by James Lawson.

17 Nov 2011: Bringing customer data derived from web behaviour into a Single Customer View (SCV) database is a small but growing area in database marketing. Behavioural information has huge potential to inform not just web marketing and email targeting, but all other channel marketing too. So where do the challenges lie?

Rich information
“We’re routinely exporting email data to the SCV, but not many people are doing it with web analytics,” says Charles Ping, Client Director at Communisis Data Intelligence. “You need good web analysis software and it’s probably most relevant for clients that need to sell products and services at a certain point in the customer journey.”
Right on the leading edge of online marketing practice, very few companies are exploiting their knowledge of individual online customer behaviour at the moment. Even fewer take that learning out to other channels. But individual browsing, response and purchase data gleaned from web and email can be a treasure trove of behavioural information.
Staying on the website, it can be used to drive personalised content, ads or offers, and can inform email retargeting (abandoned baskets, browse/didn’t buy) or subsequent standalone email campaigns. Anonymous browsers can receive personalised offers based on their previous visits, and their history linked in when they later purchase, register or otherwise make themselves known.
Brought into an SCV, web behavioural data has massive potential application across other channels, from working out the effect of offline loyalty promotions on online browsing to exploring any “halo effect” linkage such as that between the PPC keyword used and the subsequent LTV of the buyer.
Another example – that Ping alludes to above – might be to employ the SCV to examine the points in the purchase cycle across multiple channels including the web, and use that to predict behaviour via a model. This might suggest the most effective mix of time, proposition and channel used for different segments at critical stages of the customer journey.
As for any other feed to an SCV, the more advance planning, the less grief when it comes to importing and using the data. The traditional advice applies: consider in advance the information to be gathered, and so which variables will be needed for analysis and reporting.
The frequency and level of detail at which that data should be captured is also important to make sure it will support the desired analysis. How far back in time analysts want to look will affect how often web log files are deleted and purged, for example.
“You need to consider the relationship between the variables at the outset,” says Antony Allen, Managing Director at Data8. “What data will you need to support future analysis? Without six to nine months of data, trend analysis will not be worthwhile.”
Given the volume of data that even a simple website generates, this planning is crucial to avoid analysis paralysis – or to be left without an essential variable. For example, raw clickstream data relating to an anonymous website visitor is potentially useful, but does it really belong in the SCV? Aggregating this data or holding a six-month history elsewhere and then bringing it into the SCV when the person reveals themselves may be a better option.
“Somewhere you do need to hold all the data,” says Ping. “The key is to get the right view so that the end user isn’t swamped. A marketing analyst will have a different view to a campaign manager, who will have a more ‘rolled-up’ view of the same originating data.”
Items like time of clickthrough, where the person subsequently navigates to and the propositions involved all feed into honing subsequent email targeting as well as informing numerous other activities.
“Machine-generated data is clean and easy to read, so it’s usually fairly easy to work with compared to name and address data which is inevitably dirty,” says Daniel Cross, Director of Strategy at Lateral Group. “Storage is cheap but you don’t need every single page request and it’s easy to be too greedy. You might reduce it to date, products viewed, and entry and exit pages for example.
The web analytics software must up to the challenge of tying together various cookies and linking the data within an individual-level visitor-referenced database. The website must also ááá
ááá be set up properly. For example, making sure that the data generated by page tags has a clear meaning.
“The big challenge is in moving away from aggregate data and identifying and tracking individuals,” says Ben Salmon, Global Solution Owner at PBBI. “With known customers, you can start to build a picture one piece of information at a time.”
Taking action like remarketing or making personalised offers demands an understanding of individual browsing and purchase behaviour – that is simply not possible with aggregated data. And that’s before you consider pulling the data into a single customer view.
Companies like Coremetrics, Cognesia and Omniture have long recognised the value of logging individual-level data for use in multichannel marketing, but the vast majority of web analytics tools report at an aggregated level on movements within websites, most popular search keywords and similar metrics. That’s fine as site optimisation is all most companies want to do, but it does mean that free tools like the widely-employed Google Analytics are useless for individual-level tracking.

Identity related data
Allen highlights the value of identity related data captured via surveys, registration or purchase which can be tied to information on browsing behaviour gathered via web analytics as well as other channel data in the SCV.
“A customer might not buy from the catalogue but then starts looking at the website six months later,” he says. “If you can ID them online, then you can send a reactivation communication. You need cookie and registration data to tie it all together.”
Another key issue is the required speed of integration. If web data must be brought together and used in close to real time, then some form of partial SCV might be employed, one that is later fully reconciled with the reference SCV – rather like the difference between an operational OLTP database and a data warehouse.
“We now have the Customer Interaction Repository (CIR),” says Richard Lees, Managing Director of dbg. “Clients are looking to trigger on events and only then update the SCV, so online activity increasingly has to happen without a closed update cycle.”
Lees gives the example of a client portal hosted by dbg where a login has to be immediately linked to the right set of rules for that individual, so that the CMS can serve the correct personalised content. The CIR provides an intermediate partial integration that is done very very quickly.
If an interaction depends on information from the SCV then it may have to wait for execution or instead rely on the CIR, which in turn may need to be updated by the SCV with other channel data, model scores and so forth. What resides where and how each database reads from or feeds the SCV at whatever interval can make for a complex infrastructure.
“Email can be updated quickly but older POS systems might work on a 24 hour batch turnaround,” notes Lees. “Update frequency is often governed by how often the client can give us updates from their transactional systems. The back end is moving a lot slower than the front end these days.”
Bringing records together requires some sort of match key. The obvious choices are name, address, and email, but each one has definite drawbacks. For a start, name and address are often absent, while email addresses are far more volatile than postal addresses; people have more than one email address and change these far more frequently than they move house.
“Email is not unique unless you can link it to an address or a person,” says Ping. “Yahoo recycles old emails, and people have multiple email addresses too.”
Tim Beadle, Director at Atrium UK, picks “browser fingerprinting” as another way to link visits and visitors. This involves logging the mix of data associated with an individual browser: preference settings, operating system version, add-ins, version number and so on. “It gives you a pretty good fingerprint and an accuracy of 85 to 90% is possible,” he says.
So there are no hard and fast solutions, with the key being used depending entirely on the variables available in each channel. According to Cross, email is fine as are other identifiers like IP and session ID.
“You can use name and address and tie back multiple emails to that name and address or other registration data,” he says. “You can end up with cascading groups of linked information using different linkages for each one, and then link the groups together using yet another key type.”
Now for the bad news. Attractive as individual-level tracking and targeting is, it may not yet make it out of the website and into the SCV: the amended PECR legislation that reached the UK’s statute book in May could stop it dead.
Its main requirement is that site visitors give “informed consent” to the use of any cookie. Some other countries have implemented the original EU Directive such that users’ browser settings control whether or not a site can set a cookie, but the UK has gone for full opt-in.
Current browsers’ inability to reflect consent to the level required by the UK ICO means that websites will have to explicitly warn users via their terms and conditions or a mix of pop-ups, alerts and so on. The ICO whose own website, amusingly enough, does not comply with the legislation, has given UK plc until the end of April next year to act in accordance with the legislation.

Compliance concerns
“Currently no-one is compliant with this law so integrating online data would be illegal,” says Beadle. “The ICO found that 96% of visitors to its own site opted out when given the choice. At the moment, no-one knows what the technical solution will look like. It’s a complete and utter mess.”
Without cookies, it will be very hard to track and link repeat site visits, having to fall back on less reliable methods such as IP tracking or browser fingerprinting. It would be particularly disastrous for ad serving networks that will have to ask visitors to opt in to tracking cookies.
“I know some companies make it part of their registration terms and conditions that they will be allowed to set first-party cookies,” notes Salmon. “But third-party cookie users like the BBC and the ad serving networks are still going to have a big problem.”
If the law stays as it is, it will make using cookies as they are used today almost impossible. If that happens, the future for web analysis and much of ecommerce looks grim. The current UK approach seems to be ‘ignore it altogether until the ICO takes a more pragmatic approach’, or risk seriously damaging how ecommerce works.
“It may be that in six months’ time you may not be able to use clickstream analytics,” warns Beadle. “My money is on the regulators ultimately seeing sense and realising every site in the world will break this law. But it’s likely the ICO will choose a company to take to court first.”