Feed it the real dataroom — scans, spreadsheets, whatever you actually have. It extracts, reconciles, and flags what doesn't add up, and never lets an AI invent a figure you can't verify yourself.
A generic AI reads a document and gives you an answer. It doesn't verify that answer against the balance sheet, or notice the cash flow statement doesn't tie out. Alfie is built the other way around: nothing is trusted until it's checked.
No manual re-formatting first — the same messy documents your dataroom already has.
Drag in scans, PDFs, or the real BP Excel model — several at once.
OCR and parsing pull every line item, with the source page/cell kept alongside it.
Statements are tied out automatically — balance sheet, cash flow, gross profit bridges.
Breaks, missing documents, and aggressive BP assumptions surface as real, dated flags.
Chat with the deal - every answer grounded in the same verified data, checked before shown.
Once you invest, log capital sent and (soon) actuals vs. the plan you underwrote.
What you'd actually see on screen, at each stage of a real deal.
Scans, photographed statements, ten-tab Excel models, PDFs — upload several at once by drag-and-drop. Each one is parsed, OCR'd if needed, and tracked through to done or flagged if something needs a retry.
| management_accounts_fy25.pdf | done |
| BP_model_v6.xlsx | done |
| articles_of_association.pdf | processing |
| loan_agreement_scan.pdf | error · retry |
Balance sheet ties, cash roll-forwards, and cross-document mismatches are checked automatically on every upload. Nothing is a silent pass or a silent failure — you see exactly what was checked and what came back off.
Not 300+ raw spreadsheet rows. Pricing, volumes, marketing/CAC, retention — each theme expands into the individual assumptions behind it, and any one of them can be challenged directly.
Every answer is checked against the deal's own verified data before it's shown — including a citation back to where the figure actually lives. If it can't verify something, it says so instead of making it up.
Mark a deal as invested, then log capital sent in plain English - "$50,000 for the Q2 tranche." It's parsed into a verified, dated record automatically, and rolls up across every company in your portfolio.
| 2026-07-20 | 50,000 USD | "...sent $50,000 to the team...for the Q2 tranche" |
| 2026-05-02 | 120,000 USD | "...wired $120k at close..." |
Anyone who has to actually sign off on a set of numbers before money moves.
Move fast on a term sheet without skipping the reconciliation work a rushed deal usually loses.
Run the same rigorous checks across every target in a process, not just the one you have time for.
Get institutional-grade diligence without building out a full deal team.
Keep an internal, auditable record of exactly what was checked and how, for every acquisition considered.
The same discipline that keeps numbers honest applies to how your documents are handled.
Only members of your own workspace can ever see a deal - enforced at the database level, not just in the interface.
Your documents are analyzed to answer your request, full stop - not retained to improve anyone else's results.
Numbers are never taken on faith - the model is only ever asked to quote, and every quote is checked against the real document.
Delete a document, a deal, or your whole workspace at any time - it's actually gone, not just hidden.
No - it removes the slow, mechanical part (tying out statements, chasing down every BP assumption) so your team spends its time on judgment calls, not arithmetic. Every flag still needs a human decision.
That's the normal case, not the exception. Scanned and photographed documents go through OCR automatically; a bilingual, fully-scanned legal document with no embedded text is exactly the kind of file this has been tested against.
The model is never asked to state a number - only to quote a sentence or point at a cell, and that quote is checked against the real source before anything is shown to you. If it can't find a real source, it shows nothing rather than a guess.
Per deal, agreed directly with your team - not a self-serve subscription. Reach out and we'll figure out what makes sense for your deal flow.
Yes - invite teammates to a shared workspace, see the same flags and notes in real time, and control who can invite/remove others.
A 30-minute walkthrough with a real (anonymized) dataset, or your own if you'd rather.
Request access