Short answer
Evaluate spreading software on five things: whether it can be bought without an origination platform, which tax forms it names, whether global cash flow is produced inside the product, what evidence it leaves for an examiner, and whether a named institution your size will take your call. Everything else in a demo is a variation on a template, and every vendor's template looks good on a clean audited statement.
Spreading demos are unusually similar to each other. A clean PDF goes in, a tidy spread comes out, ratios populate, and the room nods. The differences between these products are real and large, and almost none of them are visible on that file. This guide sets out how to make them visible, and flags four disclosure gaps that are close to universal in this category so you can plan around them instead of discovering them in month three.
Settle the purchase shape before the feature comparison
The first question is not what the software does. It is what you are being sold. Some vendors sell spreading as a product a credit department can order. Others sell it as one module of a commercial origination platform, priced per seat or by asset size on a multi-year term, where spreading is the entry point to a programme. Both are legitimate businesses. Confusing them is how a discrete problem becomes a three-year commitment.
The test is blunt: ask for spreading quoted on its own, with no origination modules attached, and see who produces a number. In this category the answers sort the market faster than any feature matrix, and at least one very widely recommended platform publishes no standalone path at all.
- Ask for a spreading-only quote in writing, before any capability discussion
- Ask whether other institutions have bought it that way, and for one by name
- Ask what the price does in year two, and what triggers an increase
- Ask what happens to your data and your templates if you leave
Bring the file that breaks things
Do not send a clean audited statement in advance. Bring a genuine borrower package, ideally the one your team complains about: an operating company on an 1120S with a long depreciation schedule, a real estate entity on a 1065 with four K-1s, two guarantors with 1040s and Schedule E rentals, a distribution running from the operating company to one of them, and one year that arrived as a photograph taken at an angle.
Then watch, rather than ask. Watch how the product decides which document is which. Watch what it does with the photographed page. Watch how the distribution is treated in the combined coverage figure, and ask who decided that treatment. Every meaningful difference between these products appears on this file and on nothing simpler, which is exactly why vendors prefer to demonstrate on their own examples.
Four things this category systematically does not publish
Reading across the vendors in this market, the same four gaps recur, and knowing them in advance turns a surprise into an agenda item. Price is the obvious one: almost nobody publishes a figure, so plan an evaluation calendar around quote cycles rather than a comparison of published rates.
The other three are less obvious and more consequential. Product naming is unstable, with retired brands still circulating in shortlists and consultant decks. Customer evidence is frequently anonymous, especially among the community-focused specialists, several of which have decades of history and not one named reference. And global cash flow, which is the analysis a US community bank file actually turns on, is left unstated by two of the most-recommended platforms in the category.
| Gap | How it shows up | What to ask for |
|---|---|---|
| Pricing | Quote only, no published rates or bands | A written not-to-exceed figure before any pilot |
| Product naming | Retired brands still in circulation, capability URLs returning 404 | The current product name in writing, plus what support and release notes will call it |
| Customer evidence | Anonymous testimonials by asset band, logo walls, demo data styled as results | A named institution your size, in production, willing to take a call |
| Global cash flow | Not named at all, or referenced only as an abbreviation | The specific screen that produces combined debt service coverage, shown on your file |
Judge the review trail, not the accuracy claim
Accuracy figures in this market are vendor-measured and rarely arrive with a methodology. One vendor publishes a number above 99% for tax statement categorization, another claims above 97% extraction accuracy, and both are worth exactly as much as your own test on your own returns. Do the test, but do not let it decide the purchase.
What is verifiable, and what an examiner will ask about, is the review design. Can an analyst click a figure and see the source page it came from? Are overrides logged with who made them and when? Is there a validation step, and is it software or a person? One vendor in this category validates every extracted value with a human by design, which is slower and produces a much easier answer to the review question than any percentage does.
Then price the exception path, because it sets your staffing. If four in five returns extract cleanly, the remaining one in five determines how many analysts you need. Ask how a failed extraction is surfaced, who fixes it, how long that takes, and whether the correction teaches the system anything for next year.
- Can each spread figure be traced to its source document and page?
- Are overrides and manual corrections logged with attribution?
- Who validates an extracted value, and is that step optional?
- How is a failed extraction surfaced, and what is the rework path?
- What does the product produce for model documentation without being asked?
Frequently asked questions
How long should an evaluation take?
Plan on a quarter for a spreading-only purchase and longer if origination is in scope. Most of the elapsed time is quote cycles and reference calls rather than product testing. Running two vendors through the same borrower package in the same week is the single best use of that calendar.
Should we run a paid pilot?
Where a vendor has no named reference at your size, a paid pilot with written acceptance criteria is the sensible way to price the risk. Define what success looks like in advance, on your own files, and tie a payment milestone to it. Vendors confident in the product tend to agree readily.
How much weight should AI-generated shortlists carry?
Treat them as a starting point that is systematically incomplete. Our own visibility analysis found four verified, purchasable products that AI assistants almost never name, and several recommendations delivered under brand names their vendors have retired. Useful for discovery, unreliable for ordering.
What is the most common evaluation mistake?
Testing extraction and forgetting the roll-up. A product can spread every entity in a file beautifully and still leave your analyst combining them by hand in a spreadsheet, which is often the longer job. Insist on seeing the combined debt service figure produced inside the product.