It’s Time to Redefine AI Accuracy for Collections and Loan Servicing

Almost everyday at Prodigal we talk with companies that are looking for AI-powered loan servicing and account receivable management (ARM) solutions. Most of the time, their goal is to use AI to do more with fewer agents, while improving compliance and customer satisfaction.

The accuracy of our AI is one of the most common things that future customers ask us about. AI providers frequently toss out numbers like “75% accurate,” “80% accurate,” and so on. The thing is, the definition of “accuracy” is wildly inconsistent.

We don’t think anyone set out to be misleading, but the reality is, people shopping for AI are mistakenly comparing orantangs to corgis (or apples to oranges, if you prefer). The only fair thing to do is agree on one definition of accuracy that we hold everyone to. We’ll get there, but first, let’s agree on what accuracy isn’t.

Taking Words at Face Value Doesn’t Cut It

Here’s a scenario: It’s mid-July in Arizona–a toasty 92 degrees fahrenheit–and the sun is beating down on your running group as your watch vibrates to confirm you just finished your fifth mile. A friend turns to you, panting, and says, “Jeesh, I’m on fire!” 

Do you panic, call 911, and run for a fire extinguisher? No, that’d be ridiculous. 

Even though you heard every word your friend said (100% accuracy), you’re a human that knows better than to take those words literally. You consider context and sentiment too, so you understand that your friend was just expressing that she felt overheated. 

Put simply, as a human being, you innately understand the meaning and intention behind the words people say. That’s exactly what Prodigal’s AI does too.

Prodigal’s AI Rivals Human Levels of Understanding

Prodigal’s real-time and post-call solutions are powered by a first-of-its-kind AI Intent Engine. While transcribing voice to text, rather than focusing on the literal words someone says, our AI Intent Engine simultaneously considers account data, word choice, and emotional cues–like pitch and toneand evaluates everything in the greater context of the full conversation.

So, if our AI was out with you on that midsummer desert jog, it would be smart enough to offer your friend a bottle of water, instead of spraying her in the face with a chemical fire suppressant. That’s because Prodigal’s AI grasps the meaning and intention behind people’s words–just like humans.

Legacy Definitions of AI Accuracy Fall Short

When some AI providers talk about accuracy, they mean the percentage of time their software can accurately transcribe spoken words to written words. That means, if their software can write down 85% of the words someone said, then (in their eyes) it’s 85% accurate. That sounds pretty high at first, but is it really good enough?

Consider that you’re a QA manager for a loan servicer, and you see a call transcript that says: “I’m just done with everything. I want to do more with my life. And you won’t get my new address. I’m unlisted!” That definitely sounds like someone who doesn’t plan on repaying their loan. The catch is, you don’t know the context or sentiment. Also, that transcript is only about 85% correct.

In this version of the transcript, by changing just three similar sounding words, we get a 100% accurate voice-to-text transcription: “I’m just done with Evanstown. I want to do more with my life. And you won’t guess my new address. I’m enlisted!” Well, this sheds new light on the situation. 

If you also knew that this particular customer had used a cheerful tone during the call, has always talked about joining the military, and consistently paid their loan installments early, you wouldn’t waste a QA manager’s time by asking them to review this conversation. 

It turns out that only correctly hearing 85% of what someone says is actually pretty terrible. 

Prodigal is Redefining AI Accuracy

For Prodigal, voice-to-text transcription is only table stakes–the absolute bare minimum.  When we say “accuracy” we mean something else entirely. So, without further ado…

Prodigal’s Definition of AI accuracy: How often AI can correctly understand the meaning and intention behind the words people say.

More than 90% of the time, Prodigal’s AI can do just that, comprehending the significance of conversations at a level of consistency that beats even the best human agents. 

When we hold other AI providers to Prodigal’s definition of accuracy, it’s challenging to find even one that can deliver more than 50% accuracy. Thinking back to our previous scenario, it’s then up to a coin flip whether their AI will tell you that a borrower is fleeing the country overseas versus serving their country overseas. Absolutely no one should have to settle for that unacceptably low level of AI accuracy. 

Get the Most Value from AI with Prodigal

If you’re interested in learning more about AI that’s actually accurate, please reach out. Our industry and data analytics experts are looking forward to exploring how your priorities match with Prodigal’s AI solutions. Let’s work together to exceed your operational goals. You deserve the best consumer finance AI available.

Request a custom demo

Maximize Revenue And Optimize Operations

call center operations vector

Prodigal’s intelligence powers agent productivity and
boosts profitability of lending operations

Request A Demo
Brian Reed

Posted by

Kevin Mackenzie


Subscribe now!

Thank you! You're subscribed to our newsletter.
Oops! Something went wrong while submitting the form.

Prodigal LinkedIn

Find Us



(650) 802-7795

You Might Also Like