January 14, 2022

Live Chat - ATC Hackathon

Live Chat - ATC Hackathon

The problem

WWT has a confusing set of options for handling customer support. Each department uses the tool they like best, and customers are often exposed to WWT silos, with department hand-offs that are not tracked or analyzed, and no option for deflecting cases from being created via automation.


WWT.com Request Analysis

Requests that come in through wwt.com cover a variety of topics, from lab access to HR questions and even ethics violations. 

Tier 1 support within the ATC Business Operations team currently handles approximately 400-500 requests per month. In our analysis, we believe that approximately 80% of these requests can be aided, or completely eliminated by automation via ChatBot. The analysis was based on requests in 'Completed' or 'Not Needed' statuses from 11/22/2021 - 1/5/2022, and from the perspective of the ATC Business Operations team and the matrix that they provided.


Automation : Chat Bot can respond and deliver a canned response and no further action needed.

Automation Reroute : Can define the request by keywords and/or categories, and can potentially be routed to the correct team/queue based on these keywords and/or categories. Further action is most likely required and could need live chat or tickets created within the team this was rerouted too.

Live Chat: Cannot define the request by keywords or categories, and additional information is needed prior to working or sending to different team.

Remain Ticket/Request:  These are the Schedule Lab tickets.  The ATC Business Operations team they would like these requests to remain as requests as these cannot be automated or rerouted.

Do users actually like ChatBots?

  • Compared to 2018, in 2019 twice as many consumers were willing to engage with chatbots because they were "very helpful." - Forbes
  • 74% of users prefer chatbots while looking for answers to simple questions. - PSFK
  • 65% of consumers feel comfortable handling an issue without a human agent. - Adweek

"70% of users would not use a company's bot again after a bad experience." - Salesforce

The Tool Selection

In choosing a tool, we knew that we needed to find one that allowed:

  • Support to be automated and cases to be deflected via a chat bot
  • The chat bot to be easily trained and edited by multiple people, to continue to expand its skills
  • Live agents to easily pick up a support request when the ChatBot could not complete a request
  • Cases to be created and managed by the ChatBot or Live Agent when necessary

We evaluated three tools as part of our project: Salesforce, ServiceNow and LiveChat.

File:ServiceNow logo.svg - Wikipedia



We were able to build a Service Now chat flow relatively quickly, but determined that ServiceNow was not a viable option, since it only allows you to enable Chat within the ServiceNow platform and doesn't allow you to embed its chat into a website or custom application.

File:Salesforce.com logo.svg - Wikipedia






Salesforce Service Cloud allows for customer support cases to be created via email, web or chat, and is powered by a Machine Learning Einstein Bot to handle automated responses and case deflection. Salesforce chat can easily be installed on your website or custom application.

Einstein handles Natural Language Processing to take user input and attempts to determine the user's intent, so that it can match the user with an appropriate response. We ruled out Salesforce as an option when we learned that every time you add a new phrase to train your bot or a new Bot Dialog, it takes approximately 1 hour to recompile the Bot's model in order to test it. 

While the documentation does mention that it may take a "few hours" to rebuild a model, we double checked this by attending a Salesforce "Ask The Expert" session, where it was also confirmed that rebuilding a model can take hours, and their recommendation was to make changes to your bot on a less frequent basis, which is not a recommendation we found especially useful.

LiveChat Brand Guidelines




LiveChat is the current tool being used for live chat support for Labs. LiveChat integrates with many 3rd party tools (including Salesforce), has a stellar User Experience for both customers and agents, and comes bundled with support for ChatBot, a tool that falls under the umbrella of LiveChat products, and allows you to build automated bots quickly and train them in real time. 

ChatBot supports both menu driven and natural language processing style of user input, and has a full development API that allows you to integrate ChatBot with any tool used at WWT (such as for ServiceNow ticket creation, or Salesforce case creation).

One benefit of using LiveChat is that we can allow teams to use the tool of their choice for ticket management, building integrations to make ticket creation simple.

Maybe Slower Is Better?

We didn't want to rule out Salesforce without assuming that their Machine Learning algorithm might be smarter than LiveChat's. We set up an experiment with the same automated and user phrases in both systems to see which would respond correctly.

Neither system was able to infer the correct response, but at least with LiveChat, we can train the system in real time so that it can get the answer right next time around.

Comparison Matrix

Stuff to try with Atom

  • What can you do?
  • What is WWT known for?
  • What's your favorite color?
  • Is WWT hiring?
  • Why not use Salesforce for chat?
  • Report a bug
  • Drop some knowledge