TL;DR
Microsoft's Finance Agent brings AI to collections in Dynamics 365 — prioritising accounts, drafting outreach, surfacing risk. Here's what it drives well, where the customer relationship stays human, and how to make it run on your policy.
Collections is where cash flow is won or lost, and it's mostly unglamorous follow-up: work the aged receivables, prioritise who to chase, send the right message at the right time, escalate the ones that have gone quiet. It's high-volume, rule-shaped work — exactly where AI earns its place — and Microsoft's Finance Agent for Dynamics 365 now brings a collections skill to it.
If you're asking whether AI can help you collect faster, the answer is yes — with one important line drawn clearly: the agent should drive the process, not own the relationship.
What the agent drives well
A collections agent takes the mechanical weight off a credit-control team:
- Prioritises the ledger — surfaces which accounts to work first by balance, age and risk, instead of a static aging report everyone reads top to bottom.
- Drafts the outreach — the reminder, the follow-up, tuned to where the account sits, ready for a person to send.
- Keeps the thread moving — tracks what was promised, flags the accounts that have gone silent, and surfaces the ones slipping toward a problem before they're a write-off.
The model, as everywhere in agentic ERP, is operate by exception: the agent handles the routine cadence and surfaces the accounts that need a human — a hard conversation, a payment plan, a decision to escalate. Your team stops working a list and starts working the exceptions. For most collections functions, that's a real, native win.
Where the relationship stays human
Collections is not pure process, and this is the line that matters: AI drives the follow-up; a person owns the relationship. Three judgement calls stay human:
- Who to push, and who to hold. Your best customer who's 10 days late is not your churned account who's 60 days late, and the agent must not treat them the same. That judgement is relationship context, and it belongs to your credit team — codified where possible as policy (business skills), but owned by people.
- Tone and escalation. A dunning sequence that's correct and tactless loses a customer. The agent drafts; a person decides when a real conversation replaces an automated one.
- Disputes and exceptions. A "late" invoice that's actually disputed, a credit note in flight, a special arrangement — these are hand-backs, and getting them wrong damages trust and the number.
How to make it run on your collections, not a generic one
The first-party agent gives you the engine. Making it collect the way your business does needs the specifics:
- Your policy, codified — credit terms, grace, escalation thresholds, who signs off a payment plan — so the agent prioritises and drafts by your rules, not defaults.
- Your customer and AR data reached safely — including any customized fields or ISV logic the agent can't see out of the box (governed MCP Action tools).
- The controls — approval before a hard escalation, an audit trail of what was sent and promised (agent governance).
Done right, the agent clears the cadence and your credit team spends its time on the conversations that actually move cash — which is the whole point.
If collections is eating your team's week and the follow-up is the bottleneck, tell us how your collections process and policy work — including where judgement has to stay human, because that's the part a good build protects.
Share this article
Agentic ERP
Topic Cluster
