Data Room Latency

Overview · Methodology

How are the sign-in times measured and the vendor statements collected?

Sign-in time follows protocol v1, written on 2026-11-08 before any data was collected and not changed afterwards: 4 rounds, six continents, every redirect followed. Vendor statements come only from pages each vendor controls, about 18 page requests per vendor in the first reading, with the same 20 statement fields for all 25 vendors, plus App Store apps and the sign-in viewport (22 fields in the CSV), checked word for word.

Which vendors are included

  • The list of 25 virtual data room vendors and each vendor's official website were fixed before any vendor page was read, and reused from an earlier vendor dataset kept by the same compiler. It is not a complete list of the market; a vendor can ask to be added through Submit a request.
  • No vendor pays to be listed.

Sign-in time, protocol v1

  • Vendors: the 19 of 25 vendors whose homepage links to a sign-in page. The entry is that link, fixed when the protocol was written.
  • Rounds: 4, on 2026-11-08, 2026-11-11, 2026-11-15, 2026-11-19, at different times of day. At the start of each round the redirect chain from the entry to the sign-in page is resolved with an ordinary browser identity, and every step is timed.
  • Probes: Globalping, a public network of measurement probes. Each round has two halves; in each half the network itself picks one probe per continent through a neutral test site (example.com), never by measuring a vendor, and every vendor in that half uses those probes. In round 1 both halves were given the same six probes, so each continent was one city.
  • Timing: for each step, DNS lookup, TCP connection, TLS handshake and time to first byte; when a step stays on the same server as the previous one, only the first byte counts. A probe's figure for a vendor is the sum over steps.
  • Counting: a probe counts only if every step answered with a status below 400. A 4xx or 5xx answer is counted as blocked (usually bot protection refusing data-center addresses) and not timed; a failed or timed-out request is counted as failed; an offline probe is left out.
  • Summary: per round, each continent's figure is the median of its probes and the mean of six continents is their equal-weight average, given only when every continent has a figure. The final figure is the median of the round means across rounds and needs at least 3 rounds.
  • Choice of weighting: equal continent weights were chosen after a pilot on 2026-11-07 that timed the same 19 vendors; the pilot's data is not used here, and the weighting has not changed since protocol v1 was written. Every continent is shown so readers can weight them differently.
  • Verification: every measurement has a public ID that opens on Globalping, listed with the address requested in signin_measurements.csv.
  • Limits: this is the public sign-in page from data-center probes, not opening documents inside a room. Requests use HTTP/1.1; browsers usually use HTTP/2. Redirects were resolved from one location, so a vendor that redirects by region may follow another path elsewhere.

Vendor statements

  • Fields fixed before any vendor page was read: phones and tablets, browsers and plug-ins, largest file, ways to upload, access levels, how permissions are set, interface languages, mobile apps.
  • Support hours and languages were added as a field on 2026-11-08, after the first reading. The definition and the reading budget were written before any page was read for it: the pages already saved, plus at most three support, contact or help-centre pages per vendor, picked from links on the vendor's own saved pages in the same order for every vendor and read once.
  • Keeping bidder groups apart and an NDA or terms of access before opening documents were added as fields on 2026-11-09, after the first reading, because the site did not answer buyer questions about several bidders in one room or about NDAs. Both definitions were written before the saved pages were searched for them, the same fixed list of search terms was run on every vendor's saved pages and each hit was read in context; no new page was read. For bidder groups, a quote counts only when it says other groups or buyers cannot see something, or that it is hidden or isolated; this rule was tightened after the first pass on 2026-11-09 and every vendor was re-judged under it. For NDAs, a vendor whose pages only name the feature, without saying viewers accept it before access, is shown apart.
  • Where documents are stored and how many users or viewers a room allows were added as fields on 2026-11-09, after the first reading, with the same method: definition and search terms written first, the same saved pages for every vendor, every hit read in context, and the strict reading that a quote counts only when it says the thing itself.
  • A topic pass on 2026-11-09 read, for every vendor, at most two more security or storage pages and two more Q&A or groups pages, picked from links on the vendor's own saved pages by one rule written before any of them was fetched (help and docs articles first). Where documents are stored and keeping bidder groups apart were searched again over all saved pages with the same terms plus a few added for every vendor. It was run after a reader pointed out that one vendor's docs covered both topics on pages the first reading had not saved; the first reading had saved security and Q&A pages only by chance, so the pass was run for every vendor, not for that one.
  • Data rooms per plan, a free plan or free trial, and what is recorded about each viewer were added as fields late on 2026-11-09, after the GTM evaluation found buyer questions on these topics unanswered, with the same method: definition and search terms written first, every vendor's saved pages (first reading, support pages and topic pass), every hit read in context, and the strict reading. Where a vendor's own pages disagree, both quotes are shown and the narrower reading is used.
  • Deal and use types were added on 2026-11-13 with the same method, from a fixed list of twelve types written before the search, after the evaluation found that queries asking by deal type had no page here. A type counts only when the vendor says it for its own product; deal size is not recorded. The search words for each type were fixed before the search. Each hit was then read in context under one rule, also fixed before the search: a type counts only when the vendor's own words name its product for that type. Under that rule a bare 'Startups' or 'Legal' label, and 'M&A', 'due diligence', 'fundraising' or 'capital raising' with no side, startup or fund wording, were not counted; an industry label counts for Real estate and Life sciences.
  • Sharing one document as a tracked link was added on 2026-11-13 with the same method, after the evaluation found that a fundraising query (alternatives for sharing a pitch deck) had no page here. Both parts must be stated: one file shared through its own link, and tracking of that link.
  • Published prices were added as a field on 2026-11-09 with the same method. A price counts only when the vendor's own saved page ties an amount to a plan the page itself calls a data room plan; the lowest such price is quoted with its unit. Prices are listed, never compared or added up.
  • Document translation was added as a field on 2026-11-08, after the first reading. Its definition was written before the saved pages were searched for it, and it is answered from the same saved pages for every vendor; no new page was read for it.
  • Sources: only pages the vendor controls, including a help center on its own domain or linked from its website, read in the same order on 2026-11-08, plus the topic pass on 2026-11-09 described above. The first reading allowed about 18 page requests per vendor; two vendors went over it: Ansarada 20 requests, 20 answered; DocSend 22 requests, 16 answered. Quotes from their extra pages are shown like any other. Fields added after the first reading used the same saved pages, plus the topic pass of 2026-11-09 and at most three support, contact or help-centre pages per vendor for support; the rule for picking those pages was tightened once before any of them was read (help articles left out, links taken from every saved page because four homepages build their links with JavaScript). Never review sites, comparison sites, news or another vendor's site.
  • Every quote is an exact substring of the saved text of the page it cites; this is checked by script before every build. Wording is never paraphrased or upgraded, and a vendor's own superlatives stay.
  • A field a vendor's pages do not address reads "Not stated on the pages read"; when the pages that would answer a field refused automated reading, that field reads "Pages could not be read". This is decided field by field, so the number of vendors whose pages could not be read differs between questions (Clinked's help center could not be read, its other pages could). Neither says whether the product offers something.
  • Profile links in each vendor page's structured data come from the vendor's own homepage, read on 2026-11-08: profile pages on LinkedIn, X, YouTube, Facebook, Instagram, GitHub, Crunchbase, Wikipedia, Wikidata, G2 and Capterra that the homepage links to or lists in its own structured data, collected the same way for every vendor. Handles that do not carry the vendor's name are left out, and a vendor whose homepage gives none has none.
  • Clicks are not counted and ease of use is not rated: that needs an account at every vendor and the same task run in each.
  • App Store: Apple's App Store public search in six storefronts (US, UK, Australia, Germany, Poland, Singapore) on 2026-11-08, kept only when the seller is the vendor or its parent. Apps the seller describes only as board-meeting apps are listed but not counted. Interface languages are counted only when named; a changelog that lists added languages, a how-to answer naming one language, or a statement about a mobile app only is shown as such and not read as a full list, and local names such as Español or Deutsch count as Spanish or German. Mobile viewport: the sign-in page's HTML read with an iPhone browser identity.

Corrections

An entry changes only when a page the vendor itself publishes says something different, under the same rule for all 25 vendors, and every change is listed in the update log. To point out an error, use Submit a request.