🏢 Business / Industry: B2B Sales / Sales operations

🎯 Main Goal: Automate daily sales pipeline analysis and deliver an AI-curated Telegram report — without manual data pulling, sorting, or commenting.

🔹 Context

Sales leaders typically spend 5–10 hours every week pulling deals from the CRM, sorting them by stage, calculating totals, and writing commentary by hand. By the time the report lands on Monday morning, the data is already stale — over the weekend deals close, stages shift, and the snapshot no longer reflects reality.

🔹 Problem

Manual weekly reporting is slow, expensive, and outdated on arrival. Worse, no one actually verifies these reports — reviewing would take as long as building them. The result: leadership operates on week-old data, blind spots accumulate, and CRM hygiene slowly drifts because no automated layer is watching.

🔹 Solution

An AI agent was built to run on a daily schedule, pull live data from the CRM, analyze the pipeline by stage, autonomously request additional context on problematic deals, and deliver a clean, readable summary directly to Telegram — with no human in the loop. Leadership now sees the real state of the pipeline every morning, in the same cycle the data was generated.

🔹 Process

• Scheduled trigger: A daily cron-based Schedule Trigger fires the workflow at a fixed time (e.g., before standup), guaranteeing fresh data on every reporting cycle.
• Data ingestion: The workflow pulls the full active deal list from the CRM (or a test dataset during development) — stage, owner, amount, last activity, and close date for every open opportunity.
• Data preparation: A Code Node normalizes the raw payload — converting timestamps, computing days-in-stage, aggregating amounts per stage, and pre-flagging deals that haven't moved within a defined threshold.
• AI analysis: An AI Agent powered by Anthropic Claude evaluates the prepared pipeline snapshot, identifies pattern-level issues (stage bottlenecks, stalled deals, high-value opportunities at risk) and produces a structured executive summary in natural language.
• Self-querying logic: For each flagged deal, the agent autonomously requests additional context (recent notes, last activity, deal owner) and integrates it into the final commentary — no separate follow-up workflow required.
• Telegram delivery: The finalized report is sent as a formatted Telegram message to the sales leader — pipeline totals first, per-stage breakdown next, then the list of problematic deals with AI commentary, structured for fast scanning on mobile.

🔹 Tech Stack