by James Lawson, Contributing Editor
Keeping consumers actively engaged with a website in an age of ‘micro moments’ is no small feat but thanks to improved CMS and a well-maintained SCV, marketers are beginning to use browsing behaviour and other relevant data to work towards relevant real-time responses.
To succeed in ecommerce, everyone has to be a direct marketer these days. But unlike the campaigns of yore, there’s no time to ponder which offer to make.
On a website, visitors flit from page to page in a rapid series of interactions that Google recently memorably described as “micro-moments”. If they don’t like what they see, they will be gone in seconds.
The marketing goal here is clear: use browsing behaviour and other customer data to deliver a relevant response in real time. But with only 5% of marketers yet able to recognise triggers and respond appropriately (Forrester), supporting intelligent real-time interactions still eludes most of us.
“Deciding next best action is a simple statement but a massively complex challenge to achieve,” says Rob Mullen, CEO at Smartfocus. “As much as anything, it’s about channel data integration. Done properly, you will need access to purchase history, service records, LTV metrics, time, place and even the weather can be relevant too.”
Starting out, the simplest approach is to target using only data from the current web session. A view of past behaviour is vital to both recognise the important behaviours that should trigger actions and to help decide what those actions should be. One solution is to base treatments on a set of customer personas.
“When someone arrives on a website, you need to look for patterns of clicks that match previous behaviours and pick from ranked lists of offers or combinations of offers,” says Mullen. “Depending on their response to that, you then pick the next action. We use a journey builder and cache over 15,000 different persona types along with offers for products that go together for those particular personas.”
So which web data items help marketers detect these vital moments, interpret them, then decide and act? There are many.
As a starting point, Steve Lowell, Business Consultant at Communisis, suggests, “page categories and products viewed, session length, time on each page category and total time on site.”
Other influential variables include initial search terms entered on the home page or how visitors sort page content. Ranking by price implies price sensitivity – so a discount may well appeal.
What is ‘typical’ engagement?
“Understanding what a typical engagement looks like is the critical first step,” says Chris Mair, Partner at DADI+. “As part of this, understanding the available data points within a platform allows you to craft a map of moments – interaction points where you connect with your customer.”
Referring site is another vital piece of information, particularly for new, anonymous visitors. Does this site deliver visitors that are more likely to buy? If Pinterest referrals tend to spend more, then greeting them with a first order discount offer would be worth testing.
US magazine Martha Stewart Living offers another example. It measured how long visitors to its subscription page were inactive for, with over 15 seconds’ delay triggering a pop-up offer with a deeper discount: conversion rates doubled. Similar “interrupt” techniques work well for actions like site abandonment.
Using data from multiple web sessions is the next step along the personalisation road. Here the analytics tool employed makes a real difference. Whether standalone or delivered as part of the CMS, analytics should work at individual level and make it easy to tie customer sessions together using cookies, tracking pixels or other techniques.
“Previous behaviours give you context for visitors’ current behaviours and you can then use that knowledge to personalise content, even if they haven’t registered yet,” says Katharine Hulls, VP Marketing at Celebrus Technologies. “For example, you could tell them that the product they previously searched for on their last visit is now back in stock.”
Effective tracking also means that anonymous visits can be tied together when a customer registers, buys or otherwise identifies themselves, producing a ready-made customer history.
“When you see the cookie, you know who they are without needing to ask them to log in again,” says Murdo Ross, Head of Solution Design at DST Customer Communications. “You can look up the record on the database, retrieve their segment code and other relevant fields, then use that to inform which content to push.”
Anonymous vs. known targeting
In fact, Mair sees little difference between anonymous and known targeting. “Mapping anonymous and known experiences side by side is critical,” he says. “The only thing that changes is how you engage; for example, in the anonymous space, we cannot use a customer’s first name.”
Soliciting identity data – name, email, phone number, address – also becomes more of a pressing concern at this stage. It lays the groundwork for using offline knowledge to drive personalisation and opens up further channel options for customer contact – typically through email.
“Email address is definitely a starting point to enable next best action on a website,” says Ross. “Making the link between the marketing database and ecommerce is the golden nugget.”
With previous behaviour available, there’s far more information to base the current decision on. For example, a travel site prospect that searches for flights to the USA may have searched for flights to Canada on the same dates during their previous site visit. “That might prompt you to offer flights to different North American destinations where you have excess capacity,” says Hulls.
But knowing that you have excess capacity means taking the next and most challenging step: basing decisions on data from the SCV and other business systems. Ideally deployed via a decision-engine-driven hub that manages multiple channel systems, this opens the door to detailed individual personalisation.
Here, the SCV, the CMS and the decision hub must work in tandem. The CMS knows which product visitors have just browsed while the SCV holds deeper historical data like segmentation codes and product purchases.
The content options and treatments available to the CMS must also themselves be properly structured and tagged to facilitate automated selection. Modern platforms like Idio offer automated tools to help make fragmented sets of content machine-readable in this way.
“Segmentation code, age band, gender and lifestage are the priorities in helping inform product offers,” says Ross. “Timing of the last interaction is also important in working out if they are regular buyers, lapsers or pre-lapsers. If they are new, there should be a welcome programme for them to enter immediately.”
Seasonality can also be very influential and linked to unusual purchase behaviour, especially at Christmas. What did they do this time last year?
“It’s critical to be able to identify ad hoc behaviour – like buying your partner perfume as a present – versus regular behaviour,” says Ross. “Don’t do everything based on the last purchase.”
Basing affinity offers on products that other buyers have purchased in combination – as many online retailers do – can also be fraught with peril. There must be enough historical data to be statistically significant or the result is bizarre targeting that leaves the brand looking simply daft. Again, the SCV is vital here.
At this stage, the web analytics data volumes involved can start to balloon, demanding careful aggregation and coding to keep it manageable and actionable. A host of data and insight results – current behaviour, previous purchase, media and channel preferences, demographics – must be boiled down to a simpler set of data that feeds an algorithm or set of rules that, in turn, lets the CMS or decision hub produce a suggested action.
This was a challenge that Communisis overcame when it built a segmentation to help inform content served to new and returning website visitors based on both their history (where available) and recent online behaviour. The client, a maternity services provider, wanted to segment visitors by their different stages of pregnancy.
Communisis built the merged file for segmentation analysis by bringing together the offline SCV with a web analytics data set showing the different content and product combinations viewed. Email addresses from registered users made the vital link to the SCV, producing a match rate of around 70%.
“Our partner Sub2 Technologies ran aggregations to create derived variables, so rather than a row of data for every page visited, we had counts of the types of pages visited,” says Lowell. “We worked with them to create a multi-level look-up table which categorised each page by its main header and then by its sub-categories of content. That helped make sense of each visit.”
Leading the way
Virgin Atlantic is at the leading edge here, using measures like customer propensity to buy, revenue potential and product availability to drive very accurate real-time web personalisation (see DBM June 2015). Celebrus’s web analytics platform powers that Virgin solution. Hulls notes that, “You risk sub-optimal results without the full picture of the individual but we still see very few companies doing this.”
Building the required technical foundation is the biggest barrier, though conventional management structures also hold back progress. It’s common to see different departments competing with each other for budget and access to customers.
“A big challenge for our clients is that they have channel-specific systems, structures and KPIs,” says Simon Martin, Managing Director for Cross-Channel Marketing at Experian Marketing Services. “Real time personalisation usually requires major changes to the whole data infrastructure.”
“You have to decide which channels have to be real time, then build and update a single person view for those,” he continues. “You also need to be able to connect channel systems to feed back response data in real time to react quickly enough.”
That is not a trivial task. Growing channels like mobile mean goalposts are constantly shifting and location increasingly needs to be part of the mix. In the US, 82% of smartphone users consulted their phones while deciding which product to buy in-store (Google).
“If I know you went into the store, researched online but then didn’t buy,” says Mullen, “I can then make a good offer via email to make sure I don’t lose that sale to a competitor.”
Another big issue is latency: providing the required data within a few milliseconds. Legacy systems struggle here, demanding new middleware to help one system to control the other and vice versa.
“Most data warehouses are just that: big repositories of data that act a large silo,” says Mair. “It takes a long time to query them, which takes you out of the moment with a user, potentially missing the window of engagement by days.”
However, with a good CMS, decent web analytics and links to a well-maintained SCV, it’s not too tough to get started – and the unified data all this work creates has many other valuable applications. Communisis’s client will sell its file as a premium list to partners such as Pampers while the segmentation will also inform Facebook targeting via the social network’s Custom Audience option.
In fact, display ad targeting is behind much of the current push to integrate web data with other sources. Adding SCV data to the audience information on ad tech providers’ DMPs helps sharpen on-site targeting for any third party advertisers and improves the site owner’s own retargeting work elsewhere on the web.
How to build a compliant two-way link between those very different data stores is a current puzzle, with huge potential gains for direct targeting in all channels. But that’s another real-time data integration story altogether.