Coming down from an Olympic high after watching the recent Vancouver events, points and scoring were heavily on my mind. How many points can we get, how do we get a high average score, how many points do we need to win, what is the record number of points. With the Oscars was coming up, I was also reminiscing on “Up In the Air” and George Clooney’s quest for 10,000,000 frequent flyer miles.
As lead management/nurturing continues to be one of the most talked about subjects on my clients’ agendas, I started thinking about lead scoring and all of the different approaches.
Using algorithms to develop lead scores feels like the norm right now. Many of the lead management tools like Eloqua, Marketo and Lead Life, as well as some surprising vendors like Web Trends, provide an application to assign a value (or, more endearingly, points) to customer and prospect activities. If you click on an email, you can get points. If you download a white paper, you can get some points. If you come in through a search engine, you can get some points. Certain demographics/firmographics can also produce points. For example, if the size of the organization is large, maybe you get more points.
Typically, these vendors capture the events through web page tagging. When certain actions are taken by the user, the page tagging sends signals to the application that registers the event. Events are associated with actual customers through cookies. Typically, email marketing activities create the original association between cookie and customer information.
WebTrends takes a little different perspective in that it can comb through mounds of clickstream data to search for events. WebTrends takes a more analytical approach and, consequently, provides nice “what-if” modeling to see how different customers/customer segments will score based on different scenarios.
The beauty of algorithmic scoring is its simplicity. Even though many clients can create complex algorithms with many events or different point values, you are still talking about simple addition or weighted averages. It is easy for the organization to comprehend that leads and customers with higher scores should get more expensive follow ups, have a higher chance of conversion, and higher value.
The biggest gripe about algorithmic scoring is how to create your buckets. Typically, the thresholds that activate a different marketing tactic, follow up, or transfer to sales is some sort of bucket. For example, 120-140 points would escalate the lead to inside sales, and more than 140 points would send the lead to the field sales force. But how do you create the buckets? How do you know that 120 is the right number? And, in some cases, how do you know that the purely additive approach is correct? Maybe one action on its own should cause a lead to be sent to sales.
Predictive modeling solves these issues by looking at historical information and providing confidence levels based on who has actually converted in the past and the actions they took to get there. In this case, longitudinal analysis over your sales timeline (for some, this could be years) can show the migration path of how leads start with mild interest and evolve into buyers. I have seen predictive models used as a complete replacement of algorithmic scoring, and I have also seen predictive models shape the thresholds for the different buckets.
The downside of predictive modeling
is that not every organization has the level of expertise to create these models. These specialized skills aren’t cheap, and sometimes the models are viewed as block boxes and, consequently, with much skepticism.
Lately, we have seen great interest in wanting to understand multi-channel scores. Whereas page tagging provides an in-depth look at online behavior, Marketo and Unica can also support sales, telemarketing, customer service, brick and mortar, and other channel transactions. Multi-channel scoring allows organizations to gain a true understanding of how all of the different sales and marketing efforts are contributing to upward and downward migration. It also allows organizations to collect purely off-line customer interactions and scoring models.
If you were like me, you got to a place (a bar) early to get a good seat for the gold medal hockey game between the U.S. and Canada. While we waited for the game to begin, we watched the cross country skiing marathon – 50K. Some athletes were coming in almost 15 minutes after the winner crossed the finish line. Of course, everyone commented that it’s better to have some score than none at all. But not all scores are created equally. Algorithmic scoring is a great way to get going with lead scoring and is an intuitive way for people to understand how customers are moving through their life cycle. Predictive modeling provides that next step that fine tunes your algorithms and provides a richer set of insight.
SOURCE: Sales and Marketing Analytics: Lead Scoring
Recent articles by Larry Goldman