AI-assisted due diligence

Every number, traced back to the page it came from.

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.

0
Figures invented by the AI
100%
Numbers traced to a source
3,159
Line items, largest deal reconciled
24/7
Reviewing new documents
Why this matters

Most diligence tools summarize. Few actually check the arithmetic.

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.

Generic AI
Reads document
States an answer
Alfie
Reads document
Reconciles it
Verified + cited
80%
of deals have at least one reconciliation break a first read misses
32
distinct BP assumptions extracted, grouped into themes automatically
2–4h
a senior analyst spends manually tying out one set of statements
1
place your whole team sees every flag, note, and open question
How it works

From raw upload to a checked, challenged deal.

No manual re-formatting first — the same messy documents your dataroom already has.

Upload

Drag in scans, PDFs, or the real BP Excel model — several at once.

Extract

OCR and parsing pull every line item, with the source page/cell kept alongside it.

Reconcile

Statements are tied out automatically — balance sheet, cash flow, gross profit bridges.

Flag

Breaks, missing documents, and aggressive BP assumptions surface as real, dated flags.

Ask

Chat with the deal - every answer grounded in the same verified data, checked before shown.

Track

Once you invest, log capital sent and (soon) actuals vs. the plan you underwrote.

See it in action

A full diligence desk, not a chatbot.

What you'd actually see on screen, at each stage of a real deal.

01 — Documents

Drop in the whole dataroom, not one file at a time.

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.

Drag and drop documents here, or browse
management_accounts_fy25.pdfdone
BP_model_v6.xlsxdone
articles_of_association.pdfprocessing
loan_agreement_scan.pdferror · retry
02 — Flags

Every break, ranked, with a plain-English reason.

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.

Reconciliation breakCash roll-forward off by 18,200 between FY24 and FY25 statements.
Missing documentBalance-sheet liability has no supporting loan agreement on file.
Tied outGross profit bridge matches across all three statements.
03 — BP Assumptions

32 real drivers, grouped into themes an investor actually scans.

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.

Pricing & Commissions6 assumptions · 95 data points
Interchange Rate — Classic Card
Monthly AUM Fee Rate
Monthly Interest Rate
Volumes18 assumptions · 265 data points
Marketing & Acquisition4 assumptions · 20 data points
04 — Chat

Ask it anything about the deal. It won't guess.

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.

What was revenue in Y3 of the plan, and does it tie to the model?
Y3 revenue is 2,140,000 across all four streams. This matches the BP model's own total - no discrepancy found.Source: BP_model_v6.xlsx, Revenue tab
05 — Portfolio tracking

Once you invest, keep a real ledger without a spreadsheet.

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.

2026-07-2050,000 USD"...sent $50,000 to the team...for the Q2 tranche"
2026-05-02120,000 USD"...wired $120k at close..."
Who it's for

Built around a real diligence workflow, not a generic AI chat.

Anyone who has to actually sign off on a set of numbers before money moves.

Venture & growth funds

Move fast on a term sheet without skipping the reconciliation work a rushed deal usually loses.

Private equity

Run the same rigorous checks across every target in a process, not just the one you have time for.

Family offices

Get institutional-grade diligence without building out a full deal team.

Corporate development

Keep an internal, auditable record of exactly what was checked and how, for every acquisition considered.

Trust & security

Your dataroom is not training data.

The same discipline that keeps numbers honest applies to how your documents are handled.

Org-scoped by design

Only members of your own workspace can ever see a deal - enforced at the database level, not just in the interface.

Never used to train models

Your documents are analyzed to answer your request, full stop - not retained to improve anyone else's results.

Every figure has a citation

Numbers are never taken on faith - the model is only ever asked to quote, and every quote is checked against the real document.

You control retention

Delete a document, a deal, or your whole workspace at any time - it's actually gone, not just hidden.

FAQ

Before you ask.

Does this replace my analysts?+

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.

What if my documents are messy - scans, photos, inconsistent formatting?+

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.

How does it avoid just making numbers up, the way other AI tools do?+

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.

How is this priced?+

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.

Can my whole team use it together?+

Yes - invite teammates to a shared workspace, see the same flags and notes in real time, and control who can invite/remove others.

Get started

Bring your next deal. We'll show you what it finds.

A 30-minute walkthrough with a real (anonymized) dataset, or your own if you'd rather.

Request access