Financial Intelligence · Research Note
The invoice is overdue. What is actually holding up payment?
Can a collections workflow help resolve payment delays—and what additional evidence is needed when reminders do not address the blocker?
Trove · Xero · Accounts receivable · Invoice reminders · Commercial completeness · Cash visibility · Commercial context
How this connects
This write-up examines one part of the central thesis:
Are we delivering the business we intended?
It answers one part of the wider question Can AI help finance detect gaps between the accounts and commercial reality—before they distort decisions?
Tested on Xero.
The app I tested
Trove
Trove is a credit-control app for businesses that invoice through Xero. It is designed to take the routine work out of chasing unpaid invoices: it reads the outstanding invoices from Xero and sends reminders and statements from your own Gmail or Outlook account, with your signature. Its listing also describes AI that varies the wording of reminders and helps handle customer replies.
What worked as designed. In my test Trove connected to Xero, picked up the unpaid invoices already recorded there and offered a clear follow-up sequence by email: reminders at 7, 14 and 21 days, monthly statements, thank-you messages and a manual-send option. For a business whose main problem is invoice chasing, that is a tidy and well thought-out workflow.
The rest of this page records where my research question went beyond what Trove sets out to do, and what I have not yet tested.
An overdue invoice tells us that cash has not arrived. It offers surprisingly little explanation.
The customer might need a reminder. They might also be waiting for a purchase order, disputing the work or following a payment arrangement finance has never seen.
Another reminder may help. It may also make everyone wonderfully busy while the actual blocker remains untouched.
Xero experiment 01 · 16 November 2026 · Collections workflow tested using Trove.
The business problem
Unpaid invoices create uncertainty about cash and consume time across finance, sales and operations. The useful question is what is preventing payment and which action could change the outcome.
I tested a collections application to understand which part of that problem its observed workflow addressed.
The question being tested
Can a collections workflow help resolve payment delays—and what additional evidence is needed when reminders do not address the blocker?
This initial test inspected the workflow. It did not measure collection results or independently test the diagnosis of payment blockers.
Connected components
Setup confirmed by the author, who made the Xero and Gmail connections inside the application.
Trove + Xero + Gmail — operated inside Trove.
| Component | Connection and role |
|---|---|
| Trove | Application used to inspect and configure the collections workflow |
| Xero | Connected through Trove’s Xero integration; supplied recorded unpaid invoices |
| Gmail | Connected to Trove for collections correspondence |
| ChatGPT / MCP | Not part of the tested collections workflow; ChatGPT helped document the experiment |
The Gmail connection here served Trove’s workflow. This was not a test of ChatGPT independently searching Gmail and Xero together. Advanced interpretation of customer replies was not independently tested.
The setup
I connected Trove and examined its default workflow alongside Xero's existing receivables.
Trove's listing describes AI-assisted credit control, including interpreting customer replies and payment promises, drafting responses and helping with collection exceptions. Its documentation describes an AI Reply Agent that reads replies to collection emails. These are relevant capabilities to test; I did not independently evaluate the advanced reply handling in this experiment.
This entry records my experience with the workflow and configuration tested. It is not a comprehensive product review.

Original marketplace screenshot, captured 16 November 2026.
What I tried and observed
The workflow started with unpaid invoices already recorded in Xero, with connected email used for follow-up. I saw reminders at 7, 14 and 21 days after the due date, monthly statements, payment acknowledgements and a manual-send option.

Original workflow screenshot, captured 16 November 2026.
Xero already offers automated invoice reminders, with up to five configured reminders according to the documentation reviewed. That gives me a baseline for a later comparison of workflow effort, reply handling and collection outcomes. It does not establish that the two products deliver equivalent value. Xero documentation.
In my setup I appeared to connect one mailbox for collections correspondence. Trove also documents shared inboxes and multiple sending addresses, so this observation should not be read as a product-wide mailbox limit. AI Reply Agent documentation.
| Question | Evidence status |
|---|---|
| Does the observed workflow start from existing Xero receivables? | Observed |
| Can it interpret replies to collection emails? | Vendor-documented; not independently tested here |
| Can it discover completed work never invoiced? | Not tested or demonstrated |
| Does it improve collections compared with native reminders? | Not measured |
What this establishes
The observed workflow follows up unpaid invoices already recorded in Xero. That is a useful, defined task. Its advanced reply handling and incremental collection outcomes need their own assessment.
The wider question is whether commercial correspondence can explain why cash has not arrived and direct the appropriate response. Discovering work that was never invoiced is a further, separate question.
From finding to a measurable initiative
A next test would take a defined sample of overdue invoices and identify the apparent blocker: a reminder needed, a dispute, missing approval, incorrect billing or another cause.
Suggested actions would be compared with human review. We would track resolution time, collection timing, manual effort and incorrect interventions against a baseline. Faster collection would be assessed as a working-capital benefit, rather than counted again as new revenue.
This is a proposed test. No collection improvement, time saving or ROI has been measured in this experiment.
How this connects to the central thesis
“Are we delivering the business we intended?” asks whether leadership can connect intentions, execution and economic outcomes while there is still time to act.
This experiment starts with a familiar gap: expected cash has not arrived. It examines the follow-up workflow and defines the next question—whether additional context can identify the intervention that changes the outcome.
Supporting images


Supporting evidence
Two original screenshots captured 16 Nov 2026: (1) Xero marketplace listing with vendor claims; (2) default Trove workflow showing reminders at 7/14/21 days, statement on 1st, thanks-for-paying, and manual sending. First-person connected workflow observations. Original screenshot links included in article body.
Sources
- Original Trove marketplace screenshot, 16 Nov 2026
Vendor claims, embedded in article
- Original Trove workflow screenshot, 16 Nov 2026
Observed reminder workflow, embedded in article
- Trove — AI Reply Agent
Vendor-stated email reply and inbox capabilities
- Xero — Set up invoice reminders
Native Xero reminder capabilities
Counterarguments
The marketplace advertises substantially more than basic reminders, including email-reply interpretation, payment commitments, PO issues and escalations. The experiment's limited workflow inspection is insufficient to reject these capabilities.
Current conclusion
The observed workflow follows up recorded unpaid invoices. Advanced reply handling, diagnosis of payment blockers and improved collection outcomes were not independently established.
How my thinking has changed
The focus is the payment problem and appropriate intervention, with the application assessed against the workflow actually tested.
Research question
Other experiments on Xero
Experiment 04 · 19 Nov 2026
Can AI connect email evidence with accounting records to establish what happened, whether it was recorded and whether the entry reflects the underlying commercial reality?
Read the article: What does the inbox know that the accounts don’t?Experiment 03 · 18 Nov 2026
Can connecting email with accounting records reveal commercial context that helps explain invoicing issues and identify the right intervention?
Read the article: The invoice tells you what was billed. What does the business know that finance doesn’t?Experiment 02 · 17 Nov 2026
Which unsupported transactions can document matching resolve, and which require a broader investigation?
Read the article: Another missing invoice—or a transaction that needs a different explanation?Planned
Xero JAX