Daniella's Customer Brain

OneAI - Pipeline success - weekly meeting

4 min read

OneAI - Pipeline success - weekly meeting

Date: 2025-10-16 | Duration: 48.540000915527344 min | Account: pipeline-success

External attendees: bruna.santos@homecare.com, chris.b@pipelinesuccess.com, david@pipelinesuccess.com, ecorrons@homecare.com, jason.o@pipelinesuccess.com, ncuesta@homecare.com

Summary

The team gathered to discuss key improvements in AI agent scripts, emphasizing the need for strict guardrails to prevent errors during customization, especially for complex agents like Pipeline Success. Jason O’Neill highlighted the importance of maintaining internal variables while allowing cautious adjustments to editable fields. The ongoing challenge of enhancing agents’ human-like conversation was addressed, with user feedback comparing their agents unfavorably to competitors like Marriott. The team is set to begin development on a new inbound customer support agent, with strategies for knowledge base integration and call flow design. Additionally, technical integration for inbound calls and operational improvements were discussed, including adjustments to voicemail logic and plans to disable a problematic phone number to improve call quality. Overall, clear timelines and responsibilities were established for upcoming projects.

Key Points

🛡️ Guardrails for Script Changes: Clear boundaries are set to safely modify AI scripts without breaking functionality. 🗣️ Voice Quality Enhancements: Efforts are underway to improve agent conversation quality to match competitors. 📞 Inbound Agent Development: The team is preparing to develop a new inbound customer support agent with specific goals. 🔗 Technical Integration Plans: Integration of inbound calls and proper ID mapping are on the agenda for smooth rollout. 📅 Timelines Established: Clear timelines and responsibilities are outlined to maintain momentum in projects.

Overview

  • Team established guardrails for modifying AI agent scripts to avoid breaking functionality, especially for complex agents like Pipeline Success, emphasizing the importance of internal variables.
  • Jason O’Neill highlighted the need for careful testing with revision saves and AB testing to safely validate changes in agent scripts before public release.
  • Warm-up period of approximately three calls after script changes is necessary for agents to adapt, advising against early performance judgments.
  • User feedback has shown the current AI voice and phrasing fall short compared to competitors like Marriott, prompting ongoing refinements for improved naturalness in conversation.
  • Voice testing indicated Raquel’s performance surpasses Becca, leading to a decision to phase out Becca’s voice pending further data validation.
  • Development for a new inbound customer support agent is set to begin next week, focusing on effective call transfers and covering 90% of expected inquiries.
  • Task assigned to Daniella and Eran to ensure relevant support documentation is integrated into the agent’s knowledge base for accurate responses.
  • Clarified technical integration details will automate contact ID generation for inbound calls not linked to existing contacts, coordinating with Noel upon his return.
  • Voicemail message logic has been simplified to enhance clarity by removing the contact’s first name, addressing issues of inconsistency in recordings.
  • Team aims for proactive addition of local phone numbers for new client regions to enhance caller engagement and trustworthiness in line with operational improvements.

Action Items

Jason O’Neill Monitor the impact of disabling the monthly budget question today and revert if it breaks systems (25:39) Provide examples of calls where voicemail audio cutoffs or unnatural speech occur for troubleshooting (44:33) Prepare call flow updates and documentation for the inbound agent work starting next week (28:00) Send new client details for adding local phone numbers as needed (43:04)

Amit Ben Explain and assist in defining guardrails and metadata usage for AI agent scripting modifications (11:55) Enable disabling/enabling script sections and show how to test changes safely, including use of AI assistant for explanations and suggestions (16:46) Coordinate ingestion and indexing of caregiver help pages into the AI knowledge source alongside Daniella and Eran for accurate agent responses (32:25) Assist with webhook integration for inbound call contact ID population once Noel is back (36:37) Disable the problematic Homecare Leads phone number showing as spam and coordinate acquisition of new numbers for regional clients (41:50) Remove the automated inclusion of contact first names at the start of voicemails and implement controlled greeting phrasing (43:27)

Daniella Block Verify and ensure all caregiver support pages and documents are properly uploaded and accessible on the shared drive (29:40) Follow up with Noel on Slack message regarding inbound call ID configuration (34:15) Take note and manage the disabling and replacement process for the spam-flagged phone number (42:35)

Chris Branum Continue reviewing calls, especially to optimize the call center agent’s transfer flow and family interaction experience (41:40) Help identify further examples or issues related to voicemail recordings and first name inclusion for Amit’s investigation (43:27)

Graph View