Runa was founded a couple of years ago to solve the problem of very low conversion rates on most e-commerce websites. Companies spend lots of money driving people to their websites but only 2-3% convert to buyers. One of the biggest reasons for this is shopping cart abandonment – people put goods in their cart but then never purchase them. Runa’s premise was that pricing is a powerful lever to address this with a view to “recapturing” shoppers. While retailers do make offers to try and save customers, they typically offer it to everyone. In contrast, Runa targets each customer individually.
Runa’s approach is to take all the data known about the shopper (ad or referral words, search terms, pages viewed this visit, referral site, prior visits etc) and use this to assess the likelihood of the shopper being a buyer. If a shopper indicates they are going to abandon, the Runa system immediately offers a promotion from the merchant’s pool of promotions. This happens within the merchant website, while the shopper is still there and the shopper has the option of accepting the offer or to continue abandoning. Accepted offers are immediately applied to the shopping cart and the shopper is returned there to make the (now discounted) purchase.
Recaptured customers are a direct add to a company’s bottom line. All the costs (advertising, marketing and infrastructure) have been incurred already when the shopper decided to abandon. For example, a 12 week program some months back showed a 31% sales lift. Abandonment rate was 83% – $600,000 in actual sales and $3,000,000 or so in abandoned sales. Runa was able to rescue 10% of these shoppers with an average discount of 7.4% – much lower than typical coupon discounts. Because this additional revenue comes without additional marketing or overhead costs, Runa was able to double the merchant’s profit during the 12 week program.
Runa’s approach does not modify the merchant’s website – they describe it as like adding Google Analytics in that you tag pages and doing a one-time install of a module to integrate with your shopping cart. Offers are managed via a Runa Dashboard web-application and are injected into the user experience directly. Accepted discounts are applied directly to the shopping cart, in real-time, even as the shopper continues to browse the site.
The core analytics uses web data – page views, click-stream, search terms, and other associated data. The data is used to compute an abandonment index for each type of consumer segment. Both pre- and post-site data is used – how a shopper came to the site and what they have done since they started shopping, for instance. This data is added to a Conversion Behavior Dataset, which is now several petabytes in size. Runa stores this data in an HBase/Hadoop file system on the Amazon compute cloud. Runa has developed a number of home-grown modeling routines, built around an understanding of the structure of an e-commerce website, that allow their analytics staff to build and refine analytic models. Runa has experienced modelers look at and create these models, and have developed a domain-specific language (DSL) to make it easy to create these models. The DSL is itself written in Clojure, which is a functional language that runs on the JVM). The DSL is specific to writing models – essentially a Runa-specific rules-based approach for representing their models. This DSL is modeler-friendly but runs directly on the JVM. To manage the scale and real-time performance they have invested in software to break down every activity into lots of small tasks that can run independently and in parallel on the cloud and then be reassembled.
Besides these analytics and deployment technologies, Runa does some champion/challenger testing (to confirm that a new model works as well as expected and to determine which sub-segments of the population it should be used on) and is investing in some game theory work to understand how consumers may behave once they realize that abandoning the cart may result in a discount. Of course this might result in a kind of viral “do this and get a discount” effect but this would be largely positive as it would drive business for their customers.
Runa is also working on predicting abandonment, so that offers could be made preemptively while avoiding making unnecessary offers to those who are going to buy anyway and determining the minimum discount that will push a specific shopper over the edge.
Runa has been growing quickly. Working with merchants and adding more merchants leads also to larger dataset for better analysis, making their customers increasingly sticky. Customers pay for performance – a percent of rescued sales – using credit cards/electronic payments. This model combined with the very low cost of getting started means that retailers have little to lose in giving it a try. Runa is aimed squarely at B2C e-tailers – those focused on selling on the web – rather than manufacturers or multi-channel retailers. Their typical site is one that aggregates multiple manufacturers within a focused category.
An interesting example of beginning with a very focused micro-decision – what do we need to do to stop this person abandoning this cart – and of using a rules-based approach (more or less) to deploy analytic models rapidly and with agility. I would like to see them use non-web data too – I can’t help feeling that customer support and offline data might improve the models – but it is certainly an interesting decisioning application.