ambient personalisation is that the user gets what they want just by using the site.
So we're not asking them to go out and put together little sign into things, put together little RSS feeds, use little widgets. Basically what I'm trying to do when I talk about ambience is I'm just talking about the environment that the user is actually acting in. And using that environment to make personal, or make more personal, the site they're using. But what does that really mean? I guess I've said it before, that increasing the utility of a site by reacting to how the users use the site is what I'm after, but most people when I say that go all blank and really I have to move on. So to take from Plato, human behaviour flows from three main sources: desire, emotion and knowledge. Now, you can usually trust Plato on most things. So I looked up something like desire lines. Now, many of you might have head of desire lines. It's a fairly simple concept shown by this photo quite well. Many of you have seen this photo.
It's a picture of a park with nicely landscaped, it's got nice paved paths going in there all over the place, but as you can see running right across from the bottom right to the top left there's a path which people have preferred to use and enough people have used it, but they've made their own path. And that's what's called a desire line. Now, in the real work when something like that happens you have three choices to react to it. The first choice is block the path. Well, we kind of try to move away from blocking what users do these days, so I'll move on from that one. The next choice is just let people use the desire lines. Now, that's great. In fact, maybe that's all we need to do. But the problem with that is when something unexpected happens. For example, that nice dirt path is great, unless it rains and then it gets all muddy and people have to then go back and use the paths they don't like to use. Which moves on to the third option. Pave new paths. Okay. So we can't just pave the whole world. That doesn't make sense.
As a case study I wanted to have a look at an example of a supermarket website. And example of Jan who wants organic peas. Now, Jan goes to the supermarket site and of course when she looks at the supermarket site she has to try and find out where are the peas. Is there an organic section? A fresh vegetable section? A pea section? Probably not a pea section. In fact there's so many items, so many paths that Jan could take that she could end up looking two or three different places before she finds what she's looking for. But if we watch what she does, if we watch the paths that she takes and then watch what she finally ends up getting, we can begin to get an idea not only what she actually wanted at the end, but how she wanted it and where she expected to find it. Which is where we start to think about how we can use ambient personalisation.
Now, I thought about different ways of using personalisation, ambient personalisation. See, I'm getting really sick of saying it already. In the first part I thought about was easy ambient personalisation. This is the kind of stuff you already see a lot on the internet. If you've even gone to Amazon and you've looked for the book you'll find a section on the page, “People who bought this also bought”. Now, that's great. I personally have found a huge number of things which I wanted to buy just by looking at that section. But it doesn't go all the way. Another thing you can use is, “The last time you shopped for”. So basically someone comes to a website and they can find out last time they shopped for such and such, so that's going to lead them on. Again it goes so far. But I want to go further, which is where I start talking about hard ambient personalisation.
This is where I'm talking about things like tailoring a home page. Last time you bought organic food. So this time when you come to the website the organic food is up front. We change the structure of the actual home page. We put the stuff that you search for, and not just the items that you search for, but the related items to that, front and centre on the home page. This is all part of remembering the user's references. And here again I'm not talking about making people sign up for user accounts and that. I'm saying just where you can, things like using cookies, when you can remember all you can about the person and by remembering that we can do things like saving last time's shopping list or offering recipes on initial items based on things you've already got in the history.
Then I move on to what I call developer's nightmare ambient personalisation. This is where I think it really gets interesting and what I'm talking about here is changing the structure of the site and the hierarchy of information dependent upon user choices. What this means is very basically we're starting off with serving a completely different structured website to every person. What this means is over time gathering the way that people react to the website, gathering information on the paths they take to find the information, and then remembering that person and allowing the system to actually work to their benefit and change when they come to the website. So we're talking about a system which there is no one website. There is no website which this element is always going to be in the top left, this element is going to be above the fold. We're talking about a website where if I go to it I get a different website than if you go to it. And also - and this is the thing which the discussion comes up - a website that changes as I use it.
The idea of adapting a site to allow the user's past to become smooth is where I'm talking about paving those desire lines. Now, obviously in the real world, as I said before, you can't pave every desire line because you'd just get flat asphalt. And this is where you obviously, the problem of hard ambient personalisation comes in because if industrial action website changes, for example, every time a user goes to it they'll never be able to learn it. But between that and the way we currently do things where mostly we serve up the same information based on our taxonomy we've created every single time, I'm trying to find the middle point. And that's where I think about the fact that the taxonomy has to expand. And while we're talking about the case study of the supermarket, it's a fairly basic principle in supermarkets. When you go into them they all have the aisles and you find certain items in certain places.
And of course at this stage I'd take a quote from David Weinberger and his book Everything is Miscellaneous, where he said: In the past everything had its one place, the physical world demanded it. But now everything has its places. Multiple categories, multiple shelves. Simply put, everything is miscellaneous. And what this means in terms of expanding the taxonomy is that you can have the option to put things everywhere. You just have to work out how to do it intelligently. Which is where I start moving beyond the particular case study and I start thinking about different ways that you could actually use the idea of ambient personalisation. One of the first ones I came up with was the idea of intelligence search. Now, I thought it would be really good if I had a search that watches your search behaviour. It sees which queries you put it, it sees which links you click from those queries and from that filters.
So if you always search for computer terms and then you search for apple you get that, not that. Going beyond that, though, imagine a search which continuously learns your preferences so that essentially every time you go there it gets a little bit more accurate. Now this could, of course, be built by Google or it could be built on top of a search engine like Google where your own computer stores your own preferences and filters out the results that you get from Google. Another use I thought was email filtering agents. Not spam filters, but behaviour filters. So imagine Doug from work, he keeps sending email jokes, bad ones, and he does send the occasional work email so you can't really just block him entirely. Well, I thought to myself how about a filter which learns which ones you want to read? How about a filter which actually catches when Doug has sent you an email which has been forwarded from his latest joke newsletter and another one which catches when he sent you that meeting request that if you miss you're out of a job?
It's a difficult concept at times to think of a computer, letting a computer make these kind of decisions, but the whole idea is not that it's a destructive system, but it's a system where you can actually split things out. Of course there are a lot of issues with the whole concept of ambient personalisation which I'm sure one of the top ones in most peoples' minds is privacy. Because the whole thing is - when I was talking earlier the supermarket example, of coming back to the website and without making someone log in, presenting them with their last items shopped, suggestions and catering the website to their needs? Well that's great, but you don't want to freak people out. And that's what might happen, for example, if I start coming back to a website and that website strangely just seems to be knowing who I am. A lot of people don't like that. Another thing you don't want to do these days, of course, is break laws because privacy laws, depending upon different countries, can be quite vigorous and to break them - well....
The other thing is physical world limitations. Now, what I mean by this is that people still expect to be a physically limited hierarchy. They map their understanding of the electronic world to the physical world. And if we start organising taxonomies that aren't based on the physical world, that allow things to be in multiple places at the same time, they allow things to change places depending upon how we use it, it's very possible it could start confusing users. The other thing is, what do you do when you're not yourself? When I buy a Hello Kitty purse - for my niece - I'm not being myself. But how do I get the system to understand that I'm not being myself? Amazon, for example, in its recommendation engine you can actually go into individual items and say I don't want this to be added to my recommendations, but that's the old style of personalisation. That's making the user jump through hoops.
What if you built a system which was smart enough to actually figure out when something unusual happened? I think you should buy Hello Kitty. So if I bought the Hello Kitty purse for my niece, then the system should be able to look at that and say you know, that's unusual. Unless he buys some other Hello Kitty merchandise we won't decide that he's changed his preferences completely. I think that's a good system. It means that you don't actually, you don't buy one thing, for example as a gift, and then have your recommendations flooded with more of those kind of items. It's a tough system to implement, though, because at what point do you want to try and implement some of the changing habits and at what point do you want to ignore the stuff? One of the keys to usability is understanding users and this is really the whole point of what I'm talking about here. But I'm talking about it from a slightly different angle because how much better would it be if we designed our systems to understand our users? And sure, some people might get a little bit worried about whether or not it was good, but if we can do it to efficient enough level, once we train our systems to understand users, we can design more and more systems that enhance the usability of all our websites and all our projects. Thanks.
Question: I suppose coming from a more a traditional - and this is one of those things you go into every company and they've got a different definition - but in my experience customisation has been when the person has set something, so like Net Vibes, you've set up your flicker, whereas personalisation is when the technology itself determines what the change will be. So in essence your ambient personalisation, is that correct?
Answer: Yes, I guess at that stage you would be calling it ambient customisation. I've never particularly used that term, but ...
Question: But isn't it just, isn't ambient personalisation the same as personalisation by that definition?
Answer: Yes it is.
Question: The technology determines what the interface will look like or what options are preserved?
Answer: I think it is. I mean, part of when I talked about what Amazon does at the moment when you go to one of their pages and they have other books that are similar to the ones you're using, that is essentially what I'm talking about, but that's the simple level of it. What I was trying to bring about, to bring up, was the concept of going steps beyond that and actually having complete personalisation - well, what I'm calling personalisation - on a completely different level. And that's when I talked about the hard developer's nightmare and ambient personalisation, that's where I was really starting to think about the possibilities of what you could do if you had an intelligence system in the website which basically catered itself to either a user type or an individual user every time they went to it. And I guess that's the real difference with what I'm trying to do because the easy stuff, the things that Amazon does at the moment, anyone can do and to be honest it's pretty much, I'd say, trivial.
And some people might disagree with that because there's a lot of stuff you have to do to actually work out how to get things working together in recommendation engines. I mean, it's been a competition going with Net Flicks for about a year or so offering a million dollars for an algorithm that will get a great recommendation engine. So yes, I see exactly what you're saying. But this kind of stuff already exists. All I'm trying to say is think about what we could, how much further we could go and I guess that's really what I'm --
Question: So it's where you've got that desire line across the path that's determining what's simply me wandering across the grass and when does it actually become a path?
Answer: Yes.
Question: When does it start cutting into the grass and when does it really become --
Answer: That's right, and also deciding when to pave it and whether that makes sense to pave it or not.
Question: Paving it for that individual?
Answer: For that individual or for a group or for everyone or just, yes.
Question: You were talking about freaking people out when it's changing. I think that's like really important. I know when I'm showing my mum and dad something they have to write down every step, everything how. I think most of the things you're talking about could be implemented, but how do we figure out the line between freaking people out and providing a good service?
Answer: Well, I think that's the job of usability professionals, is figuring out where that line is a lot of the time. But we do that all the time when we're trying to figure out when we're making a contact form and we're trying to decide whether adding gender, postcode, address and family pet is going to put people signing up for something. It's part of the job, is figuring out where that line is. And I guess the thing is that this is something which you have to come back to. It can never be for like for the first time you visit a site. It's about a site learning you. It's not about a - so if you're - well, if my mother, because my mother's the same as yours I'm sure... grandmother?
Question: You weren't on *.
Answer: My parents find it hard to use websites. You know, new stuff, they find it difficult. So if they get use to something they want to get use to it. But I guess that's where the subtlety comes into it because if you come back to a website and suddenly it's completely changed, just with no warning, well that's going to freak people out. And I think what we have to think about, if this is a good idea and actually we're all trying to implement this kind of stuff, we have to try and think about how to do it in a subtle manner, but also how to do it progressively and for the right groups and stuff like that. So if you've got someone who comes in and maybe we watch a user and they come in and they're really slow to try and pick up the links and they kind of, like they sit there and their mouse dithers over the page for 30 seconds before deciding where to click. Maybe we can learn that too. Maybe we can learn that that user isn't confident and therefore not change it to suit them.
Question: You talked about the idea of websites, learning users' behaviour and preferences and so on, but don't users also learn website themselves? I'm just imagining if I were to go to a website a number of times and I realise that the link that I click the most is the third from the top, for example, and I keep going there and then the website goes ah, he's clicking that link all the time, and so the next time I go it's actually moved to the top, but I'm a spatial visual kind of person so I'm looking oh, where's the link gone?
Answer: Yes.
Question: So the two things could actually work against each other, couldn't they?
Answer: They certainly could. And one of the big things about usability is that most often when you talk about usability we talk about people not using expert interfaces, but using first time interfaces. So we're talking about people just coming to a website and being able to understand straight off what to do, whereas a lot of the stuff I'm talking about is talking about a little bit more complex interfaces where you have to learn, kind of expert interfaces. I guess when I was thinking about this I was troubled by exactly that point. It's kind of like, does it make sense to change someone's interface when they've gotten use to it? And the thing I came up with was basically if you can do it in such a way that it's more useful, then that makes sense. If it's more valuable for the user to be able to basically find more of the stuff they're interested in, then that moment of hesitation they go oh, it's moved, it's worth it. But it's basically, it's a proposition we had to work out the value over that usability principle. So usability is only one portion of what you're going for. There's also you have to work out useable, value, a whole bunch of different stuff.
Question: I guess the listened to the last round of conversation quite interestingly. I guess my question, while you're back and forth, is why not ask the user while they're in their session did you like your experience this time? Save it for next. Then the real question I wanted to ask at the beginning is you sort of danced around the cookie issue. Does the underlying technology really have to change in order for us to really see this eventualisation? Because if I'm on my browser tonight and my son's on the browser in the morning, you know, the computer doesn't know if my son got on there, right? So we've got to rethink us plugging in as humans to the internet knowing that you're interacting with you versus me or my son.
Answer: Sure. Actually, one of the things is that in any shared computer environment, if you don't already deal with the idea of preferences and signing in and out of stuff, then bad things are bound to happen. So yes, that's a problem. But again it's kind of it's a case of - if, for example, we've got a link system and both you and your son are using it, the link system's going to get confused because you're going for X and he's going for Y. So it's not going to be a perfect system.
Question: Just the practicality is I've got different volumes for me and my son, but the practicability is how many times have I rebuilt my Mozilla browser since *[inaudible] and I had to go through the whole personalisation thing all over again, right? And it's that, does the user really want to be bothered with going through that every time, or do they want it all on a plate? Usually they want it all on a plate.
Answer: Well, this is actually where one of the things which I was working on recently at my last job was the idea of progressively collecting more and more information about a user based on their sign in and the idea was when someone signed in we asked them for the ad slip minimum we needed, like an email address, that kind of stuff. But when they used something which required their address or their mobile phone number, if someone wanted an SMS alert, we added that to the thing. When they did something which required an age thing, we added that to the thing. So that was kind of the part of the same situation which led me to this realm of thinking in the first place, which is where essentially you're starting to try and gather these kind of information systems. Adding a log in system to it is great. I mean, if for example we start personalising a system and you could, as you mentioned earlier on, have something come up on the screen to say hey look, we've you know, we've changed it, is it better, if so give us your email and we'll make it the same way all the time.
Question: My premise is the underlying *[inaudible] my new Windows mobile, right? Every time I reset I've got to go through the whole episode again.
Answer: Well no, that's what I'm saying. There's nothing stopping you from adding a membership or a log in kind of element to it. And in fact that may be the way to go. You mentioned, the first thing you mentioned was could you pop something up on the page saying we've changed your interface, would you like to do it? That's exactly the same time as you might ask someone from an email and then you can actually save the stuff on the server and basically get them to sign in to do all the personalisation stuff or tie the two things together.
Question: We're missing out the step where we're involving people, or the person, in the process of personalisation.
Answer: Yes.
Question: If we keep offering them a question saying at the end of their shopping experience, did you enjoy your experience, we assume that when they say yes they were happy to say yes, but they may be saying yes because it's the only option they have. We're taking away the person from the personalisation and not allowing people to discover and to learn the system and then therefore take an ownership part in that system. In the shopping experience if you want to go and get a jar of peanut butter after a while you'll know in which aisle the peanut butter is and it becomes, you can optimise your shopping experience, you can get a small amount of joy out of knowing exactly where the peanut butter is, you actually make a person part of the personalisation process.
Answer: Yes and to be honest that's a solid argument in fact against this kind of personalisation. Having said that, at the same time you could say that the fact that you don't have to, when you walk into the actual real world supermarket and someone hands you the jar of peanut butter which you wanted and, you know, that's not a bad experience either, is it?
Question: What if they smear it on you?
Answer: That's a different kind of website entirely. Well that's the thing, that's where you're getting to freak me out. But that's kind of what I'm talking about.
Question: I'm just wondering. I like music a lot, I like Fiona Apple, I'm a PC user, I have no interest in Apple, so I think that's great that it would come to me with music with Fiona Apple all the time, but what about when my girlfriend like last night had a problem with her Apple laptop and I can't find anymore results anymore, I've lost discovery of new stuff?
Answer: Yes. I kind of, I was thinking about going into more detail than that, but if I was going to be design that interface it wouldn't be like you couldn't get other results. It'd be like you might have a special section at the top which is your filtered stuff and then, you know, and then beyond that it might be just the regular search results. If I was designing it, it would be so that you can easily get the stuff which supposedly, you know, filtered just for you, but you can also just, you know, scroll down a little bit and you've got the regular search results. I mean ideally, but... cool.
Okay, thank you very much.
Copyright © 2007, Scott Parsons (Speaker)
Contact Russ: memberhelp @ webstandardsgroup.org