Tl;dr - Predictive lead scoring is a concept with enormous potential. The typical execution, however, relies on data so inadequate that any predictions are nearly worthless. Here's a deep dive into models, data, methods, and the ultimate objective - propensity to buy modeling.
A Fascination with Lead Scoring
Sales and marketing are hard. Everyone wants an edge. Many hope that lead scoring can help.
And companies in the Martech, Salestech, and intent data spaces have responded. Here are some examples of each and how they describe lead scoring:
From Marketo - "Lead scoring is a methodology shared by sales and marketing that ranks leads to determine their sales-readiness. Leads are scored based on the interest they show in your business, place in the buying cycle, and overall ﬁt with your business."
From Bombora - "What is a Company Surge™ Score and what factors are used to calculate it? A Surge Score (aka Topic Surge score, Topic interest score, Composite score) indicates the extent to which a company is surging on a specific topic compared to its historical baseline."
From Salesforce.com - "Lead scoring is an effective model that helps sales and marketing departments identify which prospects are potentially most valuable to the company and its current sales funnel."
MarketingAIInstitute describes 6Sense's model - "6sense offers an AI-powered customer data platform that predicts where account-based marketing leads are in the buying journey. It then allows you to focus on the leads closest to closing...(by capturing) intent signals from every known and anonymous source and connect them all to prospective accounts to provide complete insight into the buying journey and accurately predict when accounts are in market."
From Segment - "lead scoring...to help your sales team focus on the companies most likely to convert into customers for your specific business."
There's a huge variety there. Let's look at what is behind the jargon.
Types of Predictive Lead Scoring
Contact level vs. Account level vs. Contact & Account
Many marketing automation and CRM based lead scoring systems focus exclusively on contact-level first-party data. In other words, pages visited, emails opened and clicked, forms filled, etc. That assumes that a contact is operating on behalf of a company in their efforts, essentially as part of a project initiative. (Hint....often it's a person doing their own research and NOT a company project.)
This simple ability was an enormous advance in the early days (even recent days for companies that are just adopting it) of the web, as cookies helped us to track and understand contact interactions with our content - in the shadows (e.g. not a phone call or meeting.)
But how predictive is it really?
In a world where Challenger research indicates that there are an average of 10.2 buyers on a complex buying team, is this enough? Many deals (some research says the majority) end in no decision because, across those 10 buyers, conflicts, personalities, departmental priorities, and other factors conspire against consensus. And many of those nearly dozen buyers may never engage directly with your content in any event. They may be disconnected from the project, or perhaps engaging with your competitors just as their colleagues are engaging with you.
Of the predictive lead scoring examples above, Marketo and Salesforce are focused just on the contact. In contrast, Bombora and 6Sense are focused on the account. Segment is likely alone, as a CDP (customer data platform), in combining contact AND account level data.
That combination of aggregate account level observations and specific contact level insights is critical to robust predictive lead scoring models that weigh all four factors of account fit, account activity, contact fit, and contact activity.
First party vs. Third party Data Components
There's also a difference in how broadly data is observed.
The common lead scoring models of marketing automation and CRM systems draw exclusively on first-party data - the observations of contacts with your own digital properties. That can include known user interactions (email opens & clicks, page visits, form fills, social interactions, etc.) More sophisticated systems may integrate third-party tracking of anonymous user first-party activity (visits to your site) that is typically collected via IP address resolution.
Building your lead scoring and prediction on only engagements with your site is obviously inadequate. There are 2 billion websites, and even if most are irrelevant for this purpose, you have many competitors, and every industry has review sites, forums and other concentrations of information which buyers use for research.
It's folly to pretend as though those don't matter simply because you can't typically observe them.
That's where third-party purchase intent data comes into play.
In order to really score a lead (beyond a contact, and at the opportunity level) it's important to understand and draw reasonable inferences from the activity of as many buying team members as possible, across as many sources as possible. (Learn more here about how intent data models differ in the scope and number of sources.) That requires buyer intent data.
Aggregated account level insights (the account level reporting common among most third-party intent data providers) help to supplement the much narrower first party data that informs typical predictive lead scoring models.
That's certainly an improvement, because in a world of complex sales, and large buying teams, a simplistic view of a single contact's activity with your site and property is not really predictive of opportunity success. It may help to identify your champion, but it will skew to appear favorable to you because it's inherently limited to the engagements with your content.
But it's still not really enough to call it "predictive" with any degree of confidence. There's still a huge gap!
The Intersection of Third Party Data and Contact Level Observation
Predictive lead scoring models often incorporate some factors for job function and seniority. A mid-level IT role looking at marketing software, for instance, indicates something different than a senior level sales role - and both are different than a senior level marketing role. Similarly, the specific pages each of these visits contribute to the modeling as well. If that IT person is looking at APIs and integrations for the marketing software, it's reasonable to assume that (s)he's part of a team considering implementation. If the senior salesperson repeatedly engages with sales enablement content, you can guess the problem they're focused on.
So it's understood that the specifics of the contact (role, seniority, topics) are important to accurate scoring.
It's odd, therefore, that when we're ready to incorporate third-party intent data, many vendors claim that account level data is all that's required. Of course, that's obviously folly. Realistically it's a reflection of the fact that it's all they can deliver!
But real predictive insight and scoring for complex sales require contact level insight from third party intent data as well. Only with contact level™ intent data insights can the scoring model consider job function, seniority, specific key terms apparently being researched, specific competitors engaged - and from those details draw inferences which include stage in the buying journey, problem to solve, outcome to achieve and more. Further, it's possible to incorporate factors for the number and variety of engagements for each contact.
With those additional, granular insights it becomes possible to draw some conclusions regarding the buying team both in aggregate and individually. It's a hugely powerful sales enablement play, for instance, to help the sales team understand which contacts seem to be more skeptical, as well as which likely buying team roles appear to not be as actively engaged.
It's really pretty silly to pretend that lead scoring based on first party data is more accurate when it incorporates details like the number and topics of page visits, but some aggregate account activity score is adequate from third party data.
The Role of AI in Predictive Lead Scoring
Our biases and assumptions, including our projections of how we buy onto prospects, will inform our predictive lead scoring workflows. There's no way around that.
Over time as we compare scores to outcomes, we can possibly detect skewed factor calculations, but the model still relies on our best guess (hopefully well informed from qualitative and quantitative research.)
That's the promise of AI - to use computing power to observe large volumes of data and to help identify the important factors which we might have missed in building our lead scoring assumptions.
And it's a perfect application for the power of AI. However, the AI algorithm itself is well recognized as a possibly biased "observer" and further, when AI models are built on top of inadequate (e.g. account level only intent data) then there's more marketing buzz than functional insight. Finally, based on the sample size of most models purporting to deliver this type of lead scoring, it also assumes that each industry, product, and service are similar. Are you comfortable with that assumption?
So yes, AI offers the tantalizing potential to inform predictive lead scoring with powerful improvements. But for most companies, it's still vaporware - because it doesn't consider the nuance of the circumstances of their sale and doesn't build upon contact level third party data.
Moving from Predictive Scoring to Robust Propensity to Buy Models
But let's pause for a moment and recognize that predictive lead scoring is really only an intermediate step anyway, on the longer and more aspirational journey toward Propensity to Buy modeling.
What's the difference? As we saw above, generally a predictive lead scoring model is intended to weigh factors to determine where salespeople should focus their efforts and when leads should be handed from marketing to sales.
A propensity to buy model continuously compares multiple factors in previous "closed-won," "closed-lost," and "dead" deals to identify which opportunities likely will close and when. Further it is designed to allocate resources (e.g. marketing ads, enablement content, sales outreach) and optimize messaging (to perfectly suit the buyer's mindset and stage in the buying journey as understood from previous similar situations.)
This extends well beyond predictive lead scoring which strives to make sales more efficient. Propensity to buy modeling will theoretically improve forecasting and win rates. It will shorten sales cycles and improve the efficiency of marketing spend. It will scientifically match marketing and sales tactics to mirror the buyer's condition.
Obviously propensity to buy modeling is still well beyond the reach of even the most sophisticated marketing and sales teams. It requires a very carefully engineered and robust marketing data stack and a powerful tech stack including a CDP and integrated AI. It requires contact level intent data signals, a large volume of buying journeys to analyze, and significant AI resources with properly (unbiased) constructed algorithms.
So while it's not currently feasible for most companies, propensity to buy modeling should be the goal. There are lots of opportunities to improve predictive lead scoring in the short-term using tools like contact level third party intent data, and lessons learned from that process should be captured for incorporation into the propensity to buy modeling project that's around the corner.
Want to improve your predictive lead scoring in the meantime? Let's chat.