Collections work is high volume, clear states, and a small set of true exceptions. The wrong design is a bot that pings everyone the same way. The right design is a system: truth, schedule, and a human queue.
Here is the pattern we ship for teams who are drowning in “just following up on invoice…” threads.
Layer 1 — One source of truth
Pick the system of record: accounting, billing, or ERP—not a spreadsheet copy.
Every automation reads open invoices from that source and writes status back. If Slack or email becomes the database, the system dies the first time someone deletes a thread.
Layer 2 — Escalation by age and amount
Define rules before you open n8n:
- Day 0: invoice issued (no chase)
- Day +X: polite reminder
- Day +Y: firmer notice + internal flag
- Amount thresholds: large balances escalate faster or to a human sooner
Templates differ by stage. One generic “friendly bump” trains customers to ignore you.
Layer 3 — Exception queue for people
Humans should only see:
- Disputes and partial payments
- Missing POs or wrong bill-to
- Accounts that bounced or unsubscribed
- Anything the rules cannot resolve
Success metric is not “messages sent.” It is how short the exception list is each morning.
Where n8n fits
Orchestration owns the glue:
- Pull open invoices
- Check last contact and stage
- Send the right channel/template
- Log the attempt
- Update status
Slack/email are surfaces. n8n (or similar) is the spine.
What we refuse to automate blindly
- Legal threats without review
- Customer-specific contracts
- Anything that can break trust if the model or template is wrong
Those stay on the exception path—with a clear owner.
Fixed-price first system
At isomore.tech we scope invoice chase as a fixed-price first system: discovery, written scope, then build against your stack. Source and data stay yours (Compass Ventures LLP / isomore.tech—no lock-in).
Next step
If collections is the workflow that hurts most, book a discovery call. Bring one sample aging report; we will say if it is automatable and what it costs.