Sales Copilot: Gong, Notion and Resend in One Panel

My AEs no longer spend half an hour writing recaps; the copilot delivers human-sounding drafts.

Context

In the fast-paced world of sales, every minute counts. Our account executives (AEs) were spending an inordinate amount of time on post-call administrative tasks, particularly writing detailed recaps and ensuring all follow-up actions were meticulously documented. This often led to "half-written notes"—brief, incomplete summaries that lacked the depth and context necessary for effective deal progression or seamless team collaboration. The consequence was a significant drain on their valuable selling time, a higher risk of missed opportunities, and a general frustration with the administrative burden.

I recognized that the solution wasn't to simply automate note-taking, but to create an intelligent sales copilot that truly understood our sales process, our unique deal stages, and the specific information required to move opportunities forward. My goal was to develop a system that could proactively "fill the gaps" in post-call documentation, transforming raw conversation data into actionable insights and ready-to-send communications. This copilot needed to be more than just a transcription service; it had to be an intelligent assistant that could synthesize information, identify key takeaways, and align with our established sales methodologies. The ultimate aim was to free up our AEs to focus on what they do best: building relationships and closing deals, while ensuring every call ended with clear, comprehensive next steps and a human-sounding draft for follow-up.

Stack I leaned on

• Gong webhooks + secure storage: Gong is our primary conversation intelligence platform, capturing and transcribing all sales calls. We leverage Gong's webhooks to trigger our copilot's workflow immediately after a call concludes. Transcripts and call metadata are then securely stored in an encrypted Supabase database, ensuring compliance and auditability.
• OpenAI GPT-4o mini with versioned prompts: The intelligence behind our copilot is powered by OpenAI's GPT-4o mini. We've developed a library of meticulously crafted and versioned prompts, stored in Git, that guide the LLM to extract key information, summarize calls, identify risks, and suggest next steps, all tailored to our sales methodology.
• Notion API for deals and next steps: Notion serves as our flexible workspace for managing deals and tracking next steps. The copilot uses the Notion API to automatically update deal pages with call summaries, identified risks, and proposed actions, ensuring all relevant information is centralized and easily accessible to the sales team.
• Resend for follow-ups built with React Email: For generating and sending personalized follow-up emails, we integrate with Resend. By using React Email, we can create highly customizable and branded email templates that the copilot populates with call-specific details, allowing AEs to send professional, human-sounding follow-ups with minimal effort.

Playbook

1. Cached transcripts in encrypted Supabase for auditability: Immediately after a call, Gong transcripts are securely cached in an encrypted Supabase database. This ensures that all conversation data is not only protected but also readily available for audit trails, compliance checks, and historical analysis, forming a robust foundation for the copilot's operations.
2. Fed prompts with deal history, stage and risks: To generate highly relevant and contextualized outputs, our AI prompts are dynamically fed with comprehensive deal history, the current stage of the opportunity, and any identified risks from our CRM. This rich context allows the copilot to understand the nuances of each interaction and tailor its summaries and suggestions accordingly.
3. Generated summaries, risks and actions straight into Notion: The AI processes the call transcripts and, based on our refined prompts, generates concise summaries, highlights potential risks, and proposes actionable next steps. These outputs are then automatically pushed directly into the relevant deal page in Notion, ensuring that all critical information is centralized and easily accessible to the sales team.
4. Drafted follow-up emails AEs can tweak in seconds: One of the most significant time-savers is the copilot's ability to draft personalized follow-up emails. Leveraging the call summary and identified next steps, the AI generates a human-sounding email that AEs can review, tweak in seconds to add their personal touch, and then send, drastically reducing the time spent on post-call communication.
5. Collected feedback and retrained prompts weekly: Continuous improvement is a cornerstone of our copilot's development. We have established a weekly feedback loop with our sales team. During these sessions, we collect their insights on the AI's performance, the accuracy of its outputs, and the effectiveness of its tone. This feedback is then used to retrain and refine our AI prompts, ensuring the copilot constantly evolves to meet the team's needs.

Key Principles of a Sales Copilot

• Contextual understanding: The copilot must deeply understand sales conversations, deal stages, and company-specific processes to provide relevant and actionable insights.
• Automation of administrative tasks: Free up sales representatives from time-consuming administrative work like note-taking, recap writing, and CRM updates.
• Human-in-the-loop control: Ensure that AI-generated content and actions are always reviewable and editable by sales reps, maintaining human oversight and personalization.
• Seamless integration with existing tools: Integrate the copilot smoothly with essential sales tools like conversation intelligence platforms, CRMs, and email services.
• Actionable insights and next steps: Beyond summaries, the copilot should proactively suggest next steps, identify risks, and recommend resources to advance deals.
• Continuous learning and adaptation: Implement feedback mechanisms to continuously refine the copilot's performance, prompt engineering, and alignment with evolving sales strategies.
• Enhanced forecast confidence: By providing comprehensive, up-to-date deal information and risk assessments, the copilot should contribute to more accurate sales forecasting.

Common Failure Modes (and Fixes)

1. Generic AI responses:
   • Problem: If the AI is not properly trained or lacks sufficient context, its summaries and follow-ups can be generic, lacking the personalization and nuance required for effective sales communication.
   • Fix: Implement robust prompt engineering. Feed the AI with rich, contextual data from the CRM (deal history, customer profile, previous interactions). Continuously refine prompts based on sales team feedback and successful outcomes.
2. Lack of AE trust and adoption:
   • Problem: Sales representatives may be hesitant to adopt an AI copilot if they feel it undermines their expertise or if the output requires extensive manual correction.
   • Fix: Emphasize the "copilot" aspect – the AI assists, not replaces. Ensure AEs have full editorial control over AI-generated content. Highlight time savings and improved deal hygiene. Involve AEs in the feedback loop from the start.
3. Data privacy and security concerns:
   • Problem: Handling sensitive customer conversation data requires strict adherence to privacy regulations and robust security measures.
   • Fix: Implement end-to-end encryption for all stored transcripts and data. Ensure compliance with relevant data protection laws (e.g., GDPR, CCPA). Clearly communicate data handling policies to the sales team and customers.
4. Over-reliance on AI leading to reduced critical thinking:
   • Problem: AEs might become overly dependent on the AI, potentially reducing their critical thinking skills or ability to adapt to unexpected conversational turns.
   • Fix: Position the AI as an augmentation tool. Encourage AEs to review and refine AI outputs, using them as a starting point rather than a final product. Integrate AI insights with sales coaching to reinforce critical thinking and strategic selling.
5. Integration complexities and data silos:
   • Problem: Integrating multiple sales tools (Gong, CRM, email) can be complex, leading to data silos or broken workflows if not managed properly.
   • Fix: Design a robust integration architecture with clear data flows. Utilize webhooks and APIs for real-time data exchange. Implement monitoring and alerting for integration failures to ensure seamless operation.

Metrics & Telemetry

• Time to follow-up: Reduced from an average of 25 minutes to a mere 4 minutes, significantly accelerating post-call engagement.
• Open tasks per deal: Decreased from 17% to 5%, indicating a substantial improvement in task completion and deal hygiene.
• Forecast confidence: Increased from 62% to 81%, reflecting more accurate and reliable sales projections due to enhanced data and insights.

What stuck with me

• AI needs guardrails; I cap length, tone and required fields: A critical insight from building this copilot is the absolute necessity of implementing robust guardrails for AI-generated content. Without clear boundaries, the AI can produce verbose, off-brand, or even inaccurate information. By explicitly capping the length of summaries, defining acceptable tones (e.g., professional, empathetic, direct), and requiring specific fields to be present, we ensure the AI's output remains consistent, high-quality, and aligned with our sales strategy. These guardrails are not limitations but rather essential controls that make the AI a reliable and trustworthy assistant.
• AEs adopt faster when they can edit before sending: Initial concerns about AI replacing human judgment quickly dissipated once AEs realized they were always in control. The ability to review and edit AI-drafted emails, summaries, and proposed actions before sending them out was a game-changer for adoption. This "human-in-the-loop" approach fosters trust, allows for personalization that only a human can provide, and ensures that the final communication always reflects the AE's unique style and understanding of the client. It transforms the AI from a black box into a collaborative partner.

What I'm building next

I'm currently exploring the exciting frontier of automatic coaching for our sales team, leveraging the rich conversation metrics captured by Gong. The idea is to develop an AI-powered system that can analyze call patterns, identify areas for improvement (e.g., talk-to-listen ratio, objection handling, discovery questions), and provide personalized coaching suggestions to AEs. This will move beyond reactive feedback to proactive, data-driven development for our sales force. If you're passionate about sales enablement and AI, and interested in joining the beta program for this initiative, please reach out and tell me more about your interest.

Want me to help you replicate this module? Drop me a note and we’ll build it together.