Governed AI agents, explained
The five controls every agent runs under, and how governed agents compare with RPA bots and chatbots.
Read the explainerAI collections agent for accounts receivable Follow-up that doesn't slip. Your team still owns every escalation.
The collections agent tracks aging in your accounting system, drafts reminders in your voice from templates you approve and recommends which accounts to escalate. Escalations and disputes stay with your team.
overdue invoice
Illustrative
- Aging reviewedOpen invoices read from your accounting system
- Agent drafts the next touchFrom your approved templates, in your voice
- Policy checkDisputes and accounts due for escalation are flagged
- Your team decides ApprovedEscalations and disputes go to a named owner
- Reminder goes outOnly on a template you approved, from your AR inbox
- Logged and reportedEvery touch recorded, exceptions in the monthly report
What it does
From invoice past due to paid or escalated, step by step.
-
01
Track aging
Reads open invoices and aging from your accounting system on the schedule you set.
-
02
Prioritize
Works out which accounts are due a reminder and which need more than a reminder.
-
03
Draft reminders
Writes each reminder in your voice from the templates you approve, with the invoice details filled in.
-
04
Recommend escalations
Recommends which accounts to escalate, with the aging and contact history attached, for your team to decide.
-
05
Log every touch
Records each reminder and decision against the account and reports exceptions every month.
What reaches your customers
- Only reminders built on templates your team has approved
- Written in your voice and on your terms
- New wording or a new template goes through you first
- A record of every reminder sent, on the account
What stays with your team
- Every escalation decision
- Every dispute, with the invoice and contact history attached
- Anything unclear, incomplete or outside the agreed rules
- The relationship with the customer
Where your team decides
The agent keeps follow-up moving. People make the calls.
These rules are written into the design and approved by your finance team before anything is built.
- Reminders follow templates you approve. Your team signs off the wording and the rules in the design. Anything outside them goes to a person first.
- Your team decides every escalation. The agent recommends which accounts to escalate and shows why. A person decides what happens next.
- Disputes stay with people. When a customer disputes an invoice, it goes to a named owner on your team with the history attached.
- Approvers are written down. The design names who owns each decision point, so the agent always knows who to ask.
- Actions come from an approved catalog. The agent only acts on the systems and actions you have approved. Anything new goes through you first.
- Every touch is logged. Each reminder, recommendation and decision is recorded against the customer account.
What it connects to
The systems your AR team already uses.
- Your ERP or accounting system
- Your AR inbox
- Your approval tool
The agent works in the systems you run today rather than a new platform your team has to learn. We confirm the systems, access and approval path in the written design, and access is granted through your own controls, limited to what the design calls for.
What you get
A working agent and everything behind it.
- A written design and approval matrix
- Reminder templates your team has approved
- The agent deployed in your cloud and model accounts
- An audit log and a monthly report on activity, decisions and exceptions
- Source code and runbook in your Git
How it's measured
Four measures, baselined before anything is built.
We size the agent on these measures in the Opportunity Scan and design. After go-live, the monthly report tracks the same measures against your baseline.
-
01
Hours returned
Time your team spends pulling aging, writing reminders and keeping track of who said what, and how much of it the agent takes on.
-
02
Cycle time
Days sales outstanding: how long it takes to collect after you invoice.
-
03
Risk and leakage
What slips today: past-due balance, and overdue accounts that go too long without a follow-up.
-
04
Cost against the alternative
The fixed-fee build and monthly run compared with hiring, outsourcing collections or leaving them as they are.
Agent or AR software
How this differs from AR and collections software.
Is an AI collections agent just old dunning software with new branding? It's a fair question, and part of the answer is yes. Scheduled reminders and aging reports have been around for years, and receivables products handle them well for many teams. The difference is less about the technology and more about what you buy, where it works and who runs it.
| Question | AR and collections software | A governed collections agent |
|---|---|---|
| What you buy | A software product and its workflow | A service built and run around your workflow |
| Reminders | Sequences you configure in the product | Drafted in your voice from templates you approve, with the account's history in view |
| Where the work happens | In the product, connected to your accounting system | In your accounting system, AR inbox and approval tool |
| Who runs it | Your team, with the vendor's support | We do, every month: monitoring, tuning and updates |
| Ownership | Licensed from the vendor | Source code in your Git, running in your accounts |
If a packaged product already fits your process and your team has time to run it, that may be the simpler answer. For how agents compare with RPA bots and chatbots, see governed AI agents, explained.
Questions
What CFOs and controllers ask first.
Pricing follows the standard model: a fixed-fee build and a monthly run.
Will the agent contact our customers without our approval?
Only through templates you approve. Your team signs off the wording and the rules in the design, and anything outside them goes to a person first.
What happens when a customer disputes an invoice?
It goes to your team. Disputes stay with people, and the agent passes on the invoice and contact history so nobody starts from scratch.
Who decides when an account is escalated?
Your team. The agent recommends which accounts to escalate and shows why, and a person decides every escalation.
Is an AI collections agent just old dunning software with new branding?
Some of the building blocks are familiar. What's different is that the agent drafts each reminder in your voice with the account's history in view, works in the systems you already run, follows approval rules written before it's built, and is operated by us every month. You own the code and the accounts it runs on.
Do we need new AR software?
No. The agent works in your ERP or accounting system, your AR inbox and your approval tool. We confirm the systems and access in the written design.
How do we know it's working?
We baseline hours returned, days sales outstanding, past-due balance and cost against the alternative during design. The monthly report shows those measures and every escalation and exception against that baseline.
What does it cost?
A fixed fee to build and a monthly subscription to run, sized in the Opportunity Scan against the agent's business case. Hosting and model costs run in your own accounts with no markup.
Go deeper
How we govern, build and run every agent.
How it works
Scan, design, build and run, with a fixed-fee build, a monthly subscription and a report against your baseline.
See the engagement modelClaude session vs. production agent
Why reminders drafted in someone's chat session aren't the same as an agent running in production.
See the side by sideManaged Intelligence Opportunity Scan
See what slow collections are really costing.
In 30 minutes, we'll walk through how follow-up works today, where your team decides, and what an agent would change. No commitment to build.