Get to Know CallRail’s CallScore

In this series, our product owners gives you a behind the scenes look at the process behind the product

What does CallScore do, and what gaps does it fill for customers?

We wanted to figure out a way to save customers time and money so they don’t have to listen to every single phone call to determine who they need to follow up with first, and which marketing channels are delivering the highest quality leads. We really wanted to harness emerging technologies and give our customers access to tools that help them drive real ROI. We’re really trying to deliver value to our customers by giving them solutions that are as easy as turning them on.

CallScore works by utilizing a machine learning model that’s focused on transcribing a call and extracting words from the audio. Once a call is completed, our system transcribes it and we pass that transcription along with other metadata we extract from the call to our machine learning model. Within a millisecond, the system analyzes the conversation, scores it and tells the user if the call is a lead.

As a product owner, how were you involved in the development of CallScore?

My role is primarily to facilitate a project from the initial idea phase and development phase, through it being released. For CallScore, I was excited to be able to become an expert on AI and machine learning. I had my hands in everything from leading the team that provided our training data set to ensure we were creating the right model for our customer’s use cases. For example, during the design stage, we worked with our technical content writers to make sure that everything in our designs was aligned with the language we’re using in the app. Then we do a final design review which leads to a development sprint. From this point, my job is to make sure everything is happening on schedule and address questions from development or other teams about things like a timeline, which icons to use, and other details needed to progress.

How was the machine learning model for CallScore created?

In order to make CallScore, we needed a lot of human data, so we have a team evaluating the quality of calls and scoring them so we can use that as training data for a machine learning model. To do that, we built an internal application to load calls into where that team can go through and analyze calls. They are mainly paying close attention to the intent of the call and the outcome. They use this to provide a score for each call and provide a reason why they scored the call that way. We use all that to build a machine learning model that’s really focused on quantifying and qualifying calls. CallScore utilizes multiple tested machine learning algorithms to do this.

How does CallScore determine what a “Lead” is?

To start, we looked at hundreds of thousands of phone calls from 10-15 of our main customer industries. We looked at the demographics of our customers, and what similarities they had in their calls. Our machine model looks at words from a transcription of a call and uses them as a quality indicator. But we also look for detractors from those words, like if a caller says “I’m looking to make an appointment.” The receptionist responds “We’re not taking appointments at this time.” CallScore picks up the word “appointment” but notes that the respondent said “not taking appointments” as a detractor. So that would not be qualified as a lead.

How can utilizing CallScore help marketers refine their content strategy and ad spend?

I think that being able to see where a majority of your leads are coming from and which mediums they’re coming from without having to listen to every single phone call or ask your sales team about calls removes the human aspect and is a little more similar to how a marketer looks at something like Google Ads (AdWords.) You can pull a report and look at which campaigns your leads came from. It allows you to decide where to spend your money based on hard data.

How can someone set up CallScore, and what’s the fee?

You can set up CallScore by accessing the CallScore configuration page through the Intelligence menu in the settings section of your account. It’s as simple as clicking a slider from “off” to “on” and clicking save. You can also customize whether you want to only score first-time callers, calls tagged a certain way in a call flow or based on call’s duration. And best of all it’s free!

How do you plan to keep developing CallScore?

This is one of the features we hope to be constantly evolving in terms of accuracy level. The nature of machine learning is that the accuracy increases over time as more predictions are made, but we definitely are investing time and money into making it more accurate by using new human data sets, new algorithms, and training new models. Being able to increase the accuracy of a computer system means literally training the computer to think of conversations in a more human manner. We’re really evaluating which algorithm or combination of algorithms work best to achieve this and match our customers’ different use cases. We want to constantly ensure that our customers are getting the highest level of accuracy.

**Visit to learn more about how CallScore can save you time and make lead qualification a breeze. **