Walmart Labs

Bot Spark, a console for training chatbots

Bot Spark: the admin console behind Walmart’s HR chatbot — finding the questions it failed to answer, getting answers from the business, and training them back in.

Case snapshot

Role: Product Designer · End-to-end flows
Scope: Three user types — admin, developer, business
Team: People Systems, Walmart Labs
Timeline: 2019–2021

The situation

A chatbot is only as good as the loop that fixes it.

Bot Spark is the tool for managing chatbots: the place where a chatbot’s admins do their work. A deployed bot fails all the time in small ways, and someone has to notice, find the answer, and feed it back.

The problems it had to solve

Find the gaps
Get the questions the bot could not answer.
Source the answers
Collect answers from the business users who hold them.
Train
Train the bot on unanswered questions and new ones, and feed knowledge into it.
Do not regress
Regression testing for the bot’s existing knowledge.

My scope

Three user types, one console, end to end.

What I owned: High-fidelity end-to-end flows for all three users — admin, developer and business — from research to deployment, and testing the designs against the brand guideline. The logo and identity too.
Why three is the hard part: An admin, an engineer and a business owner want different things from the same knowledge base. The business user has the answer but no interest in the tool; the developer has the tool but not the answer. The console had to make the handoff between them cheap.

How the work got made

Identity first, then wireframes, then the states that get hit.

The work ran from logo design and early exploration through wireframes to the final screens, and the error screens were part of the deliverable rather than an afterthought. For a tool whose whole job is handling the cases where something failed, that is where the emphasis belongs.