However, quality and quantity do not often go hand-in-hand, which means trade-offs occur alongside misalignment between teams
Typically, marketing and sales teams rely on marketing software to help bridge the gap.
For example, lead scoring has become a popular method to measure the quality of leads, but is it really helping to generate more leads at a higher quality?
This is a tricky question to answer, because the majority of lead scoring methods (and other marketing software solutions) are driven by human hypotheses and occur after you already have a lead.
The hypothesis approach is problematic because it is subjective, and often riddled with bias.
No matter how good our intentions are, our interpretation of the data is going to be influenced by experience, our role, our goals, the data we have access to or can draw trends from, and more.
The other problem is that most hypothesis-driven approaches only consider a few data points, which doesn’t account for the complexity and nuance of most online behavior.
Let’s look at lead scoring software as an example to demonstrate the challenges with traditional subjective approaches versus machine learning.
Subjective Lead Scoring
You may not think of these methods as subjective, because they are the current go-to for lead scoring. However, human bias is seen throughout each of them. Here’s just a few examples:
Classic Manual Methods (e.g. BANT)
BANT is an acronym for budget, authority, need, and timeline. Originally developed by IBM, BANT was the original “manual” lead qualification method. It acts as a guide or checklist for sales teams to ensure that the prospect is a viable lead. It asks questions such as:
- Budget – Does the prospect have the budget for your product or service?
- Authority – Does this prospect have the authority to make a purchasing decision?
- Need – Is there an actual need for your product or service?
- Timeline – What is the timeline for a decision to move forward?
These all seem like important questions to ask, but the problem is that the answers rely completely on the salesperson and their knowledge of the prospect. In other words, lots of human bias.
It also doesn’t account for nuance and complexity, rather relying on four very broad (although important) data points that don’t tell the whole story.
Finally, it’s not particularly timely - it only becomes useful once the salesperson has found time to answer each question, which means the data could no longer be relevant or worth pursuing, and that makes it hard to predictably scale this method into high volume.
Modern Manual Methods (e.g. Data-Informed)
This is another method where marketing and sales teams build their own model by assigning scores to different data points based on an understanding of past leads. This approach attempts to use historical data to see what good leads have had in common, then assigns weights to each data point.
Typically, a data-informed approach looks at far more data points than a classic manual method such as BANT.
For example, a data-informed method may look at
- Demographic information (e.g. geographic region)
- Firmographic information (e.g. company name)
- Referral information (where the lead came from)
- Engagement information (e.g. email opens or social engagement)
- Online actions (e.g. content downloads or video views)
The marketer or salesperson will usually calculate which of their previous leads have converted, and work backwards to see which steps those prospects took to help assign scores.
Once the data points have been considered, the marketer creates their own points system to score the leads. This pushes responsibility on the domain expert to have an extensive understanding of each individual data point, plus the ability to make adjustments and tweaks as new data comes in. Many companies don’t have the time or resources to build and manage these models.
Once again, this method is still subject to human bias. Who determines what score the prospect gets for opening an email? If that score was created two years ago, is it still just as relevant today with changing market conditions and technology? You can see how things start to fall between the cracks.
Automated Methods (e.g. Lead Scoring Software)
Lead scoring software helps by giving you a pre-established framework by which to configure and automate your lead scoring. In this case, the system already has a sound understanding of what data points typically lead to a conversion, so that takes the manual legwork away from your team.
An automated method will also use additional data points that a manual method doesn’t (or can’t) account for.
For example, an automated method may look at what relevant keywords prospects are searching for in search engines such as Google to determine if that prospect is looking for similar products or services to yours. This is typical of modern ABM software and is often referred to as determining the prospect’s “in-market intent”. From there, the software will automatically assign scores based on conversion patterns. However, this still requires the marketer to input a list of relevant keywords as a starting point - and that list is subject to human bias.
This approach certainly saves time in programming and assigning scores, but it still leans on the marketer to define what a quality lead is through the configuration process (such as the keyword list mentioned above). Once again, human bias is evident.
Additionally, modern automated methods still cannot account for the increasingly complex buyer journey.
If you thought about all of the steps your previous 10 customers took to buy from you, you would likely get 10 very different journeys.
But, automated methods don’t necessarily factor in the wider context and nuance of each journey.
For example, pricing page visits are highly regarded in scoring models under the assumption that prospects visiting the pricing page must be reasonably promising. This is often referred to as “page-based scoring”.
However you might be surprised to know that 64% of the visitors on your site with high buyer intent don’t ever go to your pricing page. When you have a complex sale and ABM program, that % of high intent visitors that don’t go to your pricing page is even higher - it can be as much as 90%. So if your model is based on these “typical” assumptions and you have no way to measure its accuracy, you will quickly run into problems.
You must also ask:
- How are these systems tested for accuracy?
- How long do you use a method before you know if it is accurate or not?
- Who decides how it gets configured?
- Do my colleagues have the same perspective on what a qualified lead is?
- How does the scoring model impact other team’s goals and how are they measured?
As you can see, even the most modern lead scoring methods are not without fault.
Another limitation of the automated software approach is that they are built around “known” companies that are already listed in your CRM or MAP. Therefore, they often require the identification of the prospect, or at least the company in order to score them and be useful to the marketing teams.
That leaves a lot of opportunity on the table in the form of unknown or “anonymous” website visitors.
So to summarize:
- Manual lead scoring is clearly anchored on human bias, is inconsistent, is untimely, is too simplified, and is impossible to scale.
- Automated and data-driven methods provide more consistency and ensure you are applying the same criteria to all leads, but are still subject to human bias as the code was engineered by human interpretations of data.
- Both methods will result in inaccurate leads, which means your team is being hampered by either following up with poor quality leads or missing the high-quality leads that they should be following up with.
- Both methods miss an entire (larger) segment of your opportunity, which is the unknown and anonymous prospects on your website.
Objective Lead Scoring - Machine Learning
Machine learning models simply evaluate the data without bias, find the relevant signals and patterns, then automatically make decisions.
Because they are built on data, machine learning models are also measured on data. Machine learning models measure their prediction accuracy prior to going into use and then continuously afterward, which takes away the hundreds of data points that were previously subject to human bias.
So, machine learning takes the bias out of lead scoring by inputting vast quantities of varying data, then letting the model determine outcomes repeatedly until it can do so with high accuracy. The accuracy of the model is how you know it is removing human bias from the equation.
But, not all machine learning models are the same. Many of the marketing tools today use an algorithm-based approach instead of a machine learning approach, and it’s important to note the difference.
The Difference Between Data-Informed (Algorithms) and Machine Learning
An algorithm is “a step-by-step procedure for solving a problem or accomplishing some end” - similar to the rules you would set for navigating between two points on a map.
Compare this to AI and machine learning, which is “the process by which a computer is able to improve its own performance by analyzing large data sets and creating new rules, and continuously incorporating new data into an existing model.”
So the simple analogy is this. Imagine a map on your phone that was based on simple algorithms.
The algorithm would define the rules that direct you from point A to point B based on some criteria such as shortest distance or shortest time. Any additional factors would have to be specifically programmed in on a case-by-case basis. Without additional rules, the algorithm-based map will be the same every single time for your drive from point A to point B.
Algorithms are generally built from a human hypothesis gleaned from big data trends. Those trends can be clearly seen in the data, but they often miss smaller trends or unthought of variables.
On the other hand, an AI-powered map will learn from GPS data plus much more, showing the routes all people took to get from point A to point B and then considering how long it took and what factors impacted that duration. Therefore, it can learn that the shortest route in miles was not always the fastest route because of speed limits, traffic, construction, one-way streets, school zones, time of day, weather and more. AI can process this huge, complex set of data and understand the thousands of one-off scenarios that could impact your trip - then feedback with repeatable accuracy in real-time.
The same thinking can be applied to marketing and sales software.
For example: website behavioral data is very complex and nuanced. That is why only machine learning can accurately identify your most valuable website visitors in real-time. AI can learn the behavior of visitors based on huge data sets of buyers to identify all of the patterns that result in conversions, then score the visitors’ “buyer intent” based on how their behaviors match the patterns of millions of other buyers.
Now you can take action and engage those buyers (e.g. via chat) without being bogged down with non-buyers.
Most companies don’t have the data that is required to build their own machine learning models for this application (as it requires millions of data points to build an accurate model), but tools like Lift AI are helping companies get the benefit of machine learning without having to train their own models.
Introducing Lift AI Intent Scoring
Lift AI is a machine learning model that can identify the buyer intent of website visitors in real-time with over 85% accuracy. It works on any kind of visitor, including known and anonymous visitors, which means you can unlock huge opportunities that traditional lead scoring tools can’t in the form of unknown visitors while sharpening your priority of known visitors for follow-up.
This is different to most other marketing software, which primarily focus on only the known accounts and visitors.
Aside from the game-changing intelligence, Lift AI works in real-time so you can take action to convert high buyer intent visitors before they leave your site. This is critical, as many other tools surface insights after the visitor has left the site where taking action is less effective or timely.
The Lift AI model was trained on 15 years of data gathered from various industries as well as billions of data points and over 14 million live sales conversations. That’s why it can assign accurate scores and instantly identify visitors with the highest buyer intent. It also automatically identifies and segments medium and low intent visitors.
This allows you to set up an intent-led engagement strategy. For example, if you have a chat tool (e.g. Drift), you can connect visitors with the highest buyer intent directly to your BDRs for conversion before they leave the site. At the same time, medium and low scoring visitors can be greeted with a support, nurture, or escalation chatbot instead.
In fact, Lift AI’s study analyzing over 20 million website visits for Drift’s customers has recently found that Lift AI improved their chat conversion rates by 9 times.
As a result, BDRs are always talking to the highest-quality leads, which lets them significantly improve conversion rates and improve their productivity.
Another use case for Lift AI is to integrate the buyer intent scores with your CRM or marketing automation tool - allowing your BDR team to follow up with high intent visitors that happened to be missed for conversion in chat.
That means Lift AI doesn't replace your existing tools, it just makes them stronger while also opening new opportunities in the form of anonymous visitors.
So, you’ve learned how machine learning can strip the human bias out of marketing software and how it creates a paradigm shift in converting visitors.
The advantages of AI and machine learning are undeniable and will only get better with time. As you can see, AI doesn’t replace your team but rather augments them in a way that they can become more productive and have more insights into their potential customers.