Case study · Splunk AI Chat
While building a new chat experience, I found the problem lived in several places. Fixing all of them made dropped leads nearly disappear.
Splunk.com’s sales chat performance had declined and I was asked to find out why. I conducted an audit that showed the chat was only one symptom: its flows were a maze built to patch holes in an older site navigation, and many visitors wanted product answers, not sales. The harder part was that the fixes lived in other people’s territory: sales intake, how reps used the channel, and the pages feeding visitors in. The real job was deciding how it should all work together, then earning buy-in and sign-off from multiple teams I had no formal authority over.
From a chat fix to changed habits
Found the true problem
My audit showed the chat focused too much on patching holes in the site’s navigation.
I redefined the chat’s goals around visitor intent instead of funnel stage.
Aligned six teams on new goals and experience
I used new AI capabilities of the bot platform to design dynamic answers to user questions.
Sales, sales ops, marketing, content, the bot team, and executive leadership signed off on the new goals and plan.
Pivoted and rewired
When the AI capabilities failed, I pivoted the strategy. A shorter sales intake, designed with the sales team.
Sales intake adopted it; the old steps had been a major drop-off point.
Changed how reps worked
Education for sales on why jumping in hurt attribution.
Reps stopped jumping in, users stopped getting stuck. Dropped leads nearly disappeared and sales said their processes had greatly improved.
I’d planned to use our vendor’s early AI feature to dynamically answer visitor questions in the chat, keeping them engaged and improving lead quality. Unfortunately, it crashed the platform in our environment, so I shelved it and instead created an initiative with department leadership to clarify upstream experiences, such as the product pages. Although it wasn’t the solution I wanted, efficiency and lead quality still improved.
The throughline
Good conversation design is measured by what stops happening: anti-pattern loops, stuck users, dropped leads, reps rescuing leads by hand. I measure what design controls and go back after launch to check it, reading for what changed. Now I agree on the success signal at kickoff and export the baseline before launch and the results after, so the evidence outlives the project.
© 2026 Acorn Steed. All rights reserved.