Claude vs. ChatGPT vs. Microsoft Copilot for investment firms

Three abstract apertures frame different views of one form, suggesting a comparison of AI tools.

Ask Anthropic, OpenAI and Microsoft whether their business AI tools train on your deal documents, and all three will tell you they don't by default. Then ask which plan gives your compliance team audit logs, automatic user provisioning and a retention period it controls. Now the answers depend on the product, and even more on the price tier.

This is for the COO, CFO or ops lead at a fund or lender who has to pick one tool. We're vendor-neutral and build workflows inside whichever workspace a firm already uses. Everything below reflects vendor pages as of October 2026. Prices and plans changed several times this year, so confirm current terms before you sign.

The short answer

None of the three wins outright. If your firm runs on Microsoft 365 and keeps deal files in SharePoint, Copilot works inside the permissions and retention rules you already manage. If your deal teams live in outside data rooms, long PDFs and Excel models, Claude and ChatGPT both sell plans aimed at that work. Whichever you choose, the plan you buy settles most of the compliance questions.

Our advice is to pick the tool your compliance team can approve and your people will actually use. The workflow matters more than the brand. If your templates, examples and review steps are written down, they carry over when you switch.

Be clear about which product you mean

"We have Copilot" can mean several different things. Microsoft renamed Microsoft 365 Copilot to Microsoft Copilot this year (Microsoft), and its own overview now separates three experiences (Microsoft). Copilot Chat comes with eligible Microsoft 365 plans and works from the web plus files you upload or open. It won't search your SharePoint on its own. A basic tier adds standard access inside Word and Excel. The paid add-on grounds answers in your mail, files and meetings and adds agents.

Claude and ChatGPT also have personal plans. Anthropic's consumer plans follow a separate data policy (Anthropic), and personal ChatGPT accounts deserve the same caution. Keep deal work on business plans.

List prices per user per month, as of October 2026:

PlanList priceNotes
Claude Team$20 annual, $25 monthly2 to 150 seats; Premium seat $100 annual
Claude Enterprise$20 seat, annualPlus usage at API rates
ChatGPT Business$20 annual, $25 monthlyMinimum 2 users
ChatGPT EnterpriseNot publishedContact sales
Microsoft 365 Copilot$30 annual, $31.50 monthlyAdd-on to a Microsoft 365 licence
Copilot Business$21 list, $18 promoUp to 300 users

Claude's prices are on its pricing page. Enterprise bills seats plus metered usage (Anthropic), so get a quote based on expected usage. ChatGPT's figures come from OpenAI's pricing docs. The Copilot add-on needs a qualifying Microsoft 365 licence underneath it, and monthly billing still requires an annual commitment (Microsoft). Microsoft's page says the $18 Copilot Business promo runs through December 31, 2026 (Microsoft).

The plan matters as much as the product

On Claude and ChatGPT, the controls a compliance officer asks about sit in the top tier. Claude Team includes SSO and admin controls for connectors. SCIM, audit logs, a Compliance API, fine-grained roles, custom retention, IP allowlisting and customer-managed encryption keys are listed for Enterprise (Anthropic). ChatGPT Business has SAML SSO and an admin console, and OpenAI's admin docs say SCIM is not included (OpenAI). ChatGPT Enterprise adds SCIM, role-based access, key management, the Compliance API, and retention and residency controls (OpenAI).

Copilot is different because it borrows your tenant's controls: Entra ID for identity, Purview for audit and retention, Content Search for review (Microsoft). Microsoft has also added an AI Administrator role (Microsoft). Those controls are only as good as your current Microsoft setup.

If an LP questionnaire or an examiner will ask you for usage logs, budget for Enterprise on Claude or ChatGPT, or check that Purview is configured if you go with Copilot.

How the three compare

ClaudeChatGPTMicrosoft Copilot
Training defaultOff on Team and EnterpriseOff on Business and EnterpriseNot used to train foundation models
RetentionUntil deleted; Enterprise sets a periodEnterprise controls; Business unclearYour Purview policies
Admin, entry planTeam: SSO, connector controlsBusiness: SSO, no SCIMYour Entra and Purview setup
Admin, top planEnterprise: SCIM, audit logs, rolesEnterprise: SCIM, roles, Compliance APISame, plus AI Administrator role
ExcelBuilds .xlsx; Excel add-inExcel add-in (confirm terms)Copilot in Excel; Cowork
Long PDFsUp to 1,000 pagesNo published limit we could confirmNo published limit we could find
CitationsCell citations in Excel add-inCites connected files (confirm)Cites Microsoft 365 sources
ExtensibilitySkills, plugins, MCP connectorsPlugins and skills; GPTs retiringAgents, Copilot Studio, Cowork skills
Microsoft 365 fitConnector and Office add-insExcel add-in, SharePoint connectorBuilt in

Training, retention and residency

Anthropic doesn't train on Team or Enterprise content unless someone sends feedback, which is kept for up to five years, and owners can switch feedback off (Anthropic). OpenAI says no training on business data by default for Business, Enterprise and Edu (OpenAI). Microsoft says prompts, responses and Microsoft 365 data aren't used to train foundation models (Microsoft).

A training default does not settle whether you may upload a document. Your NDA or the seller's data-room terms may still restrict it. UpLevered puts it well: "No training, no retention, no disclosure, and permission to process are four different questions" (UpLevered). We cover this in is it safe to put a CIM into ChatGPT.

Retention is where the plans diverge. Claude keeps chats until a user deletes them and purges deleted chats within 30 days (Anthropic). Enterprise owners can set a custom period with a 30-day minimum (Anthropic). If you want zero data retention on the Claude chat product, ask Anthropic in writing. OpenAI documents zero retention for its API, and says ChatGPT Work's local computer mode does not provide it (OpenAI). Ask your OpenAI rep what retention settings Business admins get, because published sources disagree. Copilot stores prompts and responses in your tenant, where Purview retention and search apply (Microsoft).

Two details catch firms out. The Claude for Excel add-in doesn't follow your Enterprise retention settings and doesn't appear in Enterprise audit logs, though sessions show up in the Compliance API beta (Anthropic). And inside Copilot, some Anthropic models are labelled "with Data Retention." They are off by default, and if an admin turns them on, Anthropic stores most inputs and outputs for up to 30 days (Microsoft). So the model you pick inside Copilot can change your data terms.

On residency, Copilot is covered by the EU Data Boundary, but Microsoft notes model calls can be routed to other regions at times of high demand (Microsoft). Anthropic models in Copilot sit outside the EU Data Boundary and are off by default in the EU, EFTA and UK (Microsoft). Claude Enterprise documents a US-only inference option (Anthropic), while Anthropic's regional residency is tied to cloud platforms like AWS and Google Cloud (Anthropic). ChatGPT Enterprise lists residency controls (OpenAI); confirm the current region list with OpenAI.

All three publish security certifications (Anthropic, OpenAI, Microsoft). Ask for SOC 2 Type II reports under NDA. They cover the vendor, not your configuration.

Excel and long PDFs

Claude publishes its limits: files up to 500 MB, 20 per chat, and PDFs up to 1,000 pages, with full visual analysis up to 100 pages (Anthropic). It can create and edit Excel, Word and PowerPoint files (Anthropic). The Claude for Excel add-in gives cell-level citations and keeps formula relationships, but doesn't support macros or data tables. Anthropic itself advises review before anything goes to a client (Anthropic).

OpenAI's help center describes a ChatGPT for Excel and Google Sheets add-in, generally available since May 2026, with Business and Enterprise usage drawn from credits (OpenAI). Confirm the current terms with OpenAI before you plan around it.

Copilot in Excel can edit sheets, formulas, charts and PivotTables. Microsoft's own FAQ says results can be inaccurate and advises against relying on them for financial decisions (Microsoft). Copilot Cowork creates and edits Word, Excel and PDF files in OneDrive and SharePoint, billed through usage credits (Microsoft).

Neither OpenAI nor Microsoft publishes a per-file limit that we could confirm. Test each tool on your longest real credit agreement and your heaviest model. For the tasks where these tools still need a careful check, see where AI isn't reliable in underwriting.

Citations and grounding in your documents

Copilot reads your Microsoft 365 data with each user's existing permissions, cites what it used, and honors sensitivity labels (Microsoft). That only works on the paid licence (Microsoft). It also means Copilot will surface anything your SharePoint permissions already expose, so an overshared deal folder becomes an AI problem. Microsoft sells SharePoint Advanced Management partly to address that (Microsoft). If your data room is an outside VDR, we found no Microsoft connector for one, so ask Microsoft or your reseller what it supports.

Claude Projects switch to retrieval when a project's files get large (Anthropic). Anthropic's financial services connectors include Box, Egnyte, SS&C Intralinks, PitchBook, FactSet and S&P Capital IQ (Anthropic, Anthropic).

OpenAI's help center says its SharePoint connector follows each user's existing permissions (OpenAI). OpenAI also announced ChatGPT for Financial Services in September 2026 (OpenAI). Reports differ on which plans get it and which data sources it connects to, so ask before you count on it.

Reusable workflows

Shared, tested workflows are where these tools start paying off (more on that here), such as a screening memo built from a CIM.

Claude uses Skills, which Team and Enterprise owners can provision across the firm. Enterprise can also restrict who creates them (Anthropic). Custom connectors use MCP (Anthropic).

On ChatGPT, custom GPTs are being retired in Enterprise workspaces and replaced by plugins. Custom actions don't transfer and have to be rebuilt (OpenAI). Older comparisons that recommend building on GPTs are out of date.

Copilot offers agents, with Copilot Studio licensed separately for custom builds (Microsoft), and Cowork supports up to 50 custom skills (Microsoft).

Keep your templates, examples and review checklists in files you own. Then a switch means repackaging work you already have, and you don't start from a blank page.

Who should pick which

If your work already lives in Outlook, Teams and SharePoint, start with Copilot. There's no new vendor, and Purview already handles retention and audit. Clean up deal-folder permissions first, and budget for the paid licence on top of your base plan. Copilot can also run OpenAI and Anthropic models as Microsoft subprocessors (Microsoft).

If your deal teams work from outside data rooms, long credit agreements and Excel models, look hard at Claude. It has published file limits, data-room and market-data connectors, and a Microsoft 365 connector on Team (Anthropic). Plan on Enterprise if you need audit logs or custom retention. Ask OpenAI the same data-room questions about its financial services offering.

If your people already use ChatGPT every day, standardizing there keeps adoption easy. Go to Enterprise if you need SCIM or role-based access, and plan for the move from custom GPTs to plugins.

For a small family office, a business tier (Claude Team, ChatGPT Business or Copilot Business) may be enough. Write down which controls you're going without, so you can answer an LP or auditor later.

Some firms run Copilot for everyone plus Claude or ChatGPT for the deal team. That works, but it doubles your policies, logs and training.

How to decide in a month

Take one credit agreement and one model you know well, give the same tasks to each shortlisted tool, and score accuracy, citations, and whether compliance can see what happened. Before anyone uploads deal material, have counsel review your NDA terms, restrict use to approved business accounts, and turn logging on.

How we help

Giantfish helps investment firms choose a workspace their compliance team can approve, then builds their templates, past work and review steps into it and trains the team. See how we help firms choose and set up a workspace.

Sources