Onsite conversion back in fashion

Back in the early days of ecommerce, visitor conversion was a really hot topic, and while most ecommerce teams continue to keep an eye on conversion, it’s not been a hot topic for a while. The beleaguered online marketer has been working hard to keep his or her head above water dealing with the shift of advertising dollars online, the nuances of search optimization, and distracted by A/B content testing, behavioral targeting and a hundred other wizzy approaches.

But in harder economic times, we all turn back to basics. Over the holiday period, while bricks and mortar channels suffered from poor like-for-like sales and discount induced margin erosion, online channels for many businesses showed healthy growth rates. Businesses have responded by focusing investments in the online channel, and shifted marketing investments online. In harder times our marketing dollars have to work harder, reflected in changes in the mix, for example banner advertising dollars shifting to organic and paid search.

As ecommerce executives have re-evaluated the new economic landscape, conversion is now right in focus. Getting prospects to the site is still critical, but if you can change the conversion ratio by only a small amount then the returns can be dramatic.

And let’s face it, onsite conversion is still very poor, averaging 31%, i.e. less than a third of those that start a basket process completes it resulting in a sale (source: http://index.fireclick.com). That means that 69% on average abandon, with as many as 83% abandoning in the Fashion and Apparel sector. There’s clearly lots of scope here to make a significant impact and drive more revenue. So how do you start?

There are many different approaches to optimizing conversion: landing page content, the detail of the promotional offer, pricing and competitive positioning, making sure you’re not promoting products that are out of stock etc. All of these (and many more) are indicative of a myriad of potential causes of abandonment, illustrating the difficulty in identifying which levers you need to pull to make an impact.

Being a data driven guy, I always start with numbers. The old adage applies: If you can’t measure it, you certainly can’t manage it, and there’s no doubt that you need to measure conversion.When it comes to measuring conversion, there are many different flavors, which is one of the reasons that it is difficult to compare conversion metrics between businesses.
What you measure is very important. I presented on this subject in some depth during a SeeWhy online executive briefing ‘Conversion Academy’ which you can see a recording of here http://video.google.com/videoplay?docid=-1942819305996695221&ei=XpCBSfjNFIL8igKDtPSZCw&q=seewhy+conversion+academy

Most companies today in our view do not have a complete view of conversion and abandonment. To explain this in a bit more detail let’s just think about the dimensions involved:

• What are you measuring (how you define conversion)
• How frequently you measure it (continuously, hourly, daily, weekly etc)
• The level at which you take measurements – every product, every page, every transaction step, every traffic source, each individual customer, or at some level of aggregation)
• How you are analyzing it (are you comparing like with like, and using ‘rolling time’ or fixed intervals like today, yesterday)

We’ve found that measuring it at the lowest level of granularity, and in real time is essential in determining exactly where and when there is a problem.

This really sums up the difference between reporting and monitoring. Traditional click analytics will show you, if you choose to look, that conversion was down yesterday but it’s often hard to get a meaningful picture from a single day. The trend is undoubtedly the best indicator, perhaps over a week since a single data point can be misleading.

Monitoring, of course is providing different information in a complimentary way: measurement is done continuously, not in end of day batches, which means that you pick up emerging trends much faster. Using rolling time periods (i.e. the previous 6 hours from any moment in time) gives you a very different picture from a static dashboard updated yesterday. This allows proactive alerts to tell you when there’s a problem, immediately pinpointing the source of the problem.

Real time conversion monitoring also allows you to tune your site much more effectively since you can see the effect of any changes you make almost immediately, enabling you to play the many different levers that affect conversion.

Stream analytics tame the data explosion

It was only when I was in discussion with a customer this week that I realized that most people think that primary benefit of real time Business Intelligence is its ability to analyze data in real time. And of course, they’d be right, but only partially. Another massive advantage is the way that data is analyzed, which is fundamentally different from traditional techniques since data is analyzed as a stream. As a consequence stream analytics enable new classes of analysis applications that were not previously possible. It is particularly relevant where data volumes are high.

Traditional data analysis relies on a person analyzing data in batches. First the data warehouse is updated, then queries can be run, and an analyst (or in some cases a business user) can then begin the search for insight. The search starts at an aggregate level, and usually enables the analyst to drill down when something he notices requires further investigation. This leads to an element of chance that any problem will be spotted. As data volumes increase beyond what can be stored in a spreadsheet, it’s almost guaranteed that significant items will be missed: it’s simply not possible to analyze everything.

Event stream processing does things rather differently.

Firstly each event, or transaction, is analysed at an individual level. This is a systematic approach where every event is evaluated individually, not at an aggregate level. This is particularly relevant to policing of business processes, compliance, data cleansing and security applications where checking every item is important.

Secondly, data is analyzed in a stream, not as a batch. This means that each event is analysed sequentially, one at a time. So as every event is checked, it is compared with previous and historical patterns of events. Detecting significant sequences of events becomes a breeze. This is particularly relevant to CRM analytics scenarios, where you might want to detect churn or cross sell signals based upon how the customer is interacting with you. We’ve also seen that when you can respond to customer events in real time the response rate to a promotion is up to 50% higher than an offer made some time after. Sequences of events are also highly relevant to fraud and data cleansing applications, to name but two more.

Of course in addition to this, the analysis is done automatically, and in real time. No analyst has to notice, technology is doing the heavy lifting for you.

The importance of 'Context'

I just found EDS’s Next Big Thing blog, and specifically a piece by Charlie Bess titled “The New Role For Business Intelligence”

Charlie is spot on with his comments about ‘context’ being really critical in order to deploy BI more broadly. Let me extend the thought a bit further.

Today we rely on the reader of the report / viewer of the dashboard to interpret the data correctly in the context of historical performance, and their knowledge and experience of the business. By sticking to a paradigm which essentially reports on historical performance, we are relying on the human analysis of data.

This is a theme that I’ve developed in by free eBook “In Search of Insight” which can be downloaded at http://www.seewhy.com/ebook

BI companies have struggled to deploy beyond the 5% barrier of potential users, and this reliance on the presentation of historical data is significantly to blame. As soon as you want to deploy more broadly, in particular into operations, then the requirements change:

(1) Latency becomes very important. I’ve lost count the number times that I’ve heard something along the lines of “the reports arrives just too late to be really useful.” In fact analysts Ventana just produced a survey http://www.intelligententerprise.com/showArticle.jhtml?articleID=192300872 which found that “100 percent of respondents who said their alerts did not provide guided analysis also said their alerts were always out of date.”

(2) Operations teams require context. In fact ‘information in the context of the business process’ is critical in an intraday environment. There just isn’t enough time to start rooting around to try and interpret the data manually. So operational Business Intelligence systems need to be able to provide all the information in one place, and in a way that paints a crystal clear illustration of the problem or opportunity at hand. This is the so called ‘actionable insight’ that has been so over hyped and rarely delivered.

Of course in a real time world, then the context can be used by computers to automatically interpret the data for you and provide you with problems or opportunities, rather than presenting data. This is a very exciting area because it enables computers to perform tasks that human analysts are spectacularly bad at: automatically checking and validating every transaction, for example, so that exceptions or errors can be spotted.

This capability opens up completely new avenues for Business Intelligence as a real time process step, and enables closed loop automated actions to be driven off the analysis.

End of drive by shooting: Free software transforms BI

It’s not often that an industry changes irreversibly, but the advent of Open Source Software / Free software (OSS/FS) is having exactly that effect. While many Open Source advocates will not be surprised by the assertion, it has particular relevance to the Business Intelligence industry which has remained largely untouched by this new model.

I was prompted to this conclusion by a discussion with a customer that had downloaded the SeeWhy Community Edition and was using it to build a pilot implementation in conjunction with Open Source Business Intelligence software from Pentaho. [Incidentally I’m really excited about this project since it elegantly demonstrates the fit between BI 2.0 and traditional query based BI.]

Business Intelligence in particular needs incremental and iterative development. Business users are generally very bad at describing what they need, and requirements in BI projects invariably change through the life of the project. Consequently starting small with a series of small incremental steps, responding to evolving requirements, generally works best for BI.
Traditional software licensing models don’t support this mode of development. The vendor wants you to pay upfront for their software, which drives you and them to make the project bigger and therefore riskier.

Here’s why: in order to justify the return on investment needed to make the business case for the software purchase, the project grows in size. The vendor will gladly help you to increase the scope of the project, seeing more commitment and services. You will be looking for a big ROI number to help smooth the purchase through the purchasing process.
OSS/FS changes this fundamentally. You do not need to produce a theoretical ROI. You do not need to build a business case justifying a purchase.

Instead you just do it. Download it. Deploy it small. Evolve it. Get value.

Only when the project is delivering value, might you then consider support or maintenance, or upgrading to a professional version as the project grows in scope. But by this point the project has been proven to succeed, with (hopefully) a measurable ROI.

This path is also fundamentally different because it engages customer and vendor into a symbiotic relationship. The risks are aligned: the vendor wants to get your application into production (as you do) not simply to do a ‘drive by shooting’ where the vendor disappears as soon as the software license is sold. Under the new model, the vendor’s subscription pricing reinforces this: you pay only when the software delivers value, and only as long as it continues to do so into the future.