[24]7 ai Case Study

[24]7 ai

  • Problem – Situation
  • Tasks – My Goals
  • Actions
  • Results

About [24]7 ai

“When you’re knee deep in a problem you know nothing about what do you do? Find the smartest guy in the room and make him your best friend.“

[24]7 ai AIVA Model Building Toolkit

[24]7 ai business goal is to build a single experience across multiple channels. The machine learning models integrate with the Human Interaction data in real time. The platform is designed to create intent models based on 40 years of industry specific conversations from their call centers.

I began working at [24]7 ai with little knowledge on how to build an Artificial Intelligence ML Model for a Virtual Agent Bot. I was introduced to the lead engineer on the project. He patiently taught me everything I needed to know about this new toolkit. He walked me through the process by having me create a model from the untagged data.

I relied on my UX process to sort through how to address the task of building a tool that could integrate the various user’s roles and goals. The data scientist would be transitioning over to this tool and move away from their varied time proven and comfortable workarounds. The business requirements and backend design of the tool was in progress. The tagging tool was complete, but the taggers were not consistently utilizing the tool. Adoption would be one big measure of success.


Key Tasks
  • Edit data tags and transformation files
  • Create a model for a particular client, intent or node
  • Edit based on accuracy report results
  • Optimize the agent model’s predictive response to an actual query to match an intent
  • Avoid drop off by the user by sending them to a dead end with a None-None intent
  • Increase accuracy of the AIVA
  • Create a highly accurate model that can be reused by different verticals with little customization across verticals
Pain Points
  • Bad initial tagging of the data creates a lot of back log work for the data scientists tasked with creating the models
  • Lack of an intuitive flow for the WiFi keeps the DSG from engaging in the new paradigm

Design Process

Problem - Situation​

The development team is in the process of building a powerful integrated AI ML toolkit. The data scientists are moving fast and don’t have the bandwidth to learn a new process that is not fully integrated into their process. They needed a designer to build a simple step by step front end for the data scientists and eventually for the client’s IT department to be able to use to build an AIVA Bot or customize the existing models for specific intents.

Tasks - My Goals

• Define user’s goals and stakeholder’s goals
• Take data from core call center company and create a library for AIVA Chat Bots across multiple channels
• Create a valuable tool for the business to expand and simplify the customization process
• Hand the reigns over to the data scientists and clients to build ML Models with less friction
• Create a happy path demo for the sales team to go out into the field with


Developed a tool that was used by the data scientists and integrated into the other tools in toolkit.


• Learn how to build an AIVA Bot ML Model
• Define the Eco-System and interview the users
• Create personas and define key tasks and pain points
• Build a simple single integrated toolkit user flow and UI design for data scientists and customers to; tag data, modify intents, test intents and add data for testing and training a model
• Integrate natural conversation derived from live agents experiences
• Work with the language specialists on how they were utilizing the tool and using work arounds. Create a way for them to not need to leave the tool at any point
• Work in an agile manner with the development team and the design team

Results and Outcome

• I created: user journeys personas, flow charts, brain storming sessions with: stakeholders, developers, product managers and design team
• I created: wireframes with visuals and presentations to the stakeholders for approvals
• Design team together created: visuals, copy, icons, fonts, colors and documentation for the scrum team

Journeys, Flows and Wires

“ I need to run a consistency test with the default config file. Then I need to run an experiment. After that, I may need to get easily back to the data set to retag it or I may need to make edits to the config file.”

Edit a configuration file
Mary opens the “Configure a File” section. She selects a file from the scrolling list on the left. She can now edit the file. She will then save the edited config file with the new project that she has just created.

Next she will test for accuracy. If accuracy test score is not high enough, then she will have the option to go back and edit the configuration file and run the test again. She will repeat this process until the score is sufficient.

When the model comes back with a high enough accuracy score she can select save and the new model will
be finalized. At that point she will have a new model with a new data set.

Start a new model

“ I like the way this is walking me through the process. I want to be able to go to my report and compare my results from the previous test.“

~Data Scientist ​

Business Value

24 7 ai

The new flow that I proposed was reviewed by data taggers and data scientists in depth.
I incorporated the feed back into the wireframes and prototypes for review by the LOB and development.

There was significant back and forth ideation. The new toolkit was needing to also integrate into the new archiving system and Bot Builder.

The scrum team created stories for the back log and defined an MVP happy path for the sales team to take out as a working demo.

The more feature heavy versions were scheduled and the internal team would not adopt the toolkit into the work flow until the full set of features all connected seamlessly.

The business value was dependent upon this tool being adopted and fully connecting all tasks internally. In other words, the moment a task was taken out of the process to finish, the likelihood of the user to return was minimal, meaning the data left the tool for further testing. It is an important project for the company to move forward in creating revenue and speeding up the ML Model process.


This new Tool Kit allowed the company to leverage the 20 plus years of HI data that they have collected.
The data scientists and dev team engaged in the new design as they were a big part of the input, this created “by in” with the necessary players.