Extractive vs. Abstractive Summarization: How Does it Work?

Yellow and orange sticky notes on an eraser board to show abstractive summarization.

Concatenation vs. Understanding. Words vs. Concepts. Reductive vs. Illuminative. Extractive vs. Abstractive. We’ll walk you through it.

Extractive summarization involves identifying important sections from text and generating them verbatim which produces a subset of sentences from the original text.

Abstractive summarization uses natural language techniques to interpret and understand the important aspects of a text and generate a more “human” friendly summary. 

To give an analogy, extractive summarization is like a highlighter, while abstractive summarization is like a pen. While each has its strengths and appropriate uses, abstractive often gives better results for conversations where information is convoluted and unstructured.

How We Work 

We get asked all the time how our product works. We’re happy to give the details on our blog, product pages, and through our teams. But there’s one question we’ve only had the chance to answer in person. So it’s time to write the answer down. 

That question is: Which type of summarization does ProNotes use? 


Why? Abstractive summarization leverages contextual learning to generate powerful summaries. These summaries are more human-readable, making them easier for agents to consume. That’s part of our goal with ProNotes: Enable the agent.

Apart from just generating abstractive summaries, the ProNotes model also has an additional layer wherein all of these mini summaries are weighted and only the relevant information is put into the final note. This ensures conflicting or confusing elements are not present in the final output, which further supports downstream workflows for productivity.


Agent: Hi this is Rashida Jones, calling on a monitored or recorded line. Am I speaking to Michael?

Borrower: Yes, this is Michael.

Agent: To ensure I am speaking to the correct Michael, can you please verify the last four digits of your social security number and your date of birth?

Borrower: one two three four and 11 March 1985

Agent: Thank you for your input. This call is from a debt collector. I am calling from Delta collections agency on behalf of Comcast for a balance of two hundred and fifty dollars, how do you intend to pay it?

Borrower: I don’t owe Comcast anything, this is a fraud charge.

Agent: I have notated the account as fraud and put it into dispute. We will send out a dispute letter to your address and ensure you are not called again. Thank you for your time.

Borrower: Thank you

Look at the results!

Resulting Extractive Note: one two three four and 11 March 1985,  don’t owe Comcast anything, this is a fraud charge, I have notated the account as fraud and put it into dispute. We will send out a dispute letter to your address and ensure you are not called again.

Resulting Abstractive Note (ProNotes): QA Disclosure, name, SSN & DOB verified, RPC, MM, DISPUTE - possible fraud, we are to send dispute letter, concluded. 

One of the above is clearly triumphant as a viable contact center solution. Want to learn more about how it works — and why it matters?

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

Akshat Vaidya


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