Financial Intelligence · Research Note
The invoice tells you what was billed. What does the business know that finance doesn’t?
Can connecting email with accounting records reveal commercial context that helps explain invoicing issues and identify the right intervention?
Xero · Gmail · ChatGPT · MCP · Clearspend · Transaction investigation · Email evidence · CFO intelligence · 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.
An invoice records an amount, a customer and a due date. The explanation for what happens next may be somewhere else entirely.
A revised agreement. A disputed scope. An approval nobody forwarded. A customer saying they have already paid.
Finance sees an exception. Someone elsewhere in the business may already know the explanation. Unfortunately, the explanation has not attended the finance meeting.
This experiment explored whether connecting email and accounting records through ChatGPT could bring commercial context into the investigation—and change the questions we ask. Revised agreements, disputes and missing approvals describe the broader problem; they were not all demonstrated in this test.
Xero experiment 03 · 18 November 2026 · Commercial-context investigation using ChatGPT and connected plugins.
The business problem
The financial record and the business’s understanding of an event can drift apart. A payment notification, an explanation of what a receipt covers or an acknowledgement in correspondence may clarify an entry that looks incomplete or misleading on its own.
When that context remains scattered, people have to reconstruct it across systems before deciding what needs attention. A matching result is useful, but it may leave the commercial question open.
The question being tested
Can connecting email with accounting records reveal commercial context that helps explain invoicing issues and identify the right intervention?
The strategic interest is whether that context can support better decisions about cash timing, billing completeness and confidence in financial information. This test examined selected examples; it did not establish those wider outcomes.
Connected components
ChatGPT + Gmail connector + Xero MCP + Clearspend MCP — queried through connected plugins in ChatGPT.
| Component | Connection and role |
|---|---|
| ChatGPT | Conversational workspace used to direct searches, compare evidence and develop explanations |
| Gmail plugin / connector | Gave ChatGPT direct access to search and read email evidence, separately from Clearspend’s Gmail connection |
| Xero plugin via MCP | Gave ChatGPT access to accounting records through the exposed Xero tools |
| Clearspend plugin via MCP | Gave ChatGPT access to Clearspend’s transaction listings and document-match results |
Xero and Clearspend were queried through their MCP connections. Gmail used its own connector; this experiment does not label it a Gmail MCP connection. ChatGPT was the assistant using those connections.
Simpler configuration discussed: ChatGPT + Gmail connector + Xero MCP. This could investigate accessible Xero transactions against email evidence without Clearspend’s results. We did not run a separate controlled test with Clearspend disconnected, so this is not a fourth completed experiment. Full raw bank-feed coverage remains unverified.
What the correspondence added
What had been paid—and why
For a selected customer receipt, accounting information could be compared with payment notifications and customer correspondence. The emails helped explain what the payment related to and when it had been initiated and received.
That is more useful than matching an amount alone: it helps establish the commercial purpose and timing of the receipt before deciding whether further follow-up is appropriate.
Evidence beyond a formal invoice
For a selected accountant payment, correspondence included the fee discussion, confirmation that payment had been sent and an acknowledgement.
This supported an explanation of the payment. Whether a formal invoice should also be obtained remained a separate question. An explanation and sufficient accounting documentation are related, but they are not interchangeable.
A reason to question the description
Some HubSpot-labelled entries appeared more consistent with Mercury card repayments when compared with surrounding evidence. Other examples required gross payments, processing deductions and net deposits to be considered together.
Those were findings for review. They did not establish which system or rule introduced each label, nor justify automatically changing the records.
What we established—and what remains open
| Evidence level | Result |
|---|---|
| Additional context retrieved | Selected correspondence and payment notifications helped explain payment purpose and timing |
| Candidate explanations developed | Some descriptions appeared inconsistent with the surrounding evidence |
| Broader commercial scenarios | Disputes, revised scope, missing approvals and unbilled work remain further tests |
| Root causes and interventions | Not established or implemented for every discrepancy |
| Economic outcomes | No measured improvement in cash timing, close time, billing completeness or ROI |
Clearspend returned 128 unmatched entries in the examined dataset. We investigated selected examples, not the entire population. No financial records were changed.
From context to a measurable initiative
The next step would be a defined sample of invoicing and payment issues. For each, record what the accounting system shows, what correspondence adds, which explanation a reviewer accepts and what action follows.
Possible actions include confirming payment, obtaining missing support, referring a disputed item to its owner or reviewing a classification. An action should follow the evidence rather than be assumed from the exception label.
We could then measure the time to an accepted explanation, human corrections, time to resolve the issue and the relevant business outcome. If similar issues recur, a process intervention could be tested against a baseline.
Faster cash collection, reduced manual effort and more reliable information are outcomes to evaluate. They are not benefits this experiment has already proved.
How this connects to the central thesis
Email can contain part of the commercial explanation behind financial performance. Connecting it with accounting records may help leadership distinguish an administrative backlog from a payment issue, a billing problem or a gap in the underlying process.
The opportunity is to turn knowledge scattered across the business into better-informed action—and then track whether that action delivers its intended economic benefit.
This experiment is a step from processing financial records towards understanding the commercial activity behind them. Nobody was being handed the keys to the ledger.
Supporting evidence
In a human-directed investigation of an anonymised business dataset, Clearspend returned 128 unmatched entries. Selected examples yielded additional email evidence for customer receipts and accountant payments, candidate explanations for HubSpot-labelled entries, and gross-to-net payment relationships. Underlying records and correspondence remain private.
Sources
- Private experiment evidence — 18 November 2026
Selected Xero records, Clearspend MCP unmatched results, Mercury receipt notifications, customer payment correspondence and accountant acknowledgements reviewed during the experiment. Private mailbox links and unredacted documents are omitted from this public article.
Counterarguments
This was not a controlled comparison of products. We did not resolve all 128 entries or measure comparative accuracy, time or cost. We have not verified that the current Xero connector exposes every raw unreconciled bank-feed line or attachment. Proposed explanations require review.
Current conclusion
Selected email evidence added commercial context to payment and invoicing investigations. The test did not demonstrate every scenario, establish all root causes, implement interventions or measure an economic return.
How my thinking has changed
The strategic focus is bringing commercial knowledge into financial understanding so leadership can choose an appropriate intervention and subsequently test its effect.
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 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?Experiment 01 · 16 Nov 2026
Can a collections workflow help resolve payment delays—and what additional evidence is needed when reminders do not address the blocker?
Read the article: The invoice is overdue. What is actually holding up payment?Planned
Xero JAX