There are lots of talks about personalization technology, particularly in e-commerce. And, having worked with personalization technology for almost 3 years, certain things have been bugging me. My biggest issue with those articles is that they don’t really know what personalization is – there’s not a single, standard, solid definition of personalization. I think the internet community needs this to move on effectively. Wikipedia states that personalization “involves using technology to accommodate the differences between individuals.” But I think that’s too broad. So I’d like to put forward a definition of my own.
Personalization technology enables the dynamic insertion, customization or suggestion of content in any format that is relevant to the individual user, based on the user’s implicit behaviour and preferences, and explicitly given details.
In a nutshell, I think personalization helps any format present or suggest content that is relevant to the individual at that moment in time, based on implicitly and explicitly given data – dynamically.
Personalization isn’t recommendations
In 1999, Jeff Bezos and his Amazon employees were sitting in Seattle, thinking hard about how to make money, and help customers find books they would be interested in. They started to look at collaborative filtering.
“…collaborative filtering… [is] able to help people to discover exactly what they’re looking for, saves them time and improves their lives… We have 6.2 million customers; we should have 6.2 million stores.”
Collaborative filtering is the technology behind simple recommendation algorithms, which look at the relationships between users (customers) and items (products). Whilst effective in their own way, “People who bought Product X also bought” and “People like you also looked at” are very basic, and don’t get down to an individual level.
And there are lots of companies peddling these simple product recommendations as “personalization”.
As Hank Nothhaft said in a post on Techcrunch,
“These recommendation engines were once ground-breaking, but they have failed to evolve. And more importantly, our expectations as Web consumers have evolved beyond the simple concepts of “users who purchased item X also purchased item Y.” At best, services that claim personalization based upon these aggregate metrics attempt to triangulate an identity for us as individuals based upon the galaxy of other individuals. They try to pin us down into an archetype, into a box of likes and interests, without recognizing that as humans, what we desire, want and need is in constant flux and ever-evolving.“
Recommendations look primarily at the crowd. But personalization, as the name suggests, needs to look at the individual person.
As he says, there are providers out there who claim to offer “personalization” services, but looking at their websites and implementations, it’s clear to see that they’re faking it.
What is Personalization?
“Personalization technology enables the dynamic insertion, customization or suggestion of content”– personalization doesn’t just have to be product recommendations: it can also include inserting any content like images or text (e.g. displaying a golf-orientated banner for a returning golf supplies buyer), or customizing content that is already there (e.g. “Hi Joe, we’ve got some great movie suggestions for you!”).
“…in any format”– it isn’t restricted to the web. It can be implemented for any medium or touch point, such as emails, apps, in store kiosks, etc.
“…that is relevant to the individual user, based on the user’s implicit behaviour and preferences, and explicitly given details”– finally, the most important part. Personalization uses both implicit and explicit information, derived in two ways. Firstly, a visitor might explicitly declare some information, such as their gender or date of birth.
Secondly, their behaviour can be mined and processed to help understand affinities and relationships. A good example of this in action would be on a clothes store. If you haven’t given your gender, but in the last 4 clicks you’ve only looked at men’s clothes, it’s pretty safe to say you’re shopping for some men’s clothes – so don’t recommend women’s clothes for this session. If you’re a regular returning visitor and you have bought 4 blue items and one purple item, you can be profiled as having an affinity to the colour blue – but let’s not forget that you like some purple in your wardrobe too.
And it doesn’t have to stop at simple product attributes like colour, size, gender or similar. It’s all about the data that the personalisation system has available. At Netflix, they have a team of specialists classifying each film with plot details such as “Strong Female Lead” to add more data for real-time and past behaviour profiling. By adding their expert knowledge into the system, Netflix’s personalized recommendation engine becomes smarter as it understands the subtle nuances that users can subconsciously relate to.
Coming next week: Part 2: Curation isn’t personalization.
30 May 2013