The AI-native capability set
AI-native, not AI-added.
Other data rooms bolted a chat box onto a file server. Ours was built the other way round: every document is read, indexed and cross-checked the moment it lands — so the AI can answer in plain English, draft a response to a diligence request, and tell you when two documents disagree.
Each capability below has a short, silent recording of the real product.
- 01 Builds the deal from one sentenceRequest list and folder tree, drafted for you to cut 37 sec
- 02 Reads every document as it landsThen answers in plain English, with the sources 54 sec
- 03 Flags what does not reconcileBoth sources, the gap, and the reconciling item 35 sec
- 04 Drafts the diligence responsesCited from the data room, for your team to approve 45 sec
§01 · Setup
Deal setup
Describe the deal in one sentence. It builds the room.
Type what the deal is — a sell-side carve-out of a Dutch logistics business, say — and CogniSuite drafts the diligence request list and the folder structure to hold the answers. You cut what does not apply, add what it missed, and start issuing. The list arrives written for this deal rather than inherited from the last one.
- A drafted list, not a blank page
- Start from what the AI proposes or from one of your own templates. Either way you edit before anything goes out.
- The folder tree comes with it
- A standard M&A structure is created alongside the list, with per-folder visibility set from the start.
- Nothing is issued until you send it
- The draft is yours to cut. Requests reach your team only when the deal lead releases them.
Building a deal from one sentence · 71 sec · silent
Asking the room a question · 34 sec · silent
§02 · Reading
Comprehension
It has read everything before you ask.
Every file is parsed and indexed the moment it is uploaded: the financials, the contracts, the data pack, the appendix nobody opened. So a question put to the room is answered from the documents themselves, in plain English, with the sources attached — rather than returning eleven PDFs for somebody to read.
- Questions, not search strings
- “What is the customer concentration in FY2024?” is a question. Ask it that way.
- Meaning, not keywords
- It recognises that “revenue breakdown” and “sales by segment” are the same question.
- Every answer carries its sources
- Answers cite the documents they were drawn from, so checking one takes a click rather than an afternoon.
- Zero retention
- Deal data is used to answer your questions and nothing else. It never trains a model.
§03 · Reconciliation
Cross-document accuracy
It tells you when two of your own documents disagree.
The same figure appears in the management deck, in the audited statements and in the model, and the three do not always agree. Our engine compares them across the whole deal and flags what does not reconcile, showing both sources and the size of the gap. Ask it whether a figure ties across the deck, the accounts and the schedule, and it will answer with both numbers, the difference between them and the reconciling item — citing the documents it read. That is the kind of thing normally found late, by someone reading carefully — or not found at all.
- Reads every document on upload
- Financials, contracts, and the data pack, checked as they arrive.
- Reconciles figures across files
- The same number is compared across sources, and the ones that disagree are surfaced.
- Every flag is attributed
- Each discrepancy points to its source document and page, ready for a person to resolve.
- Caught before publication
- Flags surface at review, while the answer is still internal and still yours to correct.
Catching a cross-document inconsistency · 30 sec · silent
Drafting answers to the request list · 59 sec · silent
§04 · Drafting
Diligence responses
It drafts the response. Your team approves it.
Point it at the request list and it writes an answer to each open request, grounded in what is actually in the data room and citing the documents it used. Nobody starts from an empty box. The deal team reads, edits and approves, and nothing reaches a counterparty until a person has released it. Drafting is a separate step from matching, which is further down this page.
- One pass over the whole list
- Drafts are produced across the open requests, not laboriously one at a time.
- Grounded and cited
- Each draft points at the documents it drew on, so review is checking rather than rewriting.
- Approval is always a person
- A draft stays a draft until a member of the deal team publishes it.
§05–§09 · Underneath
What the AI stands on
None of the above works without the unglamorous half: a request that has an owner, permissions that hold, a way to reach people who will never log in, and a record of who saw what. That half is below.
§05 · Requests
Workflow
Request-centric workflow
Underneath the AI, the unit of work is the request rather than the folder. Each ask carries an owner, a status and a thread. That is what gives the AI something specific to answer against, and the audit trail something exact to record.
- Structured request lists
- Issue specific, trackable requests to your team. No ambiguity about what's needed.
- Status at a glance
- Open, assigned, answered, vetted, published. Know exactly where every request stands.
- Threaded communication
- Every request keeps its full thread: comments, clarifications, and status changes.
Incoming questions
One published answer matches all three
Revenue by segment, FY2024. Already published.
Confirm once and the answer goes to everyone who asked.
§06 · Matching
Intelligence
AI matching, with human confirmation
Matching works on meaning rather than filenames. It links an uploaded document to the open request it satisfies, and a newly asked question to an answer already published on this deal. In both cases the system proposes and a person confirms, and one confirmation delivers the answer to everyone who asked the same thing.
Matching and drafting are not the same thing
- Matching
- Links an uploaded document to the request it satisfies, by meaning. A person confirms the link.
- Drafting
- Writes the prose answer to that request, citing the data room, for the deal team to approve and release.
Matching finds the evidence. Drafting writes the response. They run in that order.
- Semantic, not keyword
- It recognises that "revenue breakdown" and "sales by segment" are the same question.
- Human in the loop
- The system suggests, a person confirms. No answer goes out without explicit approval.
- Answer once, not four times
- Four bidders asking the same question in different words receive one vetted answer.
§07 · Roles
Permissions
Role-based access
Most users are workers, not deal managers. When the accountant logs in, they see their task inbox, not a deal room or a set of workstream dashboards. Just what they need to do today.
- Workstream-based access
- Finance sees Finance. Legal sees Legal. No information overload.
- Task inbox for contributors
- Assignees see only their tasks. Complete one and it routes to the next person automatically.
- Data room separation
- Participants see only published answers. Internal routing stays invisible to external parties.
Each member sees only their workstream.
From CogniSuite · Project Alpha
Request: Audited financials, FY2024
The deal team needs the audited FY2024 financials. Reply with the files attached to respond.
Delivered by email. No login required to reply.
§08 · Channels
Integrations
Work where you already work
Not everyone needs to log into another platform. Participants can respond by email, and requests route to people wherever they are.
- Email replies
- Reply to request notifications straight from your inbox, attachments included.
- No login required
- Basic responses work without an account. Full features live in the web app.
§09 · The record
Compliance
A complete record of who saw what
Every access, every download, every routing hop is logged. When compliance asks who saw what and when, you have the answer.
See how we handle securityWhat the trail captures
- Access logs
- Who viewed which document, and from where. Each entry records the IP address, timestamp, and session duration.
- Download tracking
- Every file download is recorded and watermarked with the viewer's identity for leak tracing.
- Workflow history
- The full chain of custody: who assigned, who approved, who published. Every transition is logged.
Walk through it on a live deal
Run your own request list on a 14-day free trial, or take a 30-minute walkthrough. No slide deck, no commitment.