How personal is personalisation?
Wednesday September 11 2013
In line with RetailTechnology.co.uks homepage poll, Bruce D'Ambrosio, the godfather of personalisation, discusses the difference between personalisation and segmentation
Bruce D'Ambrosio, chief scientist at Peerius
, told RetailTechnology.co.uk
how a recent Q&A session led him to ponder the difference between personalisation and segmentation and understanding how this difference can help marketers to create a different experience with average order value (AOV).
D'Ambrosio holds an emeritus faculty position in computer science at Oregon State University, where he focused on probabilistic methods in artificial intelligence, machine learning, and real-time behaviour modelling, and he founded automated personalisation engine CleverSet, before it was acquired by Art Technology Group (ATG), which is now part of Oracle. He also founded Prevision, a technology firm focused on decision analysis and strategy software, which was acquired by Fair Isaac, now FICO.
The personalisation expert explained: “At a recent e-commerce personalisation seminar a delegate asked one of the speakers if what was being described was segmentation or personalisation. Using the well-known technique when the answer is ‘I don’t know,’ the speaker offered to get back to the delegate later.
“This got me thinking. There are many solutions being touted to retailers as ‘personalisation engines’ but how many are actually delivering a cutting-edge service, and how many are just providing a piecemeal, fragmented service using out-dated methods?”
Tracing personalisation development
D'Ambrosio said: “The answer lies, in my opinion, in where a solution provider sits on the e-commerce personalisation timeline; essentially how evolved it is. This requires an understanding of the history of personalisation to date – I believe there are three generations and we are currently in the third.
“The first generation, exemplified by offerings from IBM WebSphere and ATG in the late 1990s, relied on explicit human segmentation. Marketers or merchandisers defined broad classes of visitors (new versus returning, ‘hero’ shoppers, and so on) and conditioned presentation or action based on these classes. However, manually creating segment definitions and determining appropriate actions with respect to them turned out to be quite difficult, even with the later advent of automated A/B testing, and so these methods have largely fallen into disuse.”
In the second generation, made famous by Amazon's “people who bought this also bought these,” D'Ambrosio explained how machine algorithms dynamically computed groups of ‘similar’ people and chose responses for the group. “Clustering algorithms did this statically and offline, but even sophisticated online algorithms like collaborative filtering or hidden Markov models fall into this category – essentially determining responses based on the similarities between the visitor and others,” he continued.
“While these new algorithms eliminated the problem of identifying groups, they were still merchandising to groups, and so had the primary effect of promoting already popular products. They were also largely limited to managing the product catalogue, and so constrained in their ability to address broader marketing objectives.”
Factoring in individual requirements
D'Ambrosio said the current, third generation turns this paradigm on its head, acknowledging the need to find some reference context or group, but then basing response not just on the group, but factoring in how the individual differs from his or her reference group(s).
“These methods are much more effective at merchandising the long tail of the catalogue, giving visitors a truly personalised experience and increasing AOV,” he advised. “In addition, recent application of semantic modelling across digital assets has enabled automated personalisation to address broader marketing goals like brand awareness and loyalty – communicating value throughout the customer lifecycle and not solely at the point of purchase.”
But as to who is doing what – segmentation or personalisation or something in between – D'Ambrosio replied: “Your guess is as good as mine. Indeed, the only way to find out exactly what an e-commerce personalisation provider is offering is to ask. That is, understand the personalisation technology timeline from generation one to three and find out where possible partners fit in. Just as importantly, ensure that any claims they make are supported with sales figures from named clients.”
“The point is,” he concluded, “if retailers don’t ask these questions some will find themselves buying first generation segmentation dressed up as third generation personalisation – a ‘not very’ personal personalisation solution.”
Tagged as: Personalisation | e-commerce | analytics | recommendations | testing | AOV | CRM | WebSphere | ATG | Oracle | Peerius