Business problem statement

LivePerson regarded themselves as a Conversation AI giant. It was concerning when early in 2022 they realized 2 major problems-

1. Gartner 2022 did not consider LivePerson for their Conversation AI quadrant.
2. LivePerson Brand partners had no usage data for the bot-building flagship product LivePerson owned. Instead they relied on our CSMs to create bots for them.

Research

Shadow a Customer Success Manager(CSM) building a bot for a brand - “Fly on the wall” research method

• The CSM supported brands, but in the end, they were taking charge of 90% of all the brands for their bots, like HomeDepot, Virgin Media, Delta
• Brands lacked sense of control over their bots and assumed a lengthy process to make updates in the bots
• The CSM were required to read 15+ webpage documents as training to keep up to date with the software

Competitive analysis of what’s working in the bot-building world.

Primary navigation highlighted the phases of the bot’s life-cycle, matching the AI builder’s mental model.

• Build/Design
• Tune
• Deploy/Publish
• Manage

Encompass all the steps/components required in 1 place, so AI builder doesn’t jump around

Use the flow chart approach

Inspiring delight factors

UX Problem statement

Why is this happening ?
”Creating a basic bot involves a complex and time-consuming process of going through multiple apps and webpages, lacking a guided flow and an intuitive UI, which discourages AI builders”

Configuring a basic bot required-

• ~20-30 minutes of going through 8+ processes
• Jumping between 5+ different apps to get to these processes (almost 40+ webpages) <image onkar created>
• There was no guided flow, so there was no way of predicting the next process = high learning curve
• Most companies support hybrid roles (agent manager+AI builder), who were unwilling to invest in such a high learning curve
• Bot-building UI was overloaded high amount of information with an unintuitive architecture, which discouraged the AI Builders from creating the bot

UX Strategy and Planning

Discuss business priorities with the Product Manager - Requirements Gathering

• Discoverability of features - This was one of the major requirements from a business perspective. Today most of the features were hidden, and irrelevant information was distributed unevenly on the UI, making it hard to build the bot. Therefore, we decided to prioritize the discoverability of the existing features.
• Flowchart format was postponed - ALL of our competitors used a flow chart format to create bots. However, the developer estimate for that was about 8months. As this project was chasing a tight deadline for the Release 1, we decided to postpone the flowchart format for a later release.
• Reduce the navigation of 40 pages - Navigating between different sub apps, and pages in the product was one major pain point that was discovered from the “Fly on the wall” research method. We decided to add information present in these pages to the parent app, reducing navigation, and providing a wholistic feel to the bot-building process.
• Bot life-cycle menu - All competitors used a bot life-cycle menu approach for their navigation. The AI builders were familiar with how a bot functions - Build → Deploy → Manage → Analyze. We should adopt this in our design.

First pass - Dashboard approach

A process with steps can be broken down into a series of smaller tasks or steps, each of which can be listed and documented. This documentation can then be used to create a guided flow or a step-by-step guide that outlines each task in the process and the order in which they need to be completed. By following this guided flow, users can have a better understanding of the process and can complete each step more efficiently and accurately. Additionally, having a guided flow can reduce the learning curve and make the process more accessible for new users or those in hybrid roles who may not have extensive experience with bot-building.

1. Bot life-cycle menu - The Page was divided into the 4 steps :
   Build → Deploy → Manage → Analyze.
2. Relevant processes to reduce overload - I divided the 8+ processes into these 4 steps, so that we can reveal relevant processes at relevant times. This would
   1. Reduce the elements on the screen, reducing cognitive overload
   2. Make the user rely on the information hierarchy of the UI to look for the relevant steps
   3. Have the user complete all the processes in a easy and intuitive manner

Drawback -

While this option may have had some benefits, it also had a significant drawbacks -

Reduce the importance of editing in the overall hierarchy
The way it was designed would make editing the bot one of the steps in the process. This would effectively reduce the importance of editing in the overall hierarchy of tasks required to build the bot. Additionally, accessing the editing feature would require one additional click, which could slow down the workflow of AI developers who need to access the editing feature frequently.

An extra click - every time
The drawback was particularly significant for AI developers since they use the "build" feature 90% of the time. The extra click required to access the editing feature could slow down their workflow and reduce their efficiency, which could be a significant challenge in a high-pressure work environment.

To overcome this challenge, it became essential to design bot-building tools and platforms that prioritize the most critical tasks - building and editing, which should be easily accessible to users without impeding their workflow. This can be achieved by designing an intuitive and user-friendly interface that allows users to access the editing feature with minimal clicks and effort.

Second pass - Enhance the Bot Builder app

Feedback sessions with the Product manager gave rise to a very important UX Design decision -

“If we plug in the above features of “Bot life-cycle menu” and “Relevant processes to reduce overload” within the Bot Builder app, it would eliminate the drawbacks faced in Pass 1”

I began to massage the UI of the current Bot Builder app to reduce the elements on the screen and re-organize the information architecture to make it more intuitive.

Drawing inspiration from the competitors, I introduced a new tab in the left panel.

1. Categorized Components: This left panel has a categorized list of all the different components that are required during the build stage. It also introduces a location for the processes that the user had to switch out of the app to complete.

2. Bot specific menu: The menu on the top shows the life-cycle of the app, and entails the processes that are required at each stage.

3. Dynamic right panel: The right panel will consistently provide a working area for the narrative or any components selected from the Categorized list. 

Product Manager and I spent multiple review rounds to iron out the design to make sure we cater to all the complex flows within the Bot-builder app today.

A few of the flows -

1. Add a bot user the bot

Before:- AI developer has to switch to another app called “user management” and create a new user profile. Once done, they need to copy the User ID, go back to the Bot builder app, and create a Agent connector, and use the ID to create an Agent Connector for the bot. This will allow the user to deploy the bot.

Enhanced flow :- AI developer clicks “Deploy” on the menu bar, and is led to a wizard for a step by step deployment flow. This allows the user to create a User ID and other processes required for the deployment of the bot.

2. Add training data (Domain) to the bot

Before:- AI developer has to switch to another app called “Intent Manager” and create a new domain. Next add intents to this domain and complete the training of the domain. Next they need to switch back to the bot-builder app and add the intent they created to the narrative. This also causes a confusion between intents and domains for the AI developer.

Enhanced flow :- AI developer switches tab to “Components” and click “Domain” from the categorized list. Once “Domains” are loaded in the right panel, they can choose to create a domain from right there or choose from an existing. The Domains and their intents can be created, edited, trained and managed from within the bot-building app. Also, as a delight factor, the intents can be connected to the narrative from right within this panel too, providing an easy cross link between the narrative and the training data, all without app and context switching.

Impact

We ran 10 user tests with internal team members at LivePerson, mostly consisting of Customer Success Managers (CSM) who are in charge of taking care of bot-building for the Brand Partners.

Additionally we spoke to 2 brands who use Bot-building themselves.

• Task success rate to launch a bot - 95%
  
  
• Time on task to configure a bot - Reduced to 4 min
  
  
• Overall NPS - 90

A CSM spoke about how this would help them debug the bots for brands as well!