There are 4 levels of “personalisation”. The act of “personalising” something is owned by the originator – personalised content isn’t distributed by others. Each level corresponds to an increasing technical capability. Also, note that I’m talking about the concept behind technologies – not how it can be used.
Level 1: Customisation
This is dumb personalisation. It’s not big, and it’s not clever. It’s as simple as inserting a user’s name, location or another small detail into existing copy, images or content. For example, Pizza Express’s emails – whilst very cool – are customised, not personalised.
The content already exists. A detail which relates to an individual is inserted through a merge-field. Remember the spammy “James, You Have Won £1000!” printed letters you used to get in the post? That’s customisation – nothing more than merged content.
Level 2: Crowd Personalisation
The lowest form of personalisation that any business should be seriously considering. “People Who Bought This Also Bought.” “Related Articles.” Effectively, using the wisdom of the crowd to power suggestions – because the crowd can’t be wrong. Majority rules.
This is where it starts to get interesting though. At a crowd-level, personalisation doesn’t make use of more advanced data. It’s still pretty basic. Other people looked at X, then looked at Y. If you look at X, you will also like Y. It’s what we call item to user “collaborative filtering” – in other terms, the relationships between people and content is analysed and then a rule is created.
That’s not to say that crowd personalisation doesn’t react, learn or evolve. But it’s considerably less-smart than the next level.
Level 3: Segmented Personalisation
The segment is a well-known concept, especially in email marketing: a group of people who share similar interests.
In terms of product recommendations, this is apparent in algorithms such as “People Like You Also Bought”. You might share a few details with other people in the crowd – for example, you all bought the same product before, or even just bought something from the same category.
At this level, personalisation gets a little bit more relevant because it has some more data to act on. A marketer may want to ensure that people who spend between £100 and £200 per order are shown more expensive products. An algorithm might decide that, as a group, people who spend less than £10 will never buy products over £10 so don’t show anything.
Whilst better than crowd-based, segmented personalisation doesn’t touch the performance of one-to-one personalisation.
Level 4: One-to-one personalisation
You are a 25 year old female from Hull with a preference for black high-heeled shoes and silver handbags. Someone else is a 45 year old male from Dorset interested in steel-capped hiking boots and 10-man tents.An individual; one person.
One-to-one personalisation involves collecting as much data as possible about the individual, and as much data as possible about their behaviour. If I buy a pair of Adidas trainers, I’m buying a size 11 pair of trainers, made of leather, majority colour: white, minor colour: green.
To use this data effectively, advanced algorithms take any other algorithm (from crowd-based to segment-based) and filter results to ensure they will be relevant. So, best sellers in the trainer category – but make sure they meet at least two of my preferences. If there’s not enough products to fill, then recommendations will get less targeted – but at least start off like you know me.
And one-to-one personalisation can be used as a much smarter form of “collaborative filtering”. Find people *exactly* like me, and show me what else they purchased – there’s a high chance I might be interested in them too. And the same applies as above – start super-targeted, get fuzzier as closely matching content becomes unavailable.
What do you think? Have I got it wrong?