Short answer: four different categories of tool get called AI due diligence, and they solve different problems. Document intelligence such as Hebbia, DiligentIQ and V7 Go reads a data room and answers questions with citations, but reconciles nothing. Spreading and tie-out such as Daloopa and DataSnipper structures filings and ties numbers back to source, but does not test whether the statements are true. AI quality of earnings platforms such as Finsider, Keye, Termina and Shepi work from ledger exports or from documents authored to be read. Cash reconstruction rebuilds cash from raw scanned bank statements and reconciles it to the ledger, the income statement and the tax return. That is where QoEAgent builds. Every tool in the first three categories assumes structured data already exists, an assumption that holds on a fifty million dollar deal and breaks below ten million.
“AI due diligence” describes at least four different products that solve different problems. Teams buy the wrong one regularly, because the marketing language is nearly identical across all of them.
Here is the map as we see it. We build in one of these categories and we will say so plainly, but the point of this page is the taxonomy, not the pitch.
1. Document intelligence
What it does: reads a large document set and answers questions against it with citations. Point it at a data room and ask what the change-of-control provisions say across forty contracts.
Representative tools: Hebbia, DiligentIQ, V7 Go, and general-purpose document AI.
Buy it when: your bottleneck is reading volume — a full data room, a long contract stack, a regulatory file. It compresses the first read from days to hours.
It will not: reconcile anything. It tells you what documents say, not whether the numbers in them are real.
2. Spreading and tie-out
What it does: pulls figures out of filings, PDFs and presentations into a workbook, with links back to the source page.
Representative tools: Daloopa for spreading historical financials from filings; DataSnipper as an Excel add-in that cross-references numbers in PDFs to a workbook and has become standard in audit practice.
Buy it when: you are building models from public filings, or you are an audit or diligence team that needs tie-out evidence inside Excel.
It will not: tell you whether the statements being spread reflect the cash that actually moved.
3. AI quality of earnings platforms
What it does: takes accounting-system or document data and produces diligence outputs — EBITDA bridges, working capital analysis, cohort and concentration views, adjustment suggestions.
Representative tools: Finsider.ai, which connects directly to accounting systems such as QuickBooks, Xero and NetSuite; Keye, which works from documents and the CIM and emphasizes traceable Excel output; Termina, which structures target financials and data-room documents for early review; and Shepi, which offers self-service QoE analysis with published per-deal pricing.
Buy it when: the target runs a reasonably clean ledger you can access, and your constraint is turning that ledger into deal outputs quickly.
The limit to check: what happens when there is no clean ledger to connect to. Direct accounting-system integration is genuinely powerful when the target is on a cloud system and will grant access. In the lower middle market that is frequently not the situation.
4. Cash reconstruction
What it does: rebuilds cash receipts and disbursements from raw bank statements — including scans — and reconciles them transaction by transaction to the general ledger, the income statement and the tax return.
This is where we build. It is a narrower job than the category above and a different technical problem: extraction and layout reconstruction from documents that were never meant to be machine-read, then matching that survives transfers, owner activity, merchant settlement lag and cut-off.
Buy it when: the records arrive as PDFs somebody scanned, there are more bank accounts than anyone expected, and the question you actually need answered is whether reported earnings match the money that moved.
It will not: read your data room or build your model. It answers one question thoroughly.
The distinction that gets missed
Three of these four categories start from data that is already structured, or from documents that were authored to be read — filings, CIMs, contracts, ledger exports. That is a reasonable assumption for a $50 million transaction.
It is often the wrong assumption below that. A $4 million business is one owner, a bookkeeper, six bank accounts opened at different times, statements that are photocopies, and a CPA who posts accruals once in December. There is no clean ledger to connect to, and the document intelligence tools will faithfully summarize numbers that nobody has verified.
That is not a criticism of those tools. It is a statement about which problem you have.
A practical way to choose
Ask where your hours actually go on a typical deal.
Reading — document intelligence. Spreading — a spreading or tie-out tool. Producing deal outputs from a clean ledger — an AI QoE platform. Reconstructing what the cash actually did from bad source documents — cash reconstruction.
Then run one closed deal you know the problems in through whichever you are considering, and score it on the seven tests: traceability, real-document handling, exception quality rather than headline match rate, transfer logic, monthly rather than cumulative reconciliation, deliverable integration, and human review controls.
For the record
We publish our pricing — $10,000 flat for a full proof of cash plus GAAP-formatted financials, returned within 24 and 48 hours of ingestion. On a recent engagement that meant 431 manually scanned statements across 36 bank accounts and 29,823 transactions at a 99.4% average match rate, with a $32 variance left flagged for a person.
What we do not do: read your data room, build your model, or decide whether an add-back is defensible. Those are other people’s jobs and, in the last case, a human’s. More on that line in what AI quality of earnings actually automates.
Tool descriptions above are drawn from each vendor’s published materials and from third-party buyer’s guides, current as of September 2026. Capabilities change; verify against the vendor before relying on any of it. We have no commercial relationship with any tool named here.