One Size Does Not Fit All
*Image via CastABiggerNet
Predictive vendors can seduce you with the promise of needing only one model to optimize your demand generation process. What’s not to like? This approach is easy, fast to deploy, and probably cheaper as well.
Right? Wrong! Unfortunately, building a single model is not the best way to get started. You are likely to be disappointed with the experience, and you may draw the conclusion that predictive does not work for your organization, when in fact you’re just not correctly analyzing the data.
An important consideration for multiple models creation is the incoming channel of the lead and whether or not those leads have already engaged with your brand. We call these inbound versus outbound leads. Inbound leads have expressed interest by reaching out to you, downloading a paper on your website, joining a webinar. Outbound leads may not know yet about your brand. The fact is that these leads have different dynamics and are driven by different motivating factors. Therefore, buying propensity models are likely to be different and not transferable.
Quite simply, inbound leads may enter information on a web conversion form that are going to be very predictive:
- Did they enter junk information?
- Did they use a professional email address?
- Which country are they coming from?
- Did they specify a correct job title?
- Are they actively looking for a solution?
- Have they clicked on emails we sent them?
- How many times have they visited our site?
- Have they looked at the pricing page?
Their behavior, and the information they provided, will play a crucial role in distinguishing good inbound leads vs bad inbound leads.
However, for outbound leads this information won’t apply. Outbound leads should be modeled using intent data and company fit, as they do not have any behavioral information to include in a model. In addition, leads coming to your site will have a natural interest in your product and may be different buying personas than leads you are reaching out to and trying to engage for the first time. This self-selected population is the ‘tip of the spear’ in the organization. The model built around these leads will be quite good at distinguishing those just kicking the tires versus those who are serious and ready to buy. For example, the model will naturally know how to avoid .edu email domains, as these leads don’t spend money. In contrast, a legitimate business analyst visiting your site is a good lead doing research and should be called back.
Another use case would be a company that is a known user of your competitor’s products. Such accounts are difficult to penetrate and may not be good targets for outbound leads unless they are showing buying intent. Maybe in the last month you have noticed a surge in activity around your solution on your website or other B2B websites for the same account. Perhaps they are ready to switch vendors and are now a great prospect based on this new, time-sensitive information that includes behavioral data on the web.
At first glance, a vendor might sell you on a one-size fits all approach, as it’s cheaper, faster, and easier to use. But in the end, you won’t see the return on investment that you expect. Details really do matter when extracting meaningful value from predictive modeling.
When in doubt, it’s always better to err towards building multiple predictive models. If someone promises that he can solve all your predictive needs with one magical model, be skeptical. Quality matters. Don’t exchange quick gratification and glib talk for real impact on your business.