Proof of cash tools are easy to demo and hard to evaluate. Every one of them will reconcile a clean QuickBooks export against a clean bank CSV and show you a high match rate. None of the deals you actually work on look like that.
Here is what we would test, in the order we would test it. Run one completed engagement with known problems through any platform you are considering, and score it on these.
1. Source-level traceability
Take any matched item in the output and try to get from it back to the bank statement line and the general ledger entry behind it. Then do it for an item the system classified or adjusted on its own. If you cannot get from a conclusion to its support in a couple of clicks, the tool has not saved you work — it has moved the work to whoever has to defend the number in front of a credit committee. A polished tie-out with no audit trail creates review risk rather than removing it.
Where we stand: every matched and unmatched item traces to the underlying bank statement line and general ledger detail, and exports carry that trail.
2. Real-world document handling
Test the ugly stuff deliberately. Photocopied statements. Image PDFs rather than native ones. Multiple institutions with different layouts. Statements that break a table across a page without repeating the header. A general ledger that arrives as an Excel file somebody has been maintaining by hand. This is where most tools quietly fail, because most were built and benchmarked on clean inputs.
Where we stand: this is the problem we built the platform around. A recent engagement involved 431 manually scanned statements across 36 bank accounts, from which we extracted 29,823 transactions. Scanned, skewed, low-quality source documents are the normal case for us, not the exception.
3. Exception quality, not headline match rate
A high auto-match percentage is the easiest number to optimize and the least informative. A tool that matches aggressively will pair up two similar payroll runs, or a card settlement with the wrong day’s batch, and report a beautiful match rate while quietly moving money to the wrong place. Ask for precision and recall by transaction type, then ignore the summary and read the unresolved queue. Is what is left over the interesting stuff — the items a person genuinely has to decide about? Or is it noise the system gave up on?
Where we stand: on the engagement above, the average match rate was 99.4%, and the single most useful output was not that number. It was the $32 discrepancy at one location that remained flagged at the end.
4. Transfer and sweep logic
Money moving between accounts the same owner controls is not revenue and not an expense, and a system that treats it as either will distort everything downstream. But over-suppressing transfers is equally dangerous: collapse too aggressively and you hide real cash movement. Test it with an owner who sweeps between operating and savings, an intercompany transfer, and an owner draw that goes out and comes back. Ask how the system decides, and whether that decision is visible and reversible. More on this in four things that break a proof of cash.
Where we stand: transfers are identified and netted, and the decision is exposed rather than buried — you can see what was collapsed and undo it.
5. Monthly and account-level, not cumulative
This is the test most often skipped and it matters more than almost anything else on this list. A cumulative reconciliation over a thirty-six-month diligence period can tie beautifully while concealing a cut-off problem, a timing shift, or a liability that appeared and disappeared. Require proofs by month and by bank account. Then look at whether match rates decline over time, which is often the first visible sign of deteriorating controls at the target.
Where we stand: reconciliation is per account per month, and the per-account monthly match rates are presented as a grid so a reviewer can see where the weak months are rather than being handed one aggregate figure.
6. Deliverable integration
The output has to land in the workpapers you already use: the proof-of-cash schedule, revenue testing, the EBITDA bridge, the normalized profit and loss, the net working capital analysis. If your team has to re-key results into a QoE workbook, most of the efficiency has already been spent.
Where we stand: output is GAAP-formatted monthly and twelve-month classified income statements and balance sheets with reported EBITDA, a trailing-twelve-month statement of cash flows, and a reconciling cash flows summary — designed to be the workbook rather than an input to one.
7. Human review and control
Low-confidence matches should be surfaced, not silently accepted. Analyst overrides should be visible, attributable, and preserved in the export. When a lender, an investment committee, counsel or the counterparty pushes back on a conclusion, you need to be able to say who decided what and why.
Where we stand: the system flags for human review and stops. It does not resolve ambiguity on its own, and we would not want it to.
Two places we would not oversell
Automation does not remove judgment. Everything above is data preparation. Whether an add-back is defensible, whether a customer relationship survives a change of control, whether management’s explanation of a margin swing holds up — none of that is a matching problem, and a tool that claims to decide it is a tool to be suspicious of. The value of automating the mechanical layer is that the judgment layer gets more attention, not less.
A proof of cash is not a quality of earnings report. A proof of cash establishes that reported activity matches bank activity. A full quality of earnings analysis adds normalized EBITDA, working capital and operating cycle metrics, customer concentration and change-of-control review. Tools in this category vary widely in how much of that second layer they actually cover, and the distinction is worth pinning down before you buy.
Running the test
Pick one closed deal you know the problems in — commingled personal expenses, merchant processor lag, an unrecorded liability, multi-account transfers — and put it through each candidate. Score on ingestion time, match accuracy after a reviewer has validated it, whether the unresolved queue was useful, analyst hours actually saved, auditability, and how cleanly the output moved into your workbook.
If you would like to run that test against us, get in touch. Our pricing is published: what a quality of earnings costs.