Artilium

Wednesday, October 07, 2009

Predicting Relevance

The ‘holy grail’ for the delivery of services to your mobile would be to anticipate your service requirements in real-time and deliver what you want before you need to ask for it! This concept was encapsulated by the Google service launched on 1st April 2000 (i) which anticipated your search requests! Almost 10 years on we are in reality not too far from that vision.

A cliché used to illustrate the concept of delivering relevant services is the example of walking past Starbucks, your phone beeps and you have been given a voucher for a free cookie if you buy a latte. Obviously the story ends happily with a fix of caffeine and sugar and thoughts of how the phone could be so clever and predict your need for refreshment. While there is some value in this model, the alternative view might be – if I was needing a coffee and had the time I would have gone into Starbucks anyway (without the voucher). If I was on my way somewhere I would be unlikely to stop just because a local voucher arrived, that would not be relevant and could be considered as spam.

The traditional view of location based advertising is therefore a little suspect because just being local, but not being relevant, is nothing more than localised spam. So while location is important it is pointless without being combined with relevance. Now consider that many product and service offers are about predicting who the service is relevant to in advance and offering the service to them or even creating anticipation well in advance of the likely point of sale.  Let’s modify the Starbucks example slightly. Location histories may indicate that you are often in the centre of Brussels between 9.30am and 11am on a Thursday.  An offer might then be sent to you on a Tuesday evening with a proposition of a free cookie at Starbuck’s on a Thursday between 10am and 11am with the additional incentive that a second coffee is half price. You may then think “Hey, how lucky, I will be in Brussels at that time on Thursday so why don’t I arrange to meet my friend in Starbucks for a coffee”.  This offer is more likely to be taken since it is relevant and may fit well with existing plans.

Of course the second scenario may still not be totally relevant and there are likely to much better scenario examples, however, it is possible to monitor which services or offers are accepted, where and when they are accepted and try to learn from the user behaviour.  The user responses are used to train the relevance system based on positives in a reciprocal way to training a spam filter on negatives.

The “pre-destination” concept of predicting a probable future location can be implemented using Artilium’s always-on location technology since location histories can be stored and processed. So if you combine this relevance capability with other customer intelligence parameters considered in previous blogs, such as disturbability and establishing interests, you will see that we are not that far from Google’s original spoof concept on April Fool’s day back in 2000. 

(i) http://www.google.com/mentalplex/MP_faq.html, 06/10/09