The account went quiet in March. Nobody noticed until November.
Health signals, QBR prep, renewal outreach — the account work that only gets done for the accounts somebody remembers. An agent watches every account on the same schedule and hands the at-risk ones to the CSM with the reasoning attached.
A CSM covering forty accounts is doing real work on eight of them.
Not through negligence — through arithmetic. The signals that predict a churn are spread across product usage, ticket volume, invoice history and the date of the last real conversation, and no one is reading four systems for forty accounts every week. So the review happens when the renewal date shows up in a report, which is the one moment when nothing can be done about it.
What it does every week, for every account, whether or not anyone asks.
It reads the account across the systems that hold it
Product usage from your Postgres or BigQuery, ticket volume and sentiment from the helpdesk, invoice and payment history, the last logged touch in Salesforce or HubSpot. One run per account, on the same schedule, with no favourites.
It writes the read, not just the score
A health number nobody can interrogate gets ignored by the second month. It produces the QBR pack, the usage narrative and the draft outreach, each citing the query or the ticket it came from, so a CSM can disagree with it specifically.
It stops before it speaks for you
Anything that commits the company — a discount, a roadmap promise, a renewal term, telling a customer they're at risk — is a gated action. It drafts and waits rather than sending.
The at-risk accounts go to a person, with the case attached
Not a flag on a dashboard. The CSM or AE gets the account, what changed, which signal moved, and the draft — on Slack, in time to do something about it.
Coverage that doesn't depend on who remembered.
One run per account, each with its own trace
Collection Pipelines map an agent across a list — your account table, a CRM view, a spreadsheet — as one isolated run per row. Each carries its own history, latency and cost, so “what did it look at for this customer” has an answer.
Drafts, not sends
Renewal outreach, QBR decks and check-in emails arrive as drafts on the owner's channel. Approve, edit, or ignore. The agent's job is that the draft exists on Monday for all forty accounts, not that it goes out unread.
Cost and coverage per account
Tag each run with the account it served and get cost, error rate and p95 split per account, side by side. Useful for the review, and unusually useful if you run this on behalf of someone else's book of business.
When it isn't sure
It doesn't decide who gets a discount.
Commercial judgment is the part of this job that must stay with a person, and the agent is configured to know it. Pricing and term questions route to whoever owns pricing; a renewal that needs a concession stops and asks rather than drafting one. Everything upstream of that decision — the reading, the assembly, the evidence — is already done by the time the question reaches them.
Gated actions on anything that commits the company: discounts, terms, roadmap promises, at-risk outreach
Routed by expertise — pricing to the person who owns pricing, product questions to the person who owns the roadmap
The whole case travels with the ask, so the answer is one line rather than an hour of digging
Your CRM, your warehouse, your helpdesk, your billing.
Connect a system once for the whole organization and every agent can use it from then on — with credentials held at the org and referenced by name, never pasted into a prompt.
Slack
GitHub
Jira
Google Drive
Salesforce
MongoDB
Notion
Linear
Grafana
Discord
Google Chat
Confluence
PostgreSQL
HubSpot
Airtable
Shopify
Stripe
Datadog
Sentry
GCP
BigQuery
Slack
GitHub
Jira
Google Drive
Salesforce
MongoDB
Notion
Linear
Grafana
Discord
Google Chat
Confluence
PostgreSQL
HubSpot
Airtable
Shopify
Stripe
Datadog
Sentry
GCP
BigQuery
Google Calendar
Gmail
Mixpanel
Monday
MySQL
SAP
Zendesk
Zoho CRM
Google Maps
Google Ads
Coralogix
Telegram
Apollo
Mailchimp
Calendly
Redis
Supabase
GCP Logging
gVisor
Loops
Typeform
Google Calendar
Gmail
Mixpanel
Monday
MySQL
SAP
Zendesk
Zoho CRM
Google Maps
Google Ads
Coralogix
Telegram
Apollo
Mailchimp
Calendly
Redis
Supabase
GCP Logging
gVisor
Loops
Typeform
Don't see yours? Import tools from any MCP server or request an integration.
Renewal questions, answered plainly
Mostly about what it's allowed to say to a customer.
Does it email our customers directly?
Only if you let it, and most teams don't at first. The default posture for this job is draft-and-hand-over: it writes the outreach and puts it in front of the account owner. Sending is a separate permission you grant per action.
Is this a health score product?
No. A score is one number and this produces the reading behind it — which query moved, which tickets, which invoice — because the CSM has to act on the reasoning, not the number. If you already have a health score, the agent can read it as one more signal.
How does it know what “at risk” means for us?
You tell it, in the same documents your team already uses, and it asks when a case doesn't fit them. The first month of escalations is effectively an interview about how your team actually judges an account.
What about accounts where the data is a mess?
It'll say so. A run that can't reach a signal reports the gap rather than scoring around it, and a malformed record fails on its own instead of taking the batch down.
Onboarding & Implementation
Chases what the customer still owes you, provisions what it's allowed to, and escalates the scoping calls.
Customer Support
Works the queue in your helpdesk, acts in the systems behind it, and asks before it invents a policy.
Back Office
Invoices, internal tickets, order exceptions, vendor files — the work in the queue.