RECOMMENDATION ENGINES
JUNE 2018
FROM T&I
MAKERS’ DOZEN
Recommendations engines — or ‘recommenders’ — have been around since the 1990s and are increasingly important for ambitious brands because they keep customers engaged longer.
As such they’re big business. In 2009, Netflix famously paid $1 million to an external team that was able to improve recommendations by just 10%.
And a 2017 report by Engine into the state of the U.K.’s top 25 e-commerce fashion brands, revealed that the strongest sites offering product recommendations saw 140% more page views per visit and 10% higher task completion rates.
Google — founded 20 years ago — is still powered by its PageRank algorithm and is the world’s biggest recommender.
Logo 1998
Logo 2018
Why are they important?
Collaborative filtering
The two most common techniques are collaborative filtering (information collected from users) and content-based (comparing product attributes).
Content-based filtering
buy
similar
buy
buy
similar
recommend
buy
recommend
How do recommenders work?
Around 80% of viewers discover their next Netflix binge through recommendations, as opposed to searching the site themselves.
In December 2017, the service announced customization through offering different promotional images to users based on their specific tastes.
As seen in the examples below, the service will try to make you think a show or movie is just like the other shows and movies you like.
Like Uma Thurman?
Watch Pulp Fiction!
Like John Travolta?
Watch Pulp Fiction!
Who’s doing it?
Google Maps rolled out new location-based recommendations in June 2018.
One new feature called ‘Your Match’ gives a percentage score suggesting how likely you are to enjoy a place based on your preferences, previous choices and ratings.
The idea here is that while aggregate ratings (such as reviews) are often useful, individual tastes often differ from the masses.
‘Your Match’ only works if you enable location history, and it’s currently only available for Android users.
Who’s doing it?
Uber announced in May 2018 that it was trialling new tech to match uberPOOL passengers and reduce the chance of a row between strangers.
Pool journeys are shared by up to three people, cutting 25% off the normal fare, but passengers have complained about sharing rides with troublemakers, and strangers keen to tell their life stories.
A European patent application proposes using machine learning to calculate “safety incident prediction models”.
Sharing isn’t always caring. The rise in gig economy services — such as uberPOOL - means people are coming into close contact with strangers, which doesn’t always result in positive experiences.
Image courtesy of Huffington Post.
Who’s doing it?
In May 2018 eBay announced a new personalized shopping feature available on Android and iOS in the U.S..
‘Interests’ serves up recommendations and uses automation to provide access to eBay’s extensive product inventory.
Users answer questions on their hobbies and interests, and how they would describe their personal style.
Each user’s personal store is then crafted through a combination of algorithms and eBay's massive data set.
Who’s doing it?
NOW TV recommendation revamp
The serivce could position itself as a ‘Brand for people like me’ by providing hyper relevant TV & movie recommendations.
If we built Pixoneye’s SDK into the NOW TV app, we could build anonymous customer profiles by accessing the phone’s video and photo library.
This would allow the app to make highly relevant content recommendations using AI and computer vision.
Images of young children on phone...
Kids’ content recommended
What could we do with it?
E.ON Drive holiday travel helper
An intelligent voice assistant could help recommend the best holiday driving routes for E.ON customers with Electric Vehicles.
A voice service that suggests a scenic and efficient driving route, that also includes E.ON charge-points would be incredibly helpful.
We know this would be desired by customers because voice interactions will account for half of search traffic by 20201, and voice is already being heavily used inside vehicles2.
Amazon’s Echo Show device combines voice and screen
What could we do with it?
TGI Friday ‘Mind-read your perfect order’
What better way to offer a truly personalized service than reading customers minds, and recommending the perfect TGI Fridays meal?!
Willing customers could immerse in a sensory experience that measures biometric reactions to a variety of images and experiences. Think burgers vs ribs; tacos vs nachos; wings vs skins; and whether you’re dining with family or for romance.
Laptop webcams and front-facing smartphone cameras could enable the measurement of heart rate, pulse, EEG, and galvanic skin response.
What could we do with it?
The argument against recommenders is that social media users spend their time in online echo chambers - reinforcing existing biases.
To prevent these narrow worldviews, Vice’s in-house creative agency in Copenhagen created an online tool that connects to your Facebook profile and analyses what you’ve liked, mapping out your political and ideological standpoints.
The tool then suggests pages, people and groups you’re most likely to hate, and encourages you to like those as well. The result of you ‘liking what you hate’ is that the Facebook algorithm is disrupted and your feed becomes more balanced.
Burst your filter bubble
“You give them recommendations. You throw different angles at them where, hopefully, they can get something out of it.”
Mike Butcher
More on the subject
From the Technology &
Innovation Team at PAA