SharePoint and Geistra Pro · tested October 30, 2026
SharePoint keeps the deal's files. Geistra Pro keeps the deal.
We built a realistic M&A data room, put the same files in SharePoint and in Geistra Pro, and gave Claude the same prompts. Before SharePoint could even find the files, Claude had found every insurance gap we planted, built the deal calendar with reminders, listed every consent and put it all on a dashboard for the team. Through SharePoint, Claude later found the same problems, and every one of them stayed in the chat.
The test
- A fictional $86 million acquisition: Summit Ridge Capital buys Atlas Precision Components, a Michigan parts maker.
- 47 documents in 10 folders, as a real data room would have them: contracts and amendments, insurance policies, leases, the purchase agreement, disclosure schedules, financials, employment agreements, permits.
- We planted problems a deal team must catch, and wrote them down first: 7 insurance gaps, 7 traps that only look like gaps, 12 consents and payments the sale triggers, and 29 deadlines.
- The same files went into a private SharePoint site and a Geistra Pro deal space, uploaded by hand on both sides.
- The same prompts went to Claude, each one in a fresh chat that had never seen the answer key, connected to one product at a time.
Time until Claude could find the files
SharePoint
About 70 minutes
SharePoint can't open a zip, so the folder was unzipped and uploaded. Neither Claude nor SharePoint's own search could find a single file for over an hour. Search first found them between 63 and 70 minutes after upload, matching words in the text; search by meaning would need Microsoft 365 Copilot.
Geistra Pro
62 seconds
Geistra unpacked the zip into its 10 folders and Claude found the first contract 62 seconds after upload. Every file was ready at 69 seconds; the text inside the contracts was searchable at 74.
Seven prompts, same Claude
| Prompt | SharePoint | Geistra Pro |
|---|---|---|
| “Find and track all the insurance policies and compare against the contracts to verify.” | Found all 7 too, with no false alarms: the same Claude. Then it tried to save a tracker and was refused, because the connector can't write files at its default settings. The findings stayed in the chat. 4 minutes | Found all 7 insurance gaps, including one hidden in a contract's second amendment and an injury claim missing from the insurer's loss reports. Flagged none of the 7 traps. Every finding cites the document and section, kept as a tracker the team can open. 5 minutes |
| “Build a calendar of every deadline, expiration and notice date in the Project Atlas data room, and set reminders for the deal team.” | Listed 28 of the 29 deadlines. Then it tried to put the first one on the calendar and was refused, and couldn't look up who's on the deal team. No events, no reminders: the list stayed in the chat. 5 minutes | All 29 deadlines in the answer key, no wrong dates, 55 in all, each with its source and how it was worked out. Reminder emails set up to run on their own, after a person approved them. 7 minutes |
| “Which contracts need consent or notice for the change of control, by when, and who is responsible?” | All 12 as well, plus one the Geistra Pro run didn't raise: the company's debt means a loan agreement to check, and it isn't in the room. Answered in chat, reading every document again. 4 minutes | All 12 consents, notices and payments, who signs for each side, and by when. Added to the same calendar, so the reminders already cover them. 3 minutes |
| “Build a dashboard for the deal team showing open diligence issues, consents needed and upcoming deadlines.” | Wrote a thorough dashboard as a chat reply. It couldn't be saved anywhere, so it won't update and only the person who asked can see it. 5 minutes | A four-page dashboard drawn live from what the earlier prompts kept: what needs attention now, the next deadlines, consents before closing, open issues by severity. 2 minutes |
| “What did we find on insurance earlier today, and what's still open?” | Found no record of the earlier work in SharePoint, OneDrive, email, Teams or the calendar, so it started over from 18 documents. This time it missed one of the problems the first chat had found. 2 minutes | Answered from what earlier chats had kept: every open item with its next step, the Coastline deadline first, and that nothing had been resolved since. It re-read none of the documents. 57 seconds |
| “A new cyber binder was added to the data room. Does that close any issues? Update whatever we're tracking.” | Read the binder well and narrowed the cyber gap for the same reasons, after re-reading about 20 documents. Of the 5 other open problems it named 2. With nothing kept from earlier chats, there was no tracker to update. 2 minutes | Read the binder against what was kept: the cyber gap narrowed to $5M but stays open until the policy is endorsed, the premium is paid and the customer gets its certificate. All 5 other problems kept open. Updated the tracker, with 2 new deadlines and reminders from the binder itself. 2 minutes |
| “The bank's credit agreement was added to the data room. Does it add any insurance requirements Atlas doesn't meet, and what does it mean for closing?” | Search said the agreement mentions insurance, but not where. Claude read about 50 of the 306 pages, ten at a time, and stopped before reaching page 188 or 247. It found none of the 5 requirements and said it couldn't answer yet. 3 minutes | 306 pages. Search landed on page 188, the insurance clause; Claude read 2 pages and found 4 of the 5 requirements, each cited: $3.5 million short on business interruption, no lender loss payee, the bank not an additional insured, certificates unverified. It missed the change-of-control default on page 247. 2 minutes |
Same Claude, same answers. Different afterwards.
On the first passes Claude found the same problems through both. What changed is what was left when each chat ended, and what the next chat had to start from.
| Question | SharePoint | Geistra Pro |
|---|---|---|
| Problems found against the answer key | Insurance 7 of 7, deadlines 28 of 29, consents 12 of 12, no false alarms. After the new binder, 2 of the 5 other open problems. In the 306-page agreement, 0 of 5. | Insurance 7 of 7, deadlines 29 of 29, consents 12 of 12, no false alarms. After the new binder, all 5. In the 306-page agreement, 4 of 5. |
| Kept for the team | Four chat replies. No files, no calendar events, no lists: every save was refused. | An insurance tracker, a deal calendar and a dashboard: 94 records, each citing its source, open to everyone on the deal. |
| Reminders | None. | A reminder email as each deadline nears, and a Monday summary, running on their own after a person approved them. |
| A new document mid-deal | Searchable about 11 minutes after upload. Claude rebuilt the picture from the documents and named 2 of the 5 other open problems; nothing to update. | Searchable 5 seconds after upload. Claude checked it against what was kept, left all 5 other problems open and updated the tracker, two minutes later. |
| A new chat the same day | Starts over: no record of the earlier work anywhere, 18 documents re-read, one problem missed. | Picks up where the team left off in under a minute, re-reading nothing. |
What each prompt cost Claude
Building the deal file costs more on Geistra Pro, once: Claude is also saving the tracker, the calendar and the dashboard, which SharePoint refused. Every question after that costs about a third as many tokens, a quarter of the time, and comes back right, because Claude reads what was kept instead of re-reading the documents. For a 10-person deal team asking two questions a day for six months, that is about 8 billion tokens, $4,970 and 169 hours of waiting through SharePoint, against about 2.8 billion, $2,550 and 46 hours in Geistra Pro, in a room this small. In a room of thousands of pages the gap grows; see the model below.
| Prompt | SharePoint | Geistra Pro |
|---|---|---|
| Insurance vs contracts | 2.4M read · 20k written · 50 calls · 4 min | 3.2M read · 33k written · 48 calls · 5 min |
| Deal calendar and reminders | 3.0M read · 22k written · 75 calls · 5 min | 5.0M read · 42k written · 73 calls · 7 min |
| Change-of-control consents | 1.8M read · 16k written · 53 calls · 4 min | 2.3M read · 16k written · 39 calls · 3 min |
| Deal-team dashboard | 2.4M read · 22k written · 69 calls · 5 min | 1.9M read · 15k written · 20 calls · 2 min |
| New chat: what did we find? | 1.3M read · 10k written · 39 calls · 2 min | 0.9M read · 6k written · 14 calls · 1 min |
| A new cyber binder | 2.6M read · 12k written · 40 calls · 2 min | 1.2M read · 12k written · 17 calls · 2 min |
| A 306-page credit agreement | 2.5M read · 12k written · 26 calls · 3 min · answer not found | 3.2M read · 12k written · 30 calls · 2 min · most of it searching |
| Next day: what's open, what changed? | 5.4M read · 20k written · 83 calls · 5 min | 1.2M read · 6k written · 15 calls · 1 min |
| Day after: same question | 2.5M read · 17k written · 63 calls · 4 min | 1.1M read · 6k written · 14 calls · 1 min |
| Building the deal file (prompts 1-4, once) | 9.5M read · about $6 · nothing kept | 12.4M read · about $8 · tracker, calendar, dashboard and reminders kept |
| Each later question, in a new chat (average of 3) | 3.1M read · $1.91 · 4 min · 18 to 48 documents re-read · something wrong or missing every time | 1.1M read · $0.98 · 1 min · no documents re-read · nothing wrong |
Counted from each chat's own transcript, rounded: every token Claude read, including the chat so far, which it reads again at each step (most of that is cached, at a lower price), and every token it wrote. About 63,000 of each chat's first read is the same on both sides: instructions and the descriptions of each product's tools. Dollars at Claude Opus 5.5's API prices (Anthropic's price list of October 6, 2026), with cached reads at their lower price. Claude's tokens only: Geistra Pro's reminder emails use the business's included AI allowance.
At 4,000 pages: what Claude reads
A real deal room is far bigger than ours. Here is what each question costs Claude as it grows, modelled from what we measured.
- Through Claude's Microsoft 365 connector, search says which file mentions a word, never which page. So Claude reads the file in chunks of about ten pages until it finds the part it needs or runs out of room. Nothing is kept, so every new chat reads again.
- In Geistra Pro, search lands on the page, and Claude reads that page. What it works out is kept for the team, so later chats read the findings, not the documents.
- Our deal room is 54 pages. Everything fit either way, which is why the first passes cost about the same. The gap opens as the room grows.
| What Claude reads | SharePoint | Geistra Pro |
|---|---|---|
| One clause in a 306-page agreement | Up to the whole file: about 330,000 tokens · $1.32 | The one page: about 750 tokens · under 1 cent |
| One insurance question, 4,000-page room | Every relevant document: about 900,000 tokens · $3.60 | The open findings, after a first pass of about 180,000: about 46,000 tokens · $0.18 |
| A 60-day deal, 300 questions | About 270 million tokens · $1,080 | About 14 million tokens · $56 |
| What one standard Claude chat (about 200,000 tokens) can hold | About a fifth of the documents a 4,000-page insurance question needs | All of it |
A model, not a live run. Built from what we measured (about 370 to 1,080 tokens per page; a kept finding about 510 tokens; whole files opened through the connector; no documents re-read once findings are kept) and stated assumptions: 750 tokens per page, 30% of pages in documents that touch insurance, 2 clause pages and 3 findings per document, a quarter of findings still open, 5 questions a day and 100 new documents over 60 days. Dollars are Claude's input tokens at the API list price for Claude Opus 5.5 ($4 per million, Anthropic's price list of October 6, 2026); half that on Claude Sonnet 5.5. On a Claude plan there's no per-token bill: the same tokens count against the plan's usage limits. Checked against a live run on the 306-page agreement: through SharePoint, Claude read about 50 pages (a sixth of the file, close to the model's fifth) and stopped without finding the clause; in Geistra Pro it read 2. In that run Claude Code set aside any single read over about 50,000 characters instead of reading it; other Claude apps may handle large reads differently.
What it costs
SharePoint's seats cost less. What it doesn't keep, Claude pays for again in every chat.
| Per month, 10 people | SharePoint | Geistra Pro |
|---|---|---|
| Where the files live | Microsoft 365 Business Basic, $7 a person: $70 | Included |
| The deal space | A SharePoint site, included | Team: $99 with two seats, +$35 for each of the other eight: $379. Business: $699 with ten seats. |
| Microsoft's own AI (Claude doesn't need it) | Microsoft 365 Copilot Business, $21 a person ($18 on Microsoft's offer to December 31, 2026): $180 to $210 | Not needed |
| Claude | Each person's Claude plan | Each person's Claude plan |
| What Claude reads for each question | Whole files, again in every new chat | The page it needs, then what the team already kept |
Monthly, for a 10-person deal team paying yearly, from Microsoft's and Geistra Pro's own pricing pages. Microsoft's prices are US list prices before tax, and vary by country. Seats are the smaller part on a large deal: Claude pays in tokens for everything it reads, and a whole file can be hundreds of times the size of the page that answers the question. The model above puts a number on it.
Behind the scenes in Geistra Pro
Claude does the thinking in both tests. The difference is what Geistra Pro does around it: it files every page so Claude can find it, keeps what Claude works out, runs the reminders, and remembers what changed. From the deal space's own history:
- UploadGeistra Pro files the deal room. The zip is unpacked into its 10 folders, every file is scanned for malware, and the text of every page is pulled out and indexed for search by word and by meaning. Claude could find the first contract 62 seconds after upload; SharePoint took about 70 minutes.
- Prompt 1Claude reads Geistra's own guidance first. How to work in the space, what always asks a person, and what's new since it last looked: here, 48 files. That guidance has it check which lists already exist before saving anything, so it adds to them instead of making duplicates.
- Prompt 1Search lands on the page, not the file. A search for an insurance clause returns the page it's on, so Claude reads that page instead of the whole document. In the 306-page credit agreement it read 2 pages.
- Prompts 1-3What Claude finds is saved as records. A list of the 8 insurance policies, the 28 requirements in the contracts (each linked to the policy it was checked against), and 58 deal dates. Every record cites its document and section. 94 records in all.
- Prompts 1-4Apps the team can open. An insurance tracker, a deal calendar and a dashboard, built on those records. They read the records live, so when a record changes, every app shows it at once.
- Prompt 2Reminders that run on their own. Claude asked to set up reminder emails; Geistra held them until a person approved. Since then Geistra runs them by itself: on Oct 31 it emailed the reminders for the Sentinel notice, the Pacific Rim renewal and the Coastline response, and a digest goes out every Monday.
- Every stepA history of who changed what. Each change is kept with the person, the AI app they used, and the reason. A new chat asks what changed since it last looked, and picks up from there: in prompt 5 it answered in 57 seconds without opening a single document.
- Prompt 7A new document mid-deal. The cyber binder was searchable 5 seconds after upload. Claude checked it against the saved records and changed 12 of them in one step, all or nothing, adding the binder's own deadlines with reminders. The tracker, calendar and dashboard updated with it.
- ThroughoutClaude works as the person, with their permissions. Claude sees only what the person can see in the space. Anything that sends email or can't be undone waits for that person's approval.
A saved file isn't a record
An admin can let Claude save files to SharePoint. Even then, what Claude works out is kept as one more document. In Geistra Pro each finding is a record of its own, and Geistra looks after it like a librarian.
| A file Claude saves to SharePoint | A record in Geistra Pro | |
|---|---|---|
| What changed since yesterday | The file changed. Claude opens it again and works out what's different. | Exactly which findings changed, who changed them, through which AI app and why. |
| Ten people saving findings | Ten files, each laid out its own way. | One list. Before anything is saved, Geistra checks what lists already exist, matches the field names and steers away from duplicates. |
| Two people editing at once | The last save wins, silently. | Each finding is checked on its own: if two people change the same field, the second is asked which to keep. Related changes save together or not at all. |
| Where a finding came from | Whatever the file says. | Every finding cites its document and section, and links to the policy or contract it was checked against. |
| A date in a finding | Text in a file. | A date that acts: a reminder goes out when it comes due. |
| Seeing it | Open the file. | Apps the team opens, reading the findings live, with access set list by list. |
What each one is
SharePoint
Microsoft's file sharing and intranet, in Microsoft 365.
- Files and folders, with sharing and versions, inside Microsoft 365
- Keyword search; search by meaning needs a Microsoft 365 Copilot license
- With Claude's Microsoft 365 connector: search and read files; with the connector's write tools turned on by an admin, also files, folders, email and calendar invites
- What the AI works out lives in the chat, unless someone saves it as another file
Geistra Pro
One record per deal that any AI your team uses can work on.
- The documents, read page by page, searchable by their words within about a minute, and by meaning
- What the AI works out kept as records the whole team sees: findings, policies, deadlines, consents, each citing its source
- Reminders and weekly summaries that run on their own, after a person approves them
- Apps and dashboards on those records, and every change kept with who made it and which AI app
Side by side
| Question | SharePoint | Geistra Pro |
|---|---|---|
| Files | Stored in a site's document library, with versions and sharing. | Stored as the original, scanned for malware, read page by page (scans too). |
| Finding things | Keyword search. Search by meaning, and Copilot answers, need a Microsoft 365 Copilot license; new files on shared sites are added to that index daily. | Words and meaning on every page, for every member, about a minute after upload. |
| What the AI works out | Stays in the chat. With write tools on, Claude can save it as a new file. | Kept as records, linked to each other and to the documents, open to the whole team. |
| Deadlines and reminders | Calendar invites, if the connector's write tools are on. | A deal calendar with reminder emails and a weekly summary that run on their own. |
| Tracking the deal | Lists and Power Apps, built by hand. | Trackers and dashboards Claude builds from the records, live. |
| Which AI | Microsoft 365 Copilot, or Claude and others through connectors. | Claude, ChatGPT or anything that connects through MCP, with each person's own permissions. |
Keep SharePoint for your company's files. Run the deal in Geistra Pro, from the AI you already use.
Tested October 30, 2026. The deal, its companies and its documents are fictional, written for this test. SharePoint was used as a private Microsoft 365 team site with Claude's Microsoft 365 connector at its default settings, which in this test allowed reading only; an admin can turn on its write tools (files, folders, email and calendar invites), and results with Microsoft 365 Copilot may differ. Each prompt ran in a fresh Claude chat connected to one product. Microsoft and Anthropic details are from their own pages: Semantic indexing for Microsoft Copilot (Microsoft Learn); Optimize SharePoint content for Copilot (Microsoft Learn); Enabling and using the Microsoft 365 connector (Anthropic); Microsoft 365 Business Basic (Microsoft); Microsoft 365 Copilot Business pricing (Microsoft). Products change; check there for the latest. SharePoint and Microsoft 365 are trademarks of Microsoft; Claude is a trademark of Anthropic. Geistra is not affiliated with or endorsed by either.