Join us weekly for Office Hours to discuss all things Thoughtly. RSVP here to be notified about upcoming sessions.

​

Sending Metadata to Calls [00:05:00]

• Discussion on how sending metadata into calls improves flexibility by allowing specific information such as an address or booking details to be referenced during the call.
• Example provided of using a web hook with client details like date of birth and claim numbers. To improve clarity, numbers were converted into a more natural format using an advanced prompt.
• The metadata is passed into the conversation to prevent confusion and make the call more natural and smooth.
• Emphasis on manipulating data from web hooks to provide a rich conversational experience by scraping URLs and generating summaries dynamically.

​

Using Mid Call Actions [00:15:00]

• Clarification on the inconsistency of mid call actions, specifically when scheduling or retrieving data. Performed effectively when using singular actions such as sending a text message.
• Troubleshooting discussed for a scenario where a text message link is sent for booking an appointment. The key is allowing the agent to communicate each step calmly to the caller.
• Recommendation to let the caller confirm receipt of a text message and, if not received, re-confirm their phone number before resending.
• Mid call actions were reported to be intentionally set to focus on completing the action, possibly causing abrupt transitions.

​

Calendar and Scheduling Challenges [00:30:00]

• Discussion of challenges when using versions 1 and 1.8 of Thoughtly for booking applications with different calendar providers.
• Queries on whether to practice with version 1 until version 1.8 is stabilized. V1.8 has known issues despite efforts to streamline scheduling through fewer interactions.
• Announcement of work underway to simplify scheduling into one concise action node to address the complexity and reduce redundant steps in booking workflows.

​

New Feature Discussions: Prompts and Instruction Templates [00:35:00]

• Introduction of using concise prompts instead of extensive script nodes for conversational agents, improving dynamic interaction.
• A feature for instructional improvisation allows agents to understand user sentiment and respond more naturally without always following a strict path.
• Prompts are discussed using instructions within normal brackets and parentheses to guide the conversation seamlessly.
• Debate on the format for instructions, highlighting the importance of clarity for engineering and users.

​

Automation and Workflow Integration [01:00:00]

• Clarifications sought regarding workflow automation, particularly with verifying caller details like addresses for tasks such as sending postcards.
• Insight shared into the use of forms and text message confirmations to verify user-provided information, ensuring high accuracy despite regional accents.
• Exploration of extracting preferred contact numbers to remedy initial communication errors and refining workflow calls with web hooks.

​

Demographics via Geography [01:15:00]

• Tutorial on utilizing a combination of triggers and automation to segment calls based on geographic information derived from area codes.
• Insight into defining strategic modules for demographics, enabling regional identification and tailored engagement.

​

Feature Requests and Roadmap Insights [01:35:00]

• Discussion on quality-of-life improvements sought for caller ID allocation across agents and campaign consistency.
• Offer of additional visibility on past-requested features for transparency and follow-up.
• Ongoing improvements were noted, with assurances of strategic focus despite present constraints on resources impacting deliverable timelines.

​

Conversational Intelligence and Response Generation [02:00:00]

• Review of external product integrations, including using third-party tools to enhance caller experiences through iPhone messaging and brand-integration.
• Suggestions and future plans for applications with high pickup rates due to caller ID display and message color correlation.
• Reflection on AI’s perceived trivial features having significant user perception and efficacy impacts.