HubSpot (CRM)
Last contact date, meeting and email activity, open and lost deals, lifecycle stage, account owner.
Customer
Know which customers are leaving months before they tell you
Churn rarely starts with a cancellation. It starts with a quieter inbox, a smaller order, a longer gap between purchases. The two halves of that signal live in different systems: engagement in your CRM, order behaviour in your ERP. Coheed reads both every morning, scores each account on future value and churn risk, and puts the accounts that are slipping in front of the people who can still save them.
- Systems combined
- HubSpot (CRM)Exact (ERP)
- Runs
- Daily at 8am
Sound familiar?
Teams bring us this use case when one or more of these is true.
- You find out a customer has left because the reorder simply never arrives.
- Your CRM says the relationship is healthy because someone logged a call last month, while their order value halved.
- Account owners work from gut feel about which customers need attention this quarter.
- Nobody can tell you which twenty accounts carry the most revenue at risk right now.
How it works
Coheed runs this as a scheduled insight over your connected systems, then hands the result to an agent.
- 1
Signals are collected from both sides
Every run pulls CRM engagement history (meetings, emails, deal activity, last contact) and ERP transaction history (order frequency, order value, invoice history) for the same account, matched on the cross-system account link the integration layer maintains.
- 2
Trends are compared, not snapshots
A single quiet month means nothing. Each account is compared against its own rolling baseline, so a steady customer whose order interval stretches from 30 to 55 days is flagged even while the absolute numbers still look acceptable.
- 3
Every account gets a value and a risk score
Accounts are ranked by predicted lifetime value and by churn risk, so retention effort goes where both are high instead of to whoever complained most recently.
- 4
The insight becomes an action
Flagged accounts are handed to the CLV Retention Orchestrator agent, which prepares a retention playbook per account and routes anything touching discounts or contract terms through human approval.
Which data it uses
Only the fields this use case needs are read. Nothing is copied that the insight does not use.
Exact (ERP)
Order history and frequency, order and invoice value, payment history, active contracts.
Coheed
The cross-system account link, the rolling baseline per account, and the score history that makes the trend visible.
What changes
What a team notices once this runs. No invented percentages: the effect depends on your data and your process.
- Retention conversations start while the customer is still buying, not after they stop.
- Account owners get a ranked list every morning instead of a quarterly report.
- Revenue at risk becomes a number you can report on, per account and in total.
- Discount decisions stay with a human, while the analysis and preparation do not.
From insight to action
CLV Retention Orchestrator
Running in the platform
Runs every four hours, flags accounts where CRM engagement decline and ERP order decline reinforce each other, and proposes a retention playbook per account. Discounts and contract changes are prepared as an approval pack for a human, never executed on their own.
Mode: Human-in-the-loop
Every insight can stay read-only, run with human approval, or run autonomously. You decide per action type, and you can change it later.
FAQ
Frequently asked questions
How is this different from a churn score in my CRM?
A CRM churn score only sees CRM behaviour: emails, meetings, deal stages. It cannot see that the customer's orders got smaller, because orders live in the ERP. Coheed scores on both halves, which is what makes the early signal reliable.
How much history do you need before the score is useful?
Roughly twelve months of order history per account gives a stable baseline. Accounts with less history are still scored, but marked low confidence so nobody acts on noise.
Does it write anything back into my systems?
Only what you allow. The insight itself is read-only and the agent proposes actions; writing a task, note, or property back to the CRM is a step you enable per action type.
Which systems does it need?
One CRM and one ERP is enough. The example above uses HubSpot and Exact, but the same logic runs on any CRM and ERP pair in the connector catalogue.
Built on these maturity levels
Other use cases
Lead-to-Quote Readiness
Surface correct ERP pricing, live stock, discounts, and customer agreements on every new CRM lead so sales can quote immediately.
Order Exception Monitoring
Catch stock gaps, unrealistic delivery dates, pricing errors, and missed agreements after the order, before the customer notices.
True Cost-to-Serve per Customer
Combine ERP margin, CRM service workload, and SCM exception handling to find high-revenue customers who are quietly unprofitable.
See this on your own data
Tell us your ERP, CRM, and SCM stack and we will show what this use case looks like on your systems.