Backed by Peter Gabriel, The Filter is a recommendation engine that provides users with suggestions for music, movies and videos.
I’ve been talking to CEO David Maher Roberts about how The Filter has progressed since it launched last year, its recommendation services, and its applications for other businesses…
Firstly, how have things gone since launching?
Things have been good, we launched the website in the summer, and have seen a gradual increase in visits and usage, but the area of the business that has really taken off is the B2B part where we license the technologies under the bonnet to businesses that need to connect content to people in a clever way.
So right now we are doing 2 things: (i) focusing on increasing the number of media companies using The filter on their own site and (ii) continuing to develop the technology and service on TheFilter.com to become an invaluable content filtering service for people.
How does the B2B side work? Do you have any examples of clients?
Any company that needs video or music recommendations on their site or requires personalisation of content to match an end user’s tastes – we can provide the technology to do that.
We power the recommendations in the Nokia Music Store globally When people are on an artist, album or song page, they can get suggestions provided by The Filter’s technology based on the content on that page. Other implementations can be found at We7 or DJDownload.com
We also have a number of trials running with a major music label, a TV network (both in the US) and one of the world’s biggest video aggregator sites.
How does the recommendation process work? How much user input is needed?
We use a number of different types of inputs to create a recommendation model. What people have downloaded and/or played in a session is important. But we also use other pieces of evidence data, like whether people have sent it to a friend, or saved it for later, or added it to a wish list, or commented on it, or rated it.
Basically we have a whole host of pieces of activity (we call this evidence) that we input into a model. We weigh each type of evidence differently, and then work out a probability model using a branch of A.I. called Bayesian maths.
These algorithms work out the strength of connection between every item in your database (artist, album, song, video clip, actor, and producer) which means that you can select any single item in the database, and the database will then reorder itself in decreasing order of probability, so the most likely matches are at the top.
How do you provide video recommendations?
Our work in video recommendations is very exciting. Video is quite
hard to recommend both to and from, especially User Generated Video,
but we have developed a 2 step approach.
Firstly with Tag Relevance, using algorithms to work out the
connection between tags. And then, as the usage data increases for an
individual video, our full Bayesian model kicks in to deliver
recommendations for that video clip (not just for the tags).
Using this process has seen us help our partners significantly
increase click-through rate and a whole bunch of engagement metrics
(dwell time, page views, and visits).
How useful are recommendation technologies in the current economic climate?
Critical. At a time when money is tight, content businesses need to weigh up every penny they spend, and the great thing about a recommendation technology that is effective is that it immediately delivers an uplift.
If you have a business that depends either on selling content or on increasing the amount of time people spend with that content and you can get a system in that increases the number of downloads by 10%, or increases page views by 15%, this immediately translates to more revenues.
As long as the technologies are cost effective (which The Filter is!), then it makes sense to make it a priority to add this to your current offering as a business. Ultimately content businesses have spent the last 10 years digitizing their content, and now businesses need to work out how demand is changing in a world where everything is available.
And working out demand in a digital world, is best done using intelligent technologies that can make the most out of meta-data and usage data and ultimately connect more content to more people who are more likely to enjoy it, and therefore more likely to spend time with it or buy it.
How much work was involved in launching The Filter, in terms of development time, costs etc?
The website was a 6month project; however the core (clever) technology that is under the bonnet and powers our partners is something that we have been working on for 4 years. Our technology has been in commercial environments dealing with huge amounts of data for 4 years
With 19 people in the business (including 14 engineers), then we are talking about 50+ man years of work. In terms of cost – well the business has had 3 rounds of funding which total around £5m
Are you profitable yet? If not, when do you expect to be?
The most recent round of funding was a fairly small round – this is due to the fact that we have decent revenues already, and are heading towards profitability within the next 9-10 months.
Other online music services, Pandora especially, have had problems with paying royalties to record labels? Do you have any issues here?
Well, we are all about connecting users to video or music that they will love and in all cases the music or video is held elsewhere, so we have a deal with Muze to provide us with 30-second clips, and we have a deal with We7 to provide us with full track streaming for our personalised radio service and we then link to the likes of iTunes and Amazon if people want to buy the song. We don’t host any content ourselves, we are purely the technology that finds the content and links it to the right person.
Do you have any plans to offer full songs?
If we did we would do it via an existing service. So in the same way that you can get full songs on our radio service (in the UK only) – but this is powered by We7.
With the information on user preferences etc, do you offer advertising on The Filter, or do you plan to?
Yes we do, but the advertising at the moment is not as much a focus as the content. However we plan to use the same technology eventually to target advertising as well as we do content.
How many users have you had so far?
We have had over 700,000 uniques visit the site since launch, and we currently get around 200,000 unique visitors a month. We have also had over 200,000 downloads of our desktop application (that allows you to build great playlists on WinAmp, WMP and iTunes) which is used to capture consumption habits.