Paul Cook is CEO of PositiveFeedback, a firm that provides targeting tools for online retailers and for ad agencies.
Where do you start with targeting?
There are two areas to start with for onsite behavioural targeting: collaborative filtering (‘people who bought X also bought Y’) and promotions. Matching up offers of the week to particular customers allows the problem to be simplified and, if they complement each other well, it can provide great results.
How easy is it to set up, how long does it take and how long before you see results?
First of all the site needs to be tagged in order for the data to be collected. While this is theoretically easy, most companies find getting tags implemented can be quite painfully slow. That’s why we’ve come up with a universal tagging product as part of our solution that allows the data captured in our tags to be used by other third party solutions.
Once the tags are in place and the offers loaded into the system it will probably take a month or so before you start to see significant uplift. It all depends on how much data you have, how good the placements are being optimised and how the offers complement each other.
Who owns the data – you or the client?
95% of the data is owned by the client or agency but we do have a small amount of anonymous data that we share accross all clients. This indicates how responsive people are to promotions and translates into a top level behavioural segmentation of clickers, responders etc. More specific data about interests and product purchases is not shared between clients.
Should targeting be implemented on websites with clear usability issues, or should people get their house in order before approaching firms like yours?
It is important that when an offer is made the consumer sees it so that we can learn from the fact they did not respond. Provided that the offer is on a page that is not too cluttered, targeting can provide benefits.
However, if a site has a fundamental problem that is affecting conversion then a redesign would probably produce better returns. Personalisation is really the next stage after getting the basics right.
Do you find it surprising that more targeting isn’t used in online advertising?
When I first started selling ads, which is how I got into new media, I was staggered that at that point in time, nobody could measure the response to an ad. Over time, we have seen people measure clicks and post-impression and all these sorts of things.
It does seem strange now that people are willing to put up five ads and show them to people on a random basis, even though there is this data available. They may show someone an ad again and again, even though they haven’t responded to it, and they may show someone an ad after they have already responded it. It is quite bizarre that this sort of technology hasn’t been done before but I’m sure everyone will be doing it going forward. Our software will automatically work out that one ad is generating much more interest, and will naturally start serving that ad.
How are ad agencies currently using behavioural targeting, as opposed to publishers?
There are a few things going on in the marketplace. The definition of behavioural targeting is a bit blurred – sometimes you scratch the surface and it is just normal targeting. I have seen targeting by ISP or by IP address counted as behavioural targeting, although I wouldn’t call it that.
What we do, rather than build up a segmentation of people, is essentially optimising placements for advertisers. Most of the solutions out there at the moment are for portals and publishers to help them sell more ads, and to try and get a higher CPM for people that they know are interested in skiing, for example.
What we are about is automatically using that data to change which creative is shown to someone, so that they see something they are more likely to click and respond to. Our solution isn’t about trying to capture interest from the consumer, so we don’t tend to try to share data across sites. We allow companies to optimise their advertising automatically.
Cookie-based profiling is something we would look to add in the future, but just as another thing we could optimise on. But for the time being, we use basic behavioural data, such as which ads they have clicked on in the past.
Can you give a rough guide to the uplift that’s possible?
It’s difficult to predict exactly what kind of result you will get – it depends on a number of factors. We would aim for somewhere in the region of 10% or more – but some clients maybe more, some sites maybe less.
Moving on to your targeting tools for online retailers, could you provide a quick summary of how they work and which techniques are seeing the most interest from retailers?
We allow people to automate the process of using data within their marketing. It collects data about what people are doing on websites – what they are looking at and what they have purchased – and then feeds it into a personalisation algorithm which works out which products they are likely to buy. You can come to us if you want to dip your toe in the water and start using that data to cross-sell and up-sell to customers on your website, and do it on a cost-per-recommendation basis.
Which categories of products see the biggest effect on conversion rates from personalisation, in your experience?
Some work really well and others have no uplift at all. Obviously books and films work well. There’s a website called Movielens that I’ve been a big fan of over the years. It’s a project from one of the big universities in the US [the University of Minnesota] and it recommends films that you might like based on your ratings of other films. It predicts what rating you would give a film.
I think those sorts of purchases are far more simple, so it’s easier to capture the essence of what is driving the customer’s decision. When you buy a car, there are a lot of factors that come into the equation – top speed, build quality, safety, image and so on. It’s very complicated. But when you get a book, the only thing you are worried about is the content – not whether the book will fall apart or after sales warranties.
There are far fewer things to base your decision on, whereas with a car, there are probably hundreds. With films, again, it’s less about lifestyle and more about personal choice. But that’s not to say that personalisation wouldn’t work with cars. The fact that you have bought a certain car in the past will dictate what you buy in the future. It can work in other categories, but it just works particularly well where there is less going on as far as the purchaser is concerned.
Which items work and don’t work with Amazon-style last-minute recommendations at the checkout?
As far as ‘people that bought this also bought’ recommendations go, sites are trying to make a quick buck at the till so lower unit cost items work particularly well.
Books, DVDs and CDs work well for that, but higher ticket items don’t work as well. You’re not going to spontaneously sell a fridge to someone, for example, but that’s not to say that you can’t use personalisation to sell fridges. You can use it to help people identify which fridge they want if they are looking for one.
The idea of what we are providing is we have past purchase data available and when sites need a decision to be taken, the data and the algorithm are there to make it. That data can be used for these types of ‘people that bought this also bought’ recommendations, or for selecting a specific promotion for a purchaser, or for ordering a list of fridges in likely preference.
Do you see privacy becoming more an issue for consumers as adoption of personalisation tools increases?
I consciously use Amazon because it bothers to remember my data and make my browsing experience better as a result. Because we aren’t creating profiles across sites, there are fewer privacy issues around what we do, and ultimately, people can opt out. There is a section on our website where they can opt out.
Ultimately, it is about driving down the cost of marketing, which therefore allows clients to pass that onto the consumer and lower prices. And if you talk to people, they just assume it is happening. When I speak to people outside the industry, they are quite surprised that this isn’t common practice.