Permissioned deal rooms, Q&A, and buyer administration. The seller owns the documents. Mid-deal switching can bind; that friction decays after close.
As of 19 August 2026. This memo covers permissioned sell-side data rooms, buyer access, and Q&A during an M&A process. No live-product test has been run.
The seller owns the documents. Lock-in is timed to the live deal.
Datasite is process infrastructure for a permissioned room. The seller owns the files. Hosting is not exclusive ownership. Moving a live room, with buyer permissions and Q&A history, can bind. After close that friction decays. It does not by itself price the next process or decide deal outcome.
Security, watermarking, and access logs are why unmanaged email is the wrong comparator. The overlap questions are Intralinks, Firmex, and DealRoom — named peers, not scores. Those security claims need an audit. AI extraction and Q&A drafting attack review labor, not permissioned hosting. Workflow capability and operating economics stay open without an independent product test.
Exposed vs scarce. The seller owns the files. AI review labor is an overlay. Permissioned hosting is category infrastructure, not a document asset.
Still untested. Setup time, permission error, and Q&A cycle versus named VDRs on a frozen room.
The comparison that would change the call. The same room and Q&A set on Intralinks, Firmex, or DealRoom — not versus unmanaged email.
Decision questions
On a frozen room and Q&A set, does Datasite clear a prespecified admin and response gate relative to the current process?
What isolated increment comes from permissioning, Q&A, or reporting, each changed one at a time?
What is full cost per live process, including admin labor and hosting, not the room quote?
What is the observed cost of moving a live room, if it can be moved at all?
Permission errors, Q&A delays, broken watermarking, and failed buyer access are different classes. They stay separate. A correct document with the wrong permission is a different error than a missing document with a good index.
Published benchmarks attach to the overlay on review labor. ParseBench bounds whether room documents parse. CUAD bounds extractive clause location. TAT-QA bounds a slice of financial Q&A. They do not measure permissioning, watermarking, buyer access, or Q&A cycle time. Those remain the scored job.
Workflow breakdown
Each step of the job is mapped to a published benchmark or another reputable quantitative measure only where the mapping is honest. A public figure is a floor or a labor-market context on that task type. It is never a score of this product.
A slice of drafted Q&A. Not Q&A routing or cycle time.
Permissioned hosting, buyer access, and live Q&A admin
None
No honest public floor
The scored job. No public pack measures a data room.
Findings at a glance
Capability, cost, and defensibility are assessed separately. They are not averaged into a score. Where evidence is missing, the finding stays open.
Status
Area
Finding
Open
Workflow capability
Hosting documents is not extraction quality and not time-to-close. No frozen room-admin or Q&A set is in this report.
Open
Operating economics
Page hosting, admin labor, and full cost per live process are unmeasured. A deal-room quote is a commercial term.
Judgment
Defensibility
The seller owns the documents. Mid-deal switching can bind. After close that friction decays. Security claims need an audit. AI extraction is an adjacent product, not a VDR substitute.
Vendor-stated
Vendor claims and open questions
Public materials describe virtual data rooms, Q&A, and deal administration for M&A. Those are vendor declarations.
Workflow capability
Hosting documents is not admin performance
A permissioned room, Q&A, and buyer list describe an input surface. They do not describe setup time, permission error rate, or Q&A cycle time on a frozen process. No independent Datasite test has been run.
A room can be the market default and still fail the cases that matter: a buyer seeing the wrong folder, a Q&A item that never routes, or an index that does not match the files. Those classes stay apart.
Published benchmarks
ParseBench, CUAD, and TAT-QA bound parse, contract extraction, and a slice of financial Q&A. They do not measure permissioning, watermarking, buyer access, or Q&A cycle time.
ParseBench
Room documents fail first at parse. Five official dimensions stay separate. Parse quality is not permissioning or Q&A cycle time.
Third-party leaderboard · not an Antidote run · not a product score
Current frontier VLMs on ParseBench
Official ParseBench leaderboard, observed 13 August 2026. Five dimensions kept separate. Gemini 3 Flash is Thinking High. GPT-5.6 Sol and Fable 5 are Reasoning None / default API settings in the source.
Dimension
Gemini 3 Flash
GPT-5.6 Sol
Claude Fable 5
Claude Opus 4.8
Tables
91.5%
89.34%
89.79%
89.65%
Charts
64.79%
59.57%
52.21%
49.75%
Faithfulness
90.87%
86.17%
90.02%
89.02%
Formatting
68.31%
50.42%
72.62%
71.38%
Grounding
59.77%
25.09%
49.24%
18.69%
Third-party leaderboard · not an Antidote run · not a product score
Specialist parsers vs a frontier VLM
Same ParseBench leaderboard. LlamaParse Agentic, Pulse Ultra 2, and Reducto Agentic against Gemini 3 Flash Thinking High. The source Overall column is omitted.
Dimension
LlamaParse Agentic
Gemini 3 Flash
Pulse Ultra 2
Reducto Agentic
Tables
90.74%
91.5%
75.45%
80.42%
Charts
78.11%
64.79%
90.82%
73.4%
Faithfulness
89.68%
90.87%
79.49%
86.37%
Formatting
85.24%
68.31%
73.05%
57.6%
Grounding
80.62%
59.77%
66.56%
67.07%
CUAD
Diligence rooms are full of contracts. CUAD bounds extractive clause location. It does not measure buyer access or watermarking.
Paper table · Hendrycks et al., 2021 · not an Antidote run · not a product score
CUAD precision-recall, selected models
Hendrycks et al., 2021, Table 2. AUPR and precision at 80% recall are different estimands and are not averaged.
System
AUPR
Precision @ 80% recall
Precision @ 90% recall
DeBERTa-xlarge
47.8%
44%
17.8%
RoBERTa-large
48.2%
38.1%
0%
RoBERTa-base
42.6%
31.1%
0%
ALBERT-xxlarge
38.4%
31%
0%
BERT-base
32.4%
8.2%
0%
TAT-QA
Hybrid table-and-text QA bounds a slice of buyer Q&A on financial exhibits. It is not room-admin performance.
Paper table · Zhu et al., 2021 · not an Antidote run · not a product score
TAT-QA test exact match and F1
Zhu et al., ACL 2021, Table 3, test split. TAGOP is the paper system. Human expert is 50 hybrid contexts (301 questions).
System
Exact match
F1
Human expert
84.1%
90.8%
TAGOP
50.1%
58%
NumNet+ V2
37%
46.9%
TaPas WTQ
16.6%
22.8%
BERT-RC
9.1%
18.7%
HyBrider
6.3%
7.5%
Paper table · Zhu et al., 2021 · not an Antidote run · not a product score
TAGOP by answer type on the test set
Zhu et al., ACL 2021, Table 4. Arithmetic is the weak slice. Metrics stay separate; there is no composite.
Answer type
Exact match
F1
Span
54.1%
67.9%
Spans
60%
75.1%
Counting
62.5%
62.5%
Arithmetic
42.5%
42.5%
How to read the published benchmarks
ParseBench, CUAD, and TAT-QA bound parse, contract extraction, and a slice of financial Q&A — the labor that AI extraction and drafted Q&A attack. They do not measure a permissioned room. Hosting, watermarking, and buyer access stay a process test. These numbers do not become time-to-close.
Current systems, cost, and harness
Official list prices observed 8 August 2026 bound the model bill only. They are not a product subscription and not full cost per accepted output. ParseBench cents per page is a parser-layer cost. Full cost still has to include failed attempts and reviewer minutes.
ParseBench tables vs cents per page
ParseBench official leaderboard, observed 13 August 2026. Horizontal axis is list cents per page. Vertical axis is the Tables metric (GTRM), not an overall score.
Named VDRs are the comparison, not email
The failure mode that matters for Datasite is comparing the room only to unmanaged email. That overstates the increment. The overlap questions are Intralinks, Firmex, and DealRoom on the same room contents and Q&A set.
If admin traces cannot be exported, report outcome and observed price only.
Operating economics
A room quote is not full cost per process
Hosting price is a commercial term. Full cost still has to include banker and admin hours, buyer friction, and failed access events, with those failures in the denominator.
No Datasite all-attempt cost is in this report. AI that drafts Q&A or indexes rooms compresses admin minutes. It does not retire the permissioned room. Any time-to-close story remains modeled and is out of scope besides: close timing has many causes.
Defensibility
Mid-deal lock-in is not a franchise. The documents are not Datasite's.
Room contents are the seller's. Direct substitution by Intralinks, Firmex, or DealRoom is ordinary after close. Moving a live room, with buyer permissions and Q&A history, is the reconstruction that can bind. After close that friction decays. It does not by itself establish pricing power on the next process.
AI extraction and Q&A drafting are adjacent products. They attack review labor, not permissioned hosting. Security, watermarking, and access logs are why unmanaged email is the wrong comparator. Overlap versus Intralinks, Firmex, and DealRoom is the open question, not a score. Those claims need an audit, not a brochure. Platform bundling by banks, cloud storage, and e-signature vendors is the other exposure.
Defensibility
A virtual data room is process infrastructure. The seller owns the documents. Mid-deal switching can bind. That is not a post-close data moat, and it is not an extraction product.
Dimension
Rating
Judgment
Missing evidence
Data rights
Exposed
Room contents are the seller's. Datasite hosts permissioned access. Hosting rights are not exclusive ownership of the deal corpus.
Hosting versus ownership terms and post-close retention.
Exclusivity
Exposed
Permissioning, watermarking, Q&A, and buyer administration are category features shared with other VDRs. There is no exclusive document corpus.
What, if anything, cannot be reproduced in Intralinks, Firmex, or a generic VDR.
Rebuild difficulty
Conditional
Standing up a room on another VDR is ordinary. Moving a live room mid-process, with buyer permissions and Q&A history, is the reconstruction that can bind. After close, that friction decays.
Mid-process migration time, buyer friction, and post-close archive portability.
Freshness
Open
Datasite says project interaction reinforcement stays within the originating project while anonymized aggregate attributes may support product development. Index-to-file consistency and Q&A routing freshness remain unmeasured.
Index mismatch rate and Q&A routing completeness.
Switching cost
Conditional
Buyer lists, Q&A threads, and permission trees embed the process in the room. That switching cost, if observed, is timed to the live deal. It does not by itself price the next deal or create a franchise.
A mid-process switch observation and post-close re-compete evidence.
System of record
Conditional
During the process the room is the permissioned source of documents for buyers. It is not the company's system of record, and it does not decide deal outcome.
Conflict map versus company repositories and banker file stores.
Integrations
Conditional
CRM, e-signature, and reporting connectors can add gravity during the process. Extraction and AI review tools sitting on top of the room are adjacent products, not VDR substitutes.
Integration inventory and whether traces export to those tools.
Network effects
Exposed
A room does not get better because another deal used Datasite, except via buyer familiarity. Familiarity is distribution, not a data network effect.
Buyer-side preference evidence, separately from seller-admin quality.
Trust and compliance
Conditional
Security, watermarking, and access logs are why a VDR exists instead of a file share. Those claims need an audit. They are the plausible reason a generic share is not the comparator.
Permission audit, watermarking tests, and incident history.
Distribution
Conditional
Banker and advisor channel access can win rooms. It does not block Intralinks, Firmex, DealRoom, or a bank's preferred vendor list.
Win/loss versus named VDRs, not versus email.
Services dependence
Open
Datasite's published AI Q&A workflow requires deal-team review and approval before an answer reaches buyers. The customer-side exception layer is therefore explicit; vendor services dependence remains unmeasured.
Admin hours and professional-services hours per live process.
Rate of improvement
Open
No measured setup-time or Q&A-cycle improvement curve is in this report.
Dated admin and Q&A cycle series on comparable processes.
How AI can compete
Replacement, a new AI product, a platform bundle, cheaper labor, and a customer building it themselves are different risks. They are not averaged into a score.
Mechanism
Assessment
Direct replacement
Another VDR, or a permissioned share for a simple process, can host the same files. The increment is admin, permissioning, and Q&A machinery, not the documents.
New AI product
Document-intelligence products attack extraction and Q&A drafting, not permissioned hosting. Treating Datasite as a parser or a model surface is the wrong object.
Platform bundle
Banks, cloud storage, and e-signature platforms can attach room features to existing deals infrastructure.
Labor compression
AI can draft Q&A and index rooms faster. That compresses admin minutes. It does not retire the need for a permissioned room during a live process.
Customer builds it
A bank or corporate-development team can self-host a permissioned share for a simple process. Audit, watermarking, and buyer familiarity are why they usually do not. Insourcing mid-deal is the least attractive moment to try.
Data flywheel
A flywheel needs every step below. On this evidence it is unproven. Having data is not the same as being allowed to reuse it, and reuse is not the same as a measured quality gain.
Link
Status
Note
Unique data
Vendor-stated
Deal exhaust is the seller's. Datasite's exhaust is admin telemetry.
Permission to reuse
Open
Datasite says nonarchived project documents are deleted after termination and not used to train external models. Anonymized aggregate attributes may support product development, project reinforcement stays local, and archive or program terms may preserve broader derived records.
Reliable feedback
Open
No audited permission-error or Q&A labels.
Proven quality gain
Open
No paired generic-VDR result.
Better outcomes
Open
Deal outcome is out of scope. Admin and Q&A gates are unmeasured.
More usage or data
Open
Repeat-room retention is unobserved and is not a document flywheel.
What public sources resolve
Public filings and vendor documentation can settle some structural questions. They cannot replace a test on the live product.
Question
Conclusion
Still open
Can deal data feed product improvement across projects?
Datasite says nonarchived raw project documents are deleted after termination and not used to train external models. It retains anonymized aggregate attributes for product development, keeps interaction reinforcement within the originating project, and offers an archive that can preserve deal history.Artificial Intelligence at Datasite; Terms of Use; Datasite Archive
Executed terms, anonymization validation, deletion evidence, and a measured product-quality increment from retained aggregate attributes.
Reviewer minutes, acceptance-without-edit rate, response time, and a paired generic-share workflow.
Does post-close archive reduce or increase switching cost?
Archive persistence increases governance value and historical embed, while published download and sharing features imply some exit path. Full-fidelity portability remains untested.Datasite Archive; Terms of Use
An export-and-import test covering Q&A, permissions, bidder groups, versions, audit trails, indexes, IRM, and post-close legal records.
Required comparisons
A decision-grade test would run these side by side. None has been run here.
Comparison
Note
Current workflow
Requires imported room-admin and Q&A process from a sanctioned observation.
Shared-folder floor
Permissioned file share and email Q&A without a VDR product.
Named competitor
Declared Intralinks, Firmex, or DealRoom room with the same permissioning budget.
This product
Needs authorized Datasite access and exportable admin traces.
Isolated changes
Permissioning, Q&A, and reporting must be isolated before attribution.
Open claims
Each claim stands on its own. They are not averaged into a moat score. Admin outcome and price wait on a live-product test. Deal outcome is not the room product's result. The seller owns the files. Mid-deal switching cost is not a post-close franchise.
Claim
Status
What is still missing
A live process cannot switch rooms without material delay.
Open
Migration time, buyer friction, and a mid-process switch observation.
Q&A cycle time is shorter than a generic share.
Open
Frozen Q&A items, response time, and a paired named competitor.
Datasite publishes security-control and third-party assurance statements.
Open
Captured assurance evidence, current certificate scope, target-room configuration, permission and watermarking tests, and incident history.
Room contents are the seller's, not Datasite's exclusive corpus.
Judgment
Hosting versus ownership terms and post-close retention.
Room software determines deal outcome.
Open
That claim is out of scope. Time-to-close has many causes.
What this report does not conclude
Replacement of the current room-admin process
Q&A cycle time versus Intralinks, Firmex, or DealRoom
Full cost per live process
A document moat; the seller owns the files
Deal outcome, time-to-close, or a post-close franchise
What would settle these questions
Authorized Datasite access with exportable admin traces.
A sealed room and Q&A set with independently adjudicated keys.
Imported observations of the current admin process.
A declared Intralinks, Firmex, or DealRoom cell, same contents, same budget.
Permission and watermarking audit, separate from Q&A time.
A mid-process switch observation, including buyer friction.
Admin-hour and hosting-cost measurement, including failed access.
Datasite's published terms preserve customer ownership of uploaded documents while granting Datasite the rights needed to provide the service. Separate program and AI schedules may grant broader rights over usage and aggregated data.
Public website terms are not the executed commercial VDR agreement. The target's service agreement, DPA, AI schedule, and precise data taxonomy remain controlling.
Independent auditor A-LIGN describes a multi-year assurance relationship covering Datasite's ISO, SOC, privacy, cloud, and AI-governance frameworks.
An auditor case study is not the assurance report. It does not expose scope exclusions, control exceptions, bridge periods, target configuration, or incident history.
Datasite says customer deal documents are not used to train external models; anonymized aggregated attributes may be retained for product development; project interaction reinforcement stays within that project; nonarchived project documents are deleted after termination.
Vendor policy. Archive products and separate program terms create retention and derivative-data exceptions. Anonymization quality, deletion execution, and negotiated target terms require direct verification.
Datasite publishes a control set that includes watermarking, embedded redaction, SSO, MFA, granular rights, SOC 2 Type II, and multiple ISO certifications.
Certification and feature declarations support control design, not the effectiveness of a target room's configuration or the absence of incidents.
Datasite markets automatic preservation of deal history in a searchable, reusable archive that can also be downloaded or shared. Archive persistence is a governance feature and a switching mechanism.
Vendor documentation does not establish export fidelity for Q&A, permissions, bidder groups, versions, audit trails, IRM, or portability into another VDR.
