Should I share my financial model as a file or in the room?

In the room, if the room renders it correctly. A downloaded copy leaves your control the moment it lands, and it goes stale the week after. Viewing in the room keeps the watermark on the page and the access in the audit trail. The catch is that a viewer which mangles merged headers or number formats is worse than a download, because the investor reads a wrong number and never tells you. Open your real workbook in any candidate room before you decide.

How do I share a cap table with one investor and not another?

Put each investor in their own permission group and release the cap table to the group rather than to a person. Two things to check. First, that moving an investor between groups is one action. Second, that AI features respect the same boundary, because an assistant that reads everything and filters afterwards will eventually answer an ownership question for someone who was not shown the cap table.

Does a data room make sense between rounds?

Yes, and this is where most founders leave value on the table. Existing investors get the monthly update, the current model and the board pack in one place with a record of who read them, and the next round starts with two years of material already organised rather than assembled in a panic. The economics change too: you are paying for a room that is open all year, so the billing unit matters more than the headline price.

What does a fundraising data room cost if it stays open all year?

Multiply the monthly price by twelve, then ask what grows. Storage tiers and per-page fees grow with exactly what an always-on room accumulates. Sifrsys publishes a free tier, Teams at $399/mo, Pro at $999/mo and custom Enterprise, with no per-page fees and a price-lock commitment of 90 days notice before any change. Get the equivalent commitment in writing from anyone else you shortlist.

Can investors ask the data room questions directly?

In an AI-native room, yes. The associate asks for gross margin by quarter or the change-of-control terms in the top contracts and gets an answer citing the exact page of the PDF or the exact cell range in the spreadsheet, formatted like Q3_Model.xlsx · Revenue!B14:F34. Retrieval is bounded by that participant's permissions at the database level, so an investor at an earlier disclosure stage gets an answer drawn only from what they have been shown.

Fundraising Data Room: A 2026 Buyer's Guide

What decides a fundraising data room. Rows are the four questions; columns are how investors actually receive material today.
Criterion	Email attachments	Deck-tracking tool	Legacy VDR	Sifrsys
Does the workbook render correctlyMerged headers, number formats, fills	Native Excel, but the file is gone	Often PDF only	Varies. Open your own model and check	Measured against LibreOffice ground truth. See the bars below
Cap table to one investor, not another	A new attachment and a new mistake	Link-level, rarely document-level	Permission groups	Permission groups, and the AI obeys them
Useful between rounds	Yes, and unauditable	Built for the raise window	Priced for the deal window	Same room, all year
Cost of a room open for two years	Free and expensive later	Per-seat	Not published. Storage tiers and quotes	Published: Teams $399/mo, Pro $999/mo, no per-page fees
Record of who read what	None	Opens and time on page	Full audit trail	Full audit trail of document access

What decides a fundraising data room. Rows are the four questions; columns are how investors actually receive material today.

Question one: does the workbook survive the trip?

This is the question nobody asks and everybody should. A fundraising room is mostly one spreadsheet and a folder of PDFs, and the spreadsheet is the document that decides the valuation. If the viewer drops a merged header, rounds a number format, or loses the fill that marks an assumption cell, the investor reads a different model from the one you built.

Sifrsys measures this rather than asserting it. The fidelity suite in tests/fidelity/ renders real workbooks in Chromium and compares them against LibreOffice ground truth across four producer dialects, with a committed baseline that fails the build on regression.

Display strings100%
Font styles100%
Alignment100%
Merged cells100%
KSA number formats100%
Fills99.7%

Spreadsheet rendering fidelity against LibreOffice ground truth, in real Chromium, across four producer dialects. Produced by the fidelity suite in tests/fidelity/.

If you raise from Gulf investors the last row matters more than it looks. Umm al-Qura Hijri dates and Arabic-Indic digits render correctly where LibreOffice, the ssf library and every competitor tested render them wrong. The [$-2060401] format used to produce a US dollar sign against Arabic-Indic digits. That is fixed.

Question two: who can see the cap table?

The cap table is the document with the most asymmetric consequences. A prospective investor seeing the terms you gave the last round changes the negotiation. An existing investor seeing the new round's structure before you have told them changes the relationship. It is also the document most likely to be attached to an email at 1am.

The mechanic to insist on is release to a group rather than to a person, so that promoting an investor is one action. The question to press on is what the AI can reach. If the assistant retrieves from everything and filters the answer afterwards, your ownership structure is one well-phrased question away from an investor who was never shown it. In Sifrsys the tenant and participant filter runs inside the SQL query, so a chunk outside the participant's permission is never selected. That exclusion is structural, and CI asserts it against the deployed function's source in pg_proc.prosrc on every push. The measured half is the failure that prompted the lockdown: before July 2026 an unrelated signed-in user retrieved 30 chunks across 13 documents of verbatim deal text, because the caller controlled the participant identifier the filter was checking.

Chunks of verbatim deal text, across 13 documents, that an unrelated signed-in user retrieved before the July 2026 lockdown

Measured. The filter itself is structural: asserted against pg_proc.prosrc in CI, with a denial suite that counts only SQLSTATE 42501 as a pass.

Row and column

Citations resolve to absolute sheet rows and column ranges, then wash those cells in the workbook

Product behaviour

Streaming

Purpose-built OOXML parser. The maintained libraries either drop cell styles in streaming mode or hold the whole workbook in memory to keep them.

infrastructure/hetzner/worker/src/xlsx-model/

Question three: does it stay useful between rounds?

Most founders treat the room as a thing that opens for a raise and closes after. That throws away the compounding part. Investors who read the monthly update in the room leave a record. Questions answered once in the room stay answered. And the next raise starts with two years of organised material instead of a week of archaeology.

Monthly update posted
Narrative, metrics, current model
Investors read in the room
Watermarked, logged
Questions answered once
With a citation, not an email
Next round opens
History already in place

The between-rounds cycle. Each pass leaves the next raise less work.

This changes the shortlist. A tool priced and shaped for a deal window is the wrong shape for a room that is open all year. So is a per-seat model, once you are inviting every angel on the last SAFE. What you want is a flat price you can put in the budget and a room that does not punish you for adding documents.

Question four: what does the always-on room cost?

Multiply by twelve, then ask what grows. Checked on 4 August 2026, none of the major incumbents publish a price. Datasite states pricing is customised for every transaction. Firmex offers a single-project room or an annual subscription and says the volume of data determines the price, behind a quote request. iDeals names Core, Premier and Enterprise plans, each behind a "Get price" button. Ansarada runs a quote builder and bills once the room goes live or 90 days after creation, whichever comes first. Intralinks asks you to request a quote. For a room that is open for years rather than months, an unpriced variable is a budgeting problem, not a negotiating one.

Sifrsys publishes: a free tier, Teams at $399/mo, Pro at $999/mo, custom Enterprise, no per-page fees, and 90 days notice before any price change. AI chat with page-level citations is in every paid tier rather than sold as a module. The full breakdown is on the pricing page.

The recommendation

This is one of the two segments where Sifrsys is the straightforward answer rather than a qualified one. The work is a workbook, a cap table and a recurring update, shared with a changing set of investors over years, and the four questions above are the ones it was built to answer. The honest caveat is the charts and pivot tables gap: test your own model first.

If you are mid-raise right now and the pressing problem is speed rather than cadence, the startup guide is the shorter path. If your investors are funds who will send this material into their own diligence process, the private equity guide explains what they are doing with it. And the founder walkthrough covers what belongs in the room stage by stage.

The first ten companies get different terms

Sifrsys is new, and a new vendor holding your cap table owes you more than a discount. The first ten paying companies get three commitments in writing: the price they sign at for the life of the subscription, migration of their documents, folder structure and permission setup out of their current data room at no charge, and a direct line to the founder with answers the same business day. Ten companies. When the spots are gone this section comes down. Terms are on the pricing page.

Sources

Competitor facts come from each vendor's own pricing pages and were checked on 4 August 2026: datasite.com/en/products, firmex.com/pricing, idealsvdr.com, ansarada.com/pricing, intralinks.com. Where a vendor does not publish a figure, this guide says so rather than estimating one. Fidelity and parser numbers are produced by this codebase and fail the build on regression.