This is one of those posts on the run I promised I'd try to do. I'm back in the UK for a bit now. My weekends are a bit unpredictable and will probably involve dashing hither and yon to see family and friends, but happily I'll be in the London offices of The Futures Company on workdays between Wednesday 8th and Friday 17th September, catching up with people and projects before heading out to the NY office.
One of the pleasures of all this travelling is the chance to chat without distractions to a certain other planner for eleven hours or so on long-haul flights. And yes, I'm afraid we do sometimes talk shop, which generates better thoughts and ideas than I suspect either of us would have alone.
Probably our best conversation along these lines was about the 'death of the web' idea. In particular, about the tension between still-growing web traffic and the web's loss of share of internet use. This led to a good dissection of what counted as 'web' and what didn't, with both of us querying the assertion/assumption made in the infamous Wired piece that content served in semantically-rich XML within subscriber-only platforms doesn't count as part of the web. Restricting the web to what is searchable seems rather forced, and only really serves to make sense of Wired's claim that closed networks exist as attempts to take on a Google hegemony. (By this logic, building 'walled gardens' is an act of rebellion, as well as control.)
For my money, if it has hyperlinks, is accessible over an internet connection within a browser and is reasonably device-independent (loose definition, I know), it's the web. So I see Facebook and post-paywall Times articles as absolutely a part of the web, but not Kindle eBooks or iPhone apps - though these often mirror the web or share content with it more than a little. (I'd be happy to have my definitions pushed a bit here if anyone fancies an argument.)
The 'web' vs 'rest of internet' debate is interesting but a distraction, we decided, somewhere over the Atlantic. More meaningful is the whole issue of how we discover content online and what signposts we use to get to it. By that token, search is growing, but so too are non-search-based discovery methods, especially recommendation.
Thinking about our own browsing habits and how they've developed over the last few years, we realised we both rely more and more on what we called 'nodes' to direct us to content for entertainment. We follow links from people on Facebook and Twitter, or within news websites we enjoy, or from blogs we read to ones we might like. This is the 'internet balkanisation' problem in microcosm: we like the nodes because they help us find content that resonates with us; and the more of that content we take in, the more we agree with (and like, and trust) the nodes.
On the other hand, we're the kind of people Google all day for information at work. Search, not recommendation, is our default whenever we actually need to know something.
This led us to a couple of thoughts about the current rage for social recommendation as a media tool - the deftly-placed 'three of your Facebook friends bought this t-shirt' message. We wondered about the cases where it would work best, and worst. The thought we came down to is that recommendations based on social graphs are likely to be less useful than those based on shared behaviour around well-defined nodes.
Here's a thought experiment. There are tube strikes looming in London at the moment. You normally take the District Line to work, and one day it's announced that the line will be closed. Are you more likely to take advice on an alternative route from your best friend and housemate who takes a completely different route to his office, or from a fellow commuter whom you don't really know very well but who always takes the same tube train as you, or from the underground station manager? I'd pick the manager, because alternative route planning is largely a matter of fact (and of decisions based firmly on facts). This is the equivalent of Googling something you need to know quickly. If that failed, I'd ask the fellow commuter. Only if he were unavailable would I ask my friend, probably with little confidence of getting a really savvy answer.
It seemed to us that friendship-based recommendation is inherently perilous in matters of taste as well as matters of fact. For example, I like reading, and so do some of my friends, but there are very few from whom I'd take book recommendations, because I know that our tastes differ. I probably would take a recommendation from a stranger who I knew had also bought and read a lot of the same books as I had. That's why Amazon's 'People who bought this also bought' feature would win out, for me, over a 'Your friends bought' equivalent.
Which made us wonder what the role for friendship-based recommendations in marketing is. And we thought, maybe it's for going beyond fact and detailed taste into that territory where preferences are expressed seemingly at random, as between otherwise highly commoditised brands. This got us onto a slight diversion about the extent to which people really respond in detail to all the complicated lifestyle/aspiration/etc. cues which advertisers think are built into their brand communications, and whether those cues actually function quite basically, to say to people 'this beer/car/music is/isn't for people like you', and then let word-of-mouth and social network effects between friends and peers do the rest.
Got to run again now, so will leave it at that half-developed thought. More on this later, maybe. Closing questions: how do you work out whether to recommend (and so plan digital media) based on behaviour or the social graph? How do you make use of trust in nodes, like you do trust in friends, so as to give behavioural targeting the same kind of emotionally connective kick as as friendship-based recommendation?
I'll pick this one up later...
# Alex Steer (05/09/2010)