Our sales funnel is full - how do I prioritize my leads? Who should my sales people reach out to?
Ever asked yourself questions like this? You should be: for many companies, answering these crucial questions can bring 40-60% reduction in costs and 50% increase in leads and appointments.
If you ask yourself these questions, keep reading.
After reading this blog post, you’ll walk away knowing how predictive scoring works and if your business is ready for it. Most importantly, you’ll know how to implement it on your own - no coding or data scientists required!
But before we dive into the realm of opportunities with predictive lead scoring, let’s first understand what lead scoring is.
What is Lead Scoring?
Lead scoring is a process in which marketers identify which leads are the most qualified to be sent to the sales team.
This usually happens when qualified leads outnumber the resources available in sales and marketing.
Typically, marketers collect some qualifying properties on how to explain their entities (leads). Then, they assign a score to all these properties based on how valuable they are for their business.
For example: A marketer may add a score of 5 to a lead that leaves his work email. The assumption is that people who leave their work emails are more interested in your business. However, leaving an email address may not be highly indicative of what their intent is. So, marketers may look into more serious properties and assign higher scores to things like: industry, job title, number of purchases, or frequency of visits.
The problem: This technique will give you a certain power over your data and it can be quite useful for your business. However, this process can be very slow, error prone and may result in poor follow-up with your most valuable customers. In fact, 70% of leads and sales are lost because of this. This technique also scales poorly, particularly when there are diverse & complex types of customers.
Fortunately, machine learning has advanced to the point where there is no need to do this the old way.
According to the Harvard Business Review, sales teams that adopt machine learning see a 50% increase in leads and appointments, and a 40-60% reduction in costs .
Enter, predictive lead scoring.
What is Predictive Lead Scoring?
Predictive lead scoring is a technique that provides a more powerful way to do lead scoring.
It utilizes machine learning models that enable a computer to learn the scoring parameters by itself, rather than waste a marketers time to do it manually. Remember, computers are like one very big and complex calculator, so it is only logical to skip human involvement for a task like this. Let’s utilize that!
For example: The gif below is a very simple explanation on how this works. Mainly, you’ll end up doing these five steps:
1. Upload your historical customer data;
2. The computer learns its patterns and what makes a lead qualified or not
3. You upload new never seen data
4. The computer show you who of the new customers are the most qualified
5. You download these leads and send it to your sales team.
In order to explain this technique even better, we are going to show you a step-by-step example in the next section.
Example: Predicting if a Customer will Upgrade
Let’s say you work on the business development team at ChairDB, a database company which follows a bottoms-up sales model.
Your users can create accounts and use your software for a small monthly subscription fee. However, the majority of the revenue comes from a small number of customers that pay large subscription fees, thanks to their usage volumes and use of premium features.
The challenge is: How do you get more customers to pay large subscription fees?
As a member of the business development team, your role is to identify customers that are currently on a small subscription, but that likely can be upsold to higher tiers.
Traditionally you’ve done it by eye-balling some usage data, or by proactively reaching out to users that could potentially use premium features. You are frustrated because it’s always a pain to identify which customers to reach out to.
You end up reaching out to:
- Many customers, thus annoying some; or
- Few customers, potentially missing customers that could have been upsold.
You always wished you had some way of prioritizing which users to reach out to.
In order to prioritize that, you can use lead scoring. However, you’ve learned that manual lead scoring can take a lot of your time, so you decide to go with predictive lead scoring.
Enhancing lead scoring with the prefix “predictive” means that you won’t be doing the job yourself, but you are going to train a machine learning model to do that for you.
At ChairDB, you have access to all kinds of historical information about your users:
- Product analytics data about their behavior on your platform (e.g., number of queries, amount of data uploaded, number of tables etc.),
- Marketing data (lead source, number of emails opened, webinars attended)
- Firmographic information about their company (size, industry, revenue)
- And of course, whether they upgraded to a higher tier or not
Our problem statement in this exercise is that we’d like to understand who of our customers are going to upgrade to higher tiers.
We may then say that this “Status” column in our dataset describes this target, or the specific outcomes that a customer may make: “Upgraded” or “Did not Upgrade”.
In machine learning, we define this column as the target variable for which we would like to gain a deeper understanding or make predictions of.
In this case, your historical data contains the target variable, or “Status” column, because you’ve been actively writing down notes as upgrades happen.
Let’s call this dataset: “Historical Records”. However, for your new customers this won’t be the case. The new table for them won’t contain this “Status” column, and that is what we’ll need to predict. Let’s call this new dataset “New Records”.
Your “Status” Column describes if a customer upgraded or not.
So, you could throw the “Historical Records” dataset at the computer to build a machine learning model that could classify all of these users based on your target variable, or in two groups “Upgraded” and “Did not Upgrade”.
Then, you can upload your “New Records” dataset and ask the computer to calculate which of those new customers are going to upgrade or not.
This is possible because the computer already learned so many parameters from the historical dataset.
Now, it is very easy for the computer to come up with a “calculator” to output predicted status for your new customers.
Even better: If set up correctly, the model could also output the likelihood that a user will upgrade. In other words, instead of just predicting whether a user will upgrade, your model could tell you the degree of certainty with which a user could be upsold. This will help you prioritize even better and in turn, increase revenue.
Most importantly, you can do all of this as your leads come in, without the need to wait for months to analyze their data.
How to Implement a Predictive Scoring Model?
As you probably know it, building a machine learning model is not that simple. To get started with machine learning you must be familiar with some of the following concepts:
- Linear Algebra
- Programming languages
In this post, we are not going to explain the technical part of setting up a machine learning model.
If you are in need of a solution that can handle this for your business, we might be able to help you.
Our team has built a powerful and easy to use machine learning platform for non-technical professionals.
But before you reach out to us, or start to do this on your own, let’s see if this is a right fit for your business.
Is Your Business Ready?
This technique is a great way to scale your business and analyze your data, especially when you have a lot of leads coming your way.
However, answering the next three questions are very important before you even think of implementing predictive analytics in your workflow.
- Do I have historical data? As with every machine learning use case: You can’t start without having historical data. As we showed before, the computer needs to learn from previous customers and make future predictions based on that. So, the answer to this question should be “Yes”.
- Do I have enough data? To be able to train a good machine learning model your historical data should be from a wide enough time frame to generally capture all kinds of variation you'd expect. If you’ve been collecting customer data for at least three months or have at least 500 rows of data, then you may be able to build a good machine learning model.
Please note: These are general guidelines and may differ between different businesses and datasets.
- Do I have the resources to implement this? Finally, you need to think about what is the best way to implement predictive analytics and what is the investment that you can put into this. Are you going to hire data scientists? Are you going to get a consultant? Or, are you going to use a platform that will automate your whole workflow?
If you do not fit into the criteria above, don’t worry, this blog post may serve as a tool for the future. Oftentimes, we hear companies reach out to us for guidance even when they don’t have data. Feel free to DM me on twitter if that is the case, I would love to help you out.
Finally, if this article kept you reading till the very end, you should definitely subscribe to our newsletter.
We’ll put insightful data stories, interesting visualizations and real-life examples in your inbox twice a month.