A product team built a prototype for an AI agent to help corporate travel planners negotiate hotel room block rates.
The Product Manager wrote the initial script, but it used massive, text-heavy paragraphs, overexplaining, repeating information, and leaving dead-end conversations. This text created immediate cognitive overload, buried the next steps, and risked frustrating users during a high-stakes financial negotiation.
(Generalized to protect proprietary data.)
I reviewed the prototype script and identified four problems:
The script was designed for a first-time user, providing a massive explanation and instructions upfront.
Messages were long and repetitive.
Conversations dead ended without a clear call to action.
The script had no human handoff path if the user got frustrated or the AI couldn't handle the request.
Pulling from my Conversational Style Guide, I enforced the following guidelines for the conversation:
Limit messages to 3 lines, 140 characters, and a maximum of 3 message blocks before waiting for a response.
End-load messages with a call to action at the end of the text block so users know how to respond.
Route to a human if the AI gets stuck.
I built a script template to bridge the gap between design and development, making it easy for product stakeholders to review and engineers to implement:
Defined interaction types (Say vs. Ask) and bot actions (Continue vs. Jump to) to map conversation paths.
Organized the conversation into separate flows so that multiple user paths were available.
The initial prototype included a dense block of text, over explained the system, and left the user with no clear path forward:
Say: Hi, I'm Aura, I am an AI booking agent representing Grand Horizon Hotels. My goal is to achieve win-win room block rates with you on every event booking - This project specifically is one that your organization is hoping to achieve a mutually beneficial outcome on.
Say: Hi, I'm Aura, and AI booking agent. I can help you book group room blocks.
Ask: Do you want a quick overview of how I can help, or are you ready to start?
Quick responses:
Quick overview
Start
Improved scannability by cutting down walls of text into punchy, 3-line messages that are easy for busy users to read.
Kept users moving by removing dead ends and replacing them with clear, clickable quick replies.
Sped up production by giving engineering and product teams a repeatable template that translates copy straight into code.