Churn risk early-warning alerts from Pylon to Slack

Build me an agent workflow that gives my CSM team an early warning when a high-value customer is getting frustrated in Pylon, so we can intervene before they churn.

Trigger: poll Pylon for the updated_issue event. Every time an existing Pylon issue is updated, run the agent.

What the agent should do on each fire:

1. Pull the most recent customer-facing messages on the issue using Pylon's List Issue Messages operation. Focus on the last few customer messages, not internal notes.

2. Judge customer sentiment and frustration level. Classify the conversation as positive, neutral, or negative, with a short one or two sentence summary of why. Only continue past this step when sentiment is clearly negative. Skip neutral and positive conversations silently so the channel does not get noisy.

3. Identify the account on the issue. Use Pylon's Search Accounts operation, preferring a match on the issue's account external_id when present, and falling back to a domain-based search using the requester contact's email domain.

4. Cross-reference that account in HubSpot using Search Companies, matched on the company domain. Pull tier, ARR, and lifecycle stage from the matched HubSpot company. If nothing matches in HubSpot, fall back to whatever ARR or tier fields are present on the Pylon account.

5. Apply a configurable high-value threshold. Defaults: ARR at or above fifty thousand dollars, OR tier equal to Enterprise, OR HubSpot lifecycle stage equal to Customer. Make ARR floor, tier list, lifecycle stages, and which combinator (AND vs OR) configurable at setup.

6. If sentiment is clearly negative AND the account is high-value, post a structured alert to a configurable Slack channel using Slack's Send a Message operation. The message should include: customer/company name, ARR, tier, lifecycle stage, a direct link to the Pylon issue, the short sentiment summary from step 2, and a one-line suggested next step the assigned CSM should take (for example, send a personal Loom, schedule a call, loop in the AE, or offer a credit). Keep formatting clean and skim-friendly with bold labels.

7. If sentiment is neutral or positive, OR the account is below threshold, exit silently without posting anything.

Setup-time inputs I want to be able to configure: target Slack channel, ARR floor, accepted tier values, accepted lifecycle stages, the threshold combinator (AND vs OR), and an optional override channel for very high ARR accounts (for example, a CS leadership channel for anything over two hundred fifty thousand).

Guardrails: do not post duplicate alerts for the same issue within a 24 hour window unless sentiment has clearly worsened. Do not include internal Pylon notes or PII like phone numbers in the Slack message. Always link to the Pylon issue rather than copying the full thread.