About Us Blog Events Contact us

Automated Phone Call Script Assessment.

Contact for Services
Face Recognition


The client wanted an automated system to analyze sales calls to improve sales scripts, resulting in better customer service. So, we worked on creating a system that recorded sales calls and automated the analyzing process to assess sales agents' performance and improve business operations. Compare the standard sales script with the text generated from audio speech files of the sales agent to improve the sales script and analyze the sales calls.

Face Recognition


A leading beauty e-commerce store that offers multiple brands.


Business Requirement

  • Text analysis tool to analyze the text generated from calls with standard calls scripts.
  • Integration of Twilllio to access call recordings, making and receiving calls.
  • Dashboard for visualization of analysis of calls with respect to standard calls scripts.
  • Interface for receiving and making calls.
  • Database to store logs, call recordings, text generated, and analysis .

Project Goals

  • To improve the sales script.
  • AI-assisted Automated system to analyze the sales call.
  • To understand the performance of the sales agent.
  • To enhance sales appointments and other business metrics.

Our Contribution

Our Contribution

Stage 1

Minimum Viable Product

Speech-to-Text Generation
  • Using Open-Source Libraries and Toolbox
  • Using Open-source Transformer based models

Text-to-Text Comparison
  • Using python libraries (Score based & pattern Matching) (2 Days)
  • Text Similarity using open-source Transformer based models

Deployment of the developed models on the cloud
  • Development of front end
  • Development of API
  • Deployment of the model

Testing And Validation
  • Basic Quantitative Analysis (Count of Words, word cloud)
  • Qualitative Analysis (Synonyms, semantic meanings)

Stage 2

Understanding the script & Text Analysis

Analysis of the Text Generated
  • Testing And Validation Of ML Model


Webapp to receive and make calls with a Dialpad and other call options

Webapp to show logs of calls and download recordings & generated texts

Webapp as a dashboard to visualize the analysis of the text analysis and call analysis

Database to store the scripts, & recordings of the calls integrated with the Twillio account

Model for the semantic similarity between text generated and script

Model for word for word match

Models for Phrase-to-Phrase match

Models for sentiment score and analysis

Integration of ML models with the webapp

Deployment on the local server


What Does Our Final Product Do?

What Does Our Final Product Do?
Speech-to-text generation using Twillio API
  • Using login details of the API
Script And Generated Text Comparison
  • Using Open-Source Libraries and Toolbox
  • Using Open-source Transformer based models
Analysis of the Text Generated

Quantitative Analysis of the Text Generated

  • How Many times a word or phrase occurred?
  • Impact of Count of words and phrases on business matrices
  • Business Metrics can be number of appointments made, sales increase, etc.
Qualitative Analysis of the Tools
  • The synonyms used for the word to word
  • The semantic meanings of the words
  • The phrase-to-phrase similarity
  • Sentence-Sentence Similarity
  • Houzoo2Sentiment Analysis
  • Semantic Matching
  • To make sure that the sales agent is following the script
Script Analysis
  • Measuring which script has the best ability to set appointments
  • Which features make a good script?
  • Analyze and understand the client (Who can be a prospective buyer?)
Documentation and Report
  • GitHub Repo
  • Live working Application
  • Dashboard for Analysis
  • Project Reports
  • Detailed Documentation of the Project and all of its parts

The requirements for the project are

  • Standard Sales Scripts in text format
  • Speech Audio Calls of the Sales Agents
  • Twillio API login details
  • Whisper API login details (if paid version)


Challenge 1

Twillio generates call recordings in real-time, so actual calls are required even for development.


We utilized our clients' real-time call recordings data to carry out the development process. We had to set up a system for automating data transfer every day.

Challenge 2

The models can only be trained and tested on actual data, which is not available openly.


As call recordings of individuals are unavailable openly, we had to limit our ML training to data available through the client.

Challenge 3

The standard sales scripts are not fixed and vary with business. These scripts are not available readily.


We had to do extensive research on standard sales scripts, apart from the one the client gave. We extracted data online on standard scripts to create the ideal sales pitch and assessed the call recordings compared to these standards.

The Final Result

A complete system that allows the client to utilize sales call analysis to determine the scope of improvement, along with data visualization to offer a concise and accurate assessment of recorded call analysis.

Is it Something Your Business Needs?

Booking A Free Session Today

Let's Talk