Nestoria Interview – Alan Jones, Chief Hindsight Officer Doing Words

One of Nestoria's fair leaders, Ed Freyfogle suggested Alan Jones, CHO of Doing Words  would be an interesting person for the Nestoria Australia blog to be talking to.  Which turned out to be a bit of an understatement, because I found Alan interesting and entertaining.  Hopefully you will too. 

Alan, hard at work

I began by asking Alan to indroduce himself the way he's like to be introduced.  So I'll let him do that himself: 

I was a magazine editor until the Internet came along and I realised publishing websites would one day be bigger than publishing magazines.   So after a number of contract web gigs, I joined Microsoft as a startup team member on an early 'city guide' product called Sidewalk. When I left Microsoft to off the bad karma I'd built up working for their empire, I joined a small, fast-growing startup most of my friends had never heard of, called Yahoo!. I was their second employee in Australia, and when I left in 2002 I was a product director for English language products in the Asia region.
I left to start a family, build a house and 'get the band back together' with my early Yahoo! buddies and do another startup or two. I worked on a Netflix-like DVD rental business, a couple of online music businesses and a mobile social network.
Now I consult to other early-stage online businesses; on product strategy, building a brand and managing relationships with communities of customers.

You've had more experience in search than Doing Words implies – can you give us a quick run down on your search credentials?
Thank you, though I'm not sure I have much in the way of impressive search credentials — I've just been an enthusiastic and prolific user of search technology since about 1994, and in the time-honoured tradition, sometimes if you can't find what you want elsewhere, you and some friends have to roll your sleeves up and build it yourselves.
I like big sets of data and the interesting things you can pull out of them if you know how. I've worked on search products for business listings, automotive, jobs and real estate classifieds, rental DVDs, music and social networks over the years.
I have no coding skill beyond making the screen print my name so what can I contribute? Well, I come at search from a user-centered approach. Advances in observing and monitoring user behaviour notwithstanding, I think it's very hard for most product teams to really put themselves in the mindset of a real consumer when they're working on search. If there's a spectrum of technology between 'utility' at one end and 'art' at the other, it's the products that sit at the 'utility' end of the spectrum that are the hardest for us to improve when it comes to user-centered design because they're used almost unconsciously by customers; certainly not with full attention.
I think most consumers use a search engine the same way they brush their teeth — with 90% of their attention on something else, with impatience, boredom and frustration with the whole category of search. They begin with unreasonable expectations about the quality of the result and with minimal patience for any request to contribute to the input.
As search product designers we find it very hard to really live inside that mindset since our work requires that we have 90% of our attention on the product we're designing. It's really a kind of method acting to get inside the head of a search user and I think I'm one of a number of people who are good at that kind of acting.

A couple of people I've spoken to have suggesed that personalisation is the future of search – with the opportunity for geo-locating you and your query, or using your personal search history to give closer matches. What do you say to this opinion and do you think there are any other ways of personalising search that have yet to be explored?
I think we're moving from an age of desktop and laptop computing to an age where most of us will have a device that is capable of computation in our pocket, but for the most part it will focus on helping us communicate, publish and find objects and content in the real world and in the cloud. Most of those devices will have several ways to geolocate us and orient us against a map. Whereas ten years ago, you'd have to first do all your searching and then print out a list of used cars or houses for auctions to take with you on a Saturday morning, it's going to be normal in another five for even our non-technical friends to do the searching on-the-fly, while out and about.
That leads to several big changes in consumer behaviour and affects how we interact with big datasets quite dramatically. If I don't need to invest in searching and building a shortlist of data before I leave home, I also don't need to set aside half my Saturday to dedicate some time to looking for a new apartment to rent. Most people leave that kind of activity until their lease is up or they're about to be evicted, because losing half your Saturday to rental openings is nobody's idea of a good time.
If your mobile real estate app was always checking where you were, and letting you know when it's found a great apartment that meets your needs and budget in the suburbs you frequent most often, could you spread the search for a better apartment out along the last six months of your lease instead of the last two weeks? Would that allow real estate agents to spread their openings and new listings out across the week instead of jamming them all in like lemmings on the same day of the week?

What do you think about the power of recommendations and search results that reflect the internet behaviour of your friends as well as yourself?
Personal search history is a great way to help consumers build better search queries implicitly rather than the explicit methods such as 'browseable search results' we mostly use today, where we narrow or broaden a search result set by adding or subtracting parameters from a list. An implicit strategy means observing how the consumer behaves when handed off from the search to the search result — does the consumer visit just one result in the set, was that the first, second or seventh result in the result set? Does that equate to a successful search and a good set of learnings we can apply to future searches for that consumer?
I'm a huge fan of recommendation algorithms! My first experience with recommendation was building the online DVD rental business HomeScreen (we sold it and the technology isn't deployed anywhere live anymore, sorry). When you've got 20-30,000 DVD titles in your inventory from the day you launch, you know you can't possibly hope to serve a page of 8-10 DVDs on your homepage and meet everybody's tastes in films. Nor can you expect them to craft their own successful search queries or spend hours trawling through genres and subgenres.
In those days Amazon was getting all the press for serving users product recommendations based on their past purchases. There were two big problems with that approach back then: it was 'explicit' in that it only watched what you purchased, not what you spent a lot of time considering before you made your purchase, and it assumed that a purchase was the same as being happy with the product that was being purchased. If you've ever been unhappy with something you've purchased from Amazon, you'll know that doesn't always work.
DVD rentals are great for explicit recommendation, because everybody's got an inner movie critic, and it's quite easy to get consumers to rate the DVD they've just rented out of five stars, and sometimes even write a brief review. In those days the DVDs all went to and from the consumer in the mail, and because you had to email them to confirm the rented DVD had been returned, you had these great opportunities to ask them for a rating out of five stars.
But we learned after a while that recommendation actually works better if the conscious mind isn't involved when you're observing customer behaviour. Explicit recommendation can lead to some surprising results. For instance, one of my fellow co-founders at HomeScreen constantly complained that he was unhappy with the DVD titles our algorithms were recommending for him. When we looked into it, we found he was rating each DVD out of five stars as required, but instead of rating them according to how much he enjoyed the film, he was rating the them according to how good the film was in a technical sense. We all know of films that we'd consider to be very well made — great script, great actors, great director, etc — but they are films we happen not to enjoy. (For me, almost anything by Woody Allen fits the bill). Nobody wants to sit through technically excellent movies they don't enjoy but if you're rating films that way, that's what the recommendation algorithms will deliver more of.
Implicit recommendation — observing how much time you spend in different genres, whether you favour films featuring a certain director, actor or even soundtrack composer, or even if you hover over a film three or more times without yet clicking the 'add' button — these are much better indicators of what you're likely to be grateful for when it comes to deciding what to watch on Sunday night.
I don't think we've yet scratched the surface of recommendation influenced by your social graph. HomeScreen's algorithms looked for other users who'd given similar ratings to the same films and then looked for films they'd rated highly that you hadn't yet told us you'd seen. You can see that strategy at work everywhere, for example on the iPhone community app Chomp.com, or on Last.fm or Pandora. Unfortunately that strategy generally returns a very small set of results as soon as you get beyond hardcore nerds and into regular consumers, who in most product categories, just don't have a big enough set of products/content you can draw inferences from.
What interests me (and I'd still like to see tested in the wild so let me know if you spot it first) is whether we can draw correlations out of dissimilar data sets and deliver a helpful recommendation result. If you and I are the same sex, in the same age bracket, live in the same neighbourhood of London, and check into many of the same locations on Foursquare, are we likely to share much in common when it comes to, say, music, film, live events or gaming? Or can we work that back the other way and find you a job, a car or a place to live that will make you very happy because we know about the music you listen to, the entertainment venues you frequent and the people you're connected to on Facebook? My gut says yes.

Nestoria is a vertical search engine – what are your thoughts on verticals? What will their place be in a future that many people suggest will be dominated by community based search?
In any product sector there will always be a shifting front line in the tension between vertical and horizontal plays — between the specialist and the generalist — and I don't see that tension going away or either side winning a conclusive victory. Facebook is to 2010 what Yahoo! was to 2000 — a huge audience of incredibly diverse consumers that don't actually have very much in common unless you can address them personally. So far, I'm surprised at how poorly Facebook targets third-party advertising at me, considering how much they ought to be able to infer from my behaviour.  Amazon did better six years ago.
They're probably victim to the same challenges we faced at Yahoo! a decade ago — being the big horizontal player means, as each new vertical category takes off, you either have to keep having to rapidly develop your own vertical in that category or acquire one of the leading players in that space. Either way, you're under tremendous pressure to keep up, fighting on multiple fronts, and inevitably you're short of time and resources to do a great job of development and integration.
While if you're a vertical, you live and breathe just your vertical space. You'd rather work all night as a team shipping another release more than almost anything else in the world other than sex, and clearly, you're not getting any sex any time soon. You know your reason for being, your competition and most importantly, your customers, very intimately indeed. If only you could afford to acquire millions more of them, or to spend the time to learn how to run your business well, to hire great people and maybe even manage some of them well. And the investors would like to see a return. And the corporate development people from the horizontal play keep waving a really big cheque. And they swear you and the team would really like working there...

Can you have a quick look over at Nestoria.co.au and see if you think there are any other search fields we'd benefit from? It's different in each country – for example in Germany they apparently like to search properties by the number of bathrooms, so we added that function at their request...
You mean, like, for free? You know I do this kind of thing for a living, right? ;-)
Oopps – sorry!
OK, I'll give you a couple of simple ones: Australia is the land of the great outdoors, and particularly the great outdoors entertaining, so I think it would help to add 'backyard' to your advanced search. We have some large urban areas but they are typically low-density, so backyards are common. It's also the driest, hottest continent, so 'aircon' or 'airconditioning' is going to be big with us, 'central heating' not so much.
Thanks for those.  You're right – we could definitely do with them.

You've recently launched a pilates based search site called pilatessource.com.au, can you tell us what kind of search functions you've empowered the site with, and if we can have the geo-locations of all the studios so that we can include them on Nestoria for people who really want to live near their pilates studio?
Wow, you can totally have the listings and the geocoding for them all! That's going straight to the RT queue. We're still new and we don't have the advantage of starting with someone else's big database like Nestoria does, but we're growing and we'll keep updating that dataset as we grow.
PilatesSource.com.au has instructor listings and studio listings, and many-to-many relationships between them, so an instructor can be listed as working at one or more studios, and a studio can have one or more instructors listed as working there. You can view a studio listing, see the instructors working there, click on one to see their personal profile, and click say, on one of their professional qualifications to learn about the education they've needed to achieve that qualification. That's by no means rocket-science but we seem to be first out of the gate with that idea — other health directory products seem to come at this from the old Yellow Pages approach of a flat inventory of display ads that have no relationships. Sigh.
Geolocation is critical to Pilates and similar fitness services because where a Pilates studio is can be even more important than how good it is — we all struggle to find enough time to go to a weekly class, much less find enough time to commute across town to get there. So we knew from the beginning that geo would be essential.
Some of our competitors are pre-Google Maps and just break listings down into postal code-based categories by state, region and locality, which is just a terrible user experience — you're always at least three clicks away from any listing close to you. Many Australian postal code areas are more than 100km across, since they're based on population density, not area. If you live and work on one edge of a 100km wide postal code area, you don't want to see a studio listing on the opposite edge 100km away. Instead we display studios closest to you by straight line proximity, using the Google APIs.
We don't force you to remember the postal code of your suburb — you can start typing in a suburb name and we'll autocomplete it, or start typing in a postal code and we'll autocomplete that and the suburb name too. Then we'll remember your recent searches for next time. I'm not happy with the speed we're getting for autocompletion yet but we know what we need to do.

At this point I'd like to thank Alan a huge amount for being so expansive and so interesting that I can't cut his words dn enough to fit them in one post.  He very generously went on answering questions after this, and in the interest of your time, and because I love a good cliffhanger I'm going to leave the rest of his post to be published tomorrow.  Stay tuned, in Part Two Alan talks about social media, whether everything is worthy of having 'friends', some of his favourite apps. And what kind of tool he'd come up with if he was the God of technology.

2 comments

Leave a comment...