Weekly HubSpot deal recap posted to Slack every Friday

Every Friday at 4pm local time, a cron trigger kicks off a weekly deal-motion recap for our leadership team. This is a deterministic pipeline: fixed query, fixed grouping logic, one LLM formatting step, one Slack post.

Step 1 (HubSpot Search Deals). Call HubSpot's Search Deals with a filter on hs_lastmodifieddate greater than 7 days ago. Request these properties: dealname, dealstage, amount, hubspot_owner_id, closedate, createdate, hs_lastmodifieddate, plus any custom "won reason" and "lost reason" properties the account has. Resolve owner names by calling Get Owner (or batching them) on the returned hubspot_owner_id values.

Step 2 (group in code, deterministically). Sort the returned deals into five buckets:

• New deals created — createdate falls within the last 7 days.

• Deals that advanced stages — dealstage changed in the last 7 days and the new stage sits later in the pipeline than the previous one.

• Deals won — dealstage is closedwon and closedate is within the last 7 days. Include the won-reason field if present.

• Deals lost — dealstage is closedlost and closedate is within the last 7 days. Include the lost-reason field if present.

• Deals stalled — no modification in >7 days AND closedate is in the past.

For every deal in every bucket include: deal name, amount, owner name, current stage, close date, and the relevant reason field where applicable. Also compute per-bucket count and total amount.

Step 3 (Groq Create Chat Completion). Send the grouped payload to Groq using the llama-3.3-70b model (versatile variant is fine) with a fixed prompt template that outputs a written narrative in the same structure every single week. The system message locks the format so week-over-week diffs are readable at a glance. Target output shape:

*Deal recap — week of [start] to [end]*

*Won this week:* [N] deals, $[total] — [name, owner, amount, reason for each]

*Lost this week:* [N] deals, $[total] — [name, owner, amount, reason]

*Advanced:* [N] deals — [name, owner, old stage → new stage, amount]

*New:* [N] deals, $[total] — [name, owner, amount, stage]

*At risk (stalled past expected close):* [N] deals, $[total] — [name, owner, amount, days past close]

Keep the LLM temperature low (0.2 or lower) so the structure stays stable. No branching, no reasoning about what to include — the code decides which deals are in which bucket, Groq only formats.

Step 4 (Slack Bot Send a Message). Post the returned markdown to a configurable leadership Slack channel via the Slack Bot Send a Message operation. Slack mrkdwn (single-asterisk bold) is what the LLM should emit, so the recap lands with proper formatting.

Handle empty weeks gracefully: if all buckets are empty, still post a short "Quiet week — no closed or advanced deals" note rather than skipping.