Summary
- Trove is credit control software for growing UK finance teams. Trove’s AI Reply Agent reads replies to Trove’s chasers and drafts the next step, on every plan.
- An accounts receivable (AR) agent reads customer replies to your chasers and picks out what the customer said: a promised date, a dispute or a new contact.
- An AR agent drafts the reply and proposes the ledger update, such as setting an expected payment date in Xero or Business Central.
- In a controlled setup, every action an AR agent proposes waits for approval by default. Nothing is sent or written back until someone approves it.
- Auto-approve should be switched on one action type at a time, once you’ve approved that type enough to trust it.
- Every action an AR agent takes should be logged against the customer, so anyone can see what was proposed, approved and changed.
What an AR agent reads
An AR agent works on the replies to your chasers. That’s where the work piles up: a credit controller sends a batch of reminders on Monday and spends the rest of the week reading what comes back.
- Replies in your own mailbox. The agent connects to Gmail or Outlook, including shared inboxes such as accounts@ or finance@, and reads replies to the chasers it sent.
- The whole thread. It reads the reply in context, so “yes, that one” is understood as the invoice in the previous email.
- Remittance advice. Payment details in the body of an email or in an attached PDF, matched to the invoices being paid.
Trove’s AI Reply Agent reads replies to Trove’s chasers in Gmail and Outlook, including shared inboxes, and reads remittance in the email body or an attached PDF.
What an AR agent drafts and proposes
For each reply, the agent works out what the customer said and proposes what should happen next. A proposal is a suggested action, waiting on a task list. It isn’t done yet.
| The customer says | What the agent proposes | What needs approval |
|---|---|---|
| "We'll pay on the 3rd" | Set the expected payment date in the accounting software, pause chasing, book a follow-up in case the payment doesn't land, draft a reply confirming the date | The date change and the reply |
| "We've already paid this" | Draft a reply saying you'll look out for the payment, then update the draft to confirm receipt once the payment shows in the accounting software | The reply |
| "The price on this invoice is wrong" | Draft an acknowledgement for a person to review | The reply. Resolving the dispute, and whether to keep chasing, stays with a person |
| "Please send invoices to accounts@ instead" | Add the new contact and sync it to your accounting software | The new contact |
| "Can you send a copy of the invoice?" | Draft a reply covering the request | The reply |
| Remittance advice listing three invoices | Match the payment to the invoices it covers and the date it's due | Any change to what's being chased |
The pattern is the same every time: the agent does the reading and the typing, and a person makes the decision.
What waits for approval and when to switch on auto-approve
Every action waits for approval by default. That’s the right starting point for any finance team, because a wrong date in the ledger feeds straight into the cash forecast, and a wrong email goes to a customer.
Approval is usually quick. The credit controller opens the task list, reads the reply and the proposal side by side, and approves, edits or skips each one.
Auto-approve comes later, and only for one action type at a time. A sensible order:
- Expected payment dates. Low risk and high volume. Once you’ve approved a few dozen without changing one, switch them to auto-approve.
- New contacts. Also low risk, and easy to check in the log.
- Replies to “we’ve already paid”. Usually safe once you’ve seen the drafts are right.
- Anything about a dispute. Keep these on approval. Disputes need judgement, and often someone in sales.
In Trove, individual action types can be switched to auto-approve once you’re happy with them, and everything the agent does is logged against the customer.
Worked example: one UK customer reply handled end to end
This example is is the kind of work credit controllers handle every day. Example Supplies Ltd uses Trove with Xero. Its customer Penrose Interiors Ltd is overdue by nine days paying INV-1042 for £4,860.
1. The reply arrives. Trove’s second reminder went out from Example Supplies’ shared accounts@ inbox in Outlook. Jo in Penrose’s operations team replies:
Hi, sorry for the delay. We’ll pay on the 3rd. Can you send a copy of INV-1042 to accounts@penrose.example? They handle payments, not me.
2. The agent reads it. Trove picks out three things: a promised date (3 December), a request for a copy of INV-1042, and a new contact who handles payments.
3. The agent proposes. Three tasks appear on the credit controller’s list:
- set the expected payment date on INV-1042 to 3 December in Xero
- add accounts@penrose.example as a contact for Penrose Interiors
- send a drafted reply, addressed to the new contact, confirming the 3rd and covering the copy invoice request
4. The credit controller approves. Tim at Example Supplies reads the reply, approves the date and the contact, tweaks one line of the draft and sends it. It takes under a minute.
5. Trove writes back. The expected payment date and the new contact sync to Xero. Anyone looking at the invoice in Xero now sees 3 December, and so does the cash forecast. Here’s how updating the expected payment date from an email works.
6. Chasing pauses and a follow-up is booked. No more reminders go to Penrose while the date is pending. Trove schedules a follow-up in case the money doesn’t arrive.
7. The payment lands. Penrose pays on 2 December. Xero reconciles it, Trove sees the payment and cancels the follow-up. Nobody had to remember anything.
Logging the date matters because Trove’s data shows 7 in 10 promised payment dates are met, or missed by no more than three days. When a missed date got a follow-up within three days, 76% were paid within a week, against 49% without one. The agent makes both the logging and the follow-up automatic.
AR agents vs rules-based reminders
Rules-based reminders send on a schedule. An agent reacts to what the customer says. Most teams need both.
| Rules-based reminders | AR agent | |
|---|---|---|
| What triggers it | A date: 7 days overdue, 14 days overdue | A customer reply |
| What it does | Sends the next email in the sequence | Reads the reply, proposes actions and drafts a response |
| Changes to your ledger | None | Proposes expected payment dates and contacts, written back once approved |
| Human role | Set up the schedule | Approve, edit or skip each proposal, until an action type is trusted |
| Where it goes wrong | Chasing a customer who's already promised to pay | A misread reply, which approval catches |
For the wider picture of where AI helps in credit control, and where the claims run ahead of the product, see AI in credit control: what’s useful and what’s overhyped in our credit control software guide.
Questions to ask a vendor about its AR agent
If a credit control vendor says it has an agent, ask these questions to see how much work it can take on.
- Which inbox does it read? Your own Gmail or Outlook, shared inboxes included, or only replies sent to the vendor’s address?
- What can it change in your ledger? Expected payment dates, contacts, notes? Does it write back, or only to its own system?
- What happens by default? Does every action wait for approval, or does it send from day one?
- Can you auto-approve one action type and not another? All-or-nothing is a warning sign.
- Where’s the log? You should see every proposal, who approved it and what changed, against the customer.
- What happens when the payment lands? Does it cancel the follow-up it booked, or keep chasing?
- Can it read remittance PDFs? Ask to see one matched to invoices.
Ask the vendor to run one of your own customer replies through it in the demo. Five minutes on a real reply tells you more than a slide.
FAQs about AI agents in accounts receivable
What is an accounts receivable AI agent?
An accounts receivable AI agent reads customer replies to your payment chasers, works out what the customer said, drafts the response and proposes updates to your accounting software. In a controlled setup a person approves each action before it happens. Read more about how Trove’s AI Reply Agent works.
Does an AI agent send emails without approval?
It shouldn’t by default. Trove’s AI Reply Agent puts every drafted reply and action on a task list for approval, and you can switch individual action types to auto-approve once you trust them.
Can an AI agent read remittance advice?
Yes. Trove reads remittance details in the body of an email or in an attached PDF, and picks out which invoices are being paid and when.
Does an AR agent update Xero?
Trove’s does, once you approve. Expected payment dates and new contacts sync back to your accounting software, such as Xero, QuickBooks or FreeAgent.
Which Trove plans include the AI Reply Agent?
Every Trove plan, from Essential up.
Trove runs a free 30-day trial and takes about five minutes to connect to your Xero account. You can start a free trial here or book a demo and bring a real customer reply to try it on.
