Every financial-analysis platform demos well. That isn't a criticism of the vendors — it's what a demo is built to be. The data is clean, the path is rehearsed, and the parts that break get skipped. Nobody opens a sales call by loading a rotated scan of a 1997 credit union statement.
The problem is that the tool doesn't get used on the demo file. It gets used on the box of records a client dropped off, and its output has to survive a setting most software never faces: someone whose professional job is to discredit it.
That gap — between what a demo shows and what a case demands — is where buying decisions in this field go wrong.
01Evaluate the failure modes, not the features
Feature lists are the wrong comparison surface. Every serious tool in this category claims extraction, reconciliation, tracing, and reporting. The lists converge; the products don't.
What separates them is how they behave when something goes wrong — and in forensic work, something always does. A statement is smudged. A bank changes its debit convention mid-year. A balance doesn't tie. The useful question isn't "does it have feature X" but "what happens when X meets a document that doesn't cooperate?"
What a feature list tells you
- That extraction exists
- That reconciliation is offered
- That reports can be generated
- That the vendor has competitors with the same list
What a failure mode tells you
- Whether the tool admits it hit trouble
- Whether a human can intervene
- Whether the intervention is recorded
- Whether you could explain any of it on the stand
02The three things a demo cannot show you
A demo runs on records chosen because they work. Until a tool has processed your messiest real statement — the multi-column one, the scanned one, the one with continuation pages — you've learned almost nothing about how it will handle your caseload. Ask to supply the file. A vendor's willingness to run your document live is itself a data point.
Demos are short and narrated. Daily use is neither. The functions that decide whether a tool is tolerable are the boring ones you'll touch a hundred times — filtering by account, by date, by payee. Those are also the ones most likely to break on real data, and least likely to appear in a scripted walkthrough.
No demo simulates opposing counsel asking how a figure was produced. But that moment is the entire point of the work. If the answer involves a proprietary algorithm the vendor won't detail, you've inherited a problem you cannot solve at the podium.
03Why "certainty" should make you cautious
Marketing in this category leans hard on words like verified, complete, and certainty. It's understandable positioning — buyers want reassurance. But in litigation, a certainty claim is a surface to attack.
An expert who testifies that software verified something now has to explain how. If the method is proprietary, that explanation isn't available, and the challenge shifts from your conclusions to your methodology — a considerably worse place to be defending.
You don't have to share a particular philosophy to test for this. The walk-through either exists in plain language or it doesn't.
04Seven questions worth asking every vendor
In short form, the evaluation reduces to seven questions — four about whether the work survives scrutiny, three about whether your team can actually operate the thing.
- When a figure doesn't reconcile, does the tool surface it — can your investigator correct it — and is that correction recorded somewhere it can't quietly change?
- Can the vendor explain, step by step, how a number was produced?
- Does every figure lead back to the exact page it came from, in a single step?
- Who and what handles your case data, and is any part of that outsourced or crowdsourced?
- Do the routine functions hold up on your records rather than the sample file?
- Can a small team run it without a training budget or an implementation project?
- What are you obligated to before you know whether it works?
The full guide — with what to watch for on each — is here: Choosing a Financial-Analysis Tool: 7 Questions to Ask First. There's also a one-page scorecard for running several vendors side by side.
05A note on cost
Question seven is the one that surprises people. Enterprise pricing in this category assumes a large, predictable caseload and a procurement cycle to match — annual commitments, seat counts, five-figure minimums. Forensic work rarely looks like that. Volume swings case to case, and a small unit's biggest risk isn't paying too much per page; it's committing to a year of something before a single real matter has run through it.
So ask the unglamorous questions. Is there a minimum? What happens to volume you don't use? Can you start on one matter and expand only if it earns the place? The answers say a great deal about who the product was designed for.
06The short version
Bring your own documents. Test the boring functions. Ask how a number was produced, and keep asking until you get an answer you could repeat under oath. Then find out what you're signing up for before you know whether it works.
Do that with every vendor — including this one.
