Sellside Stack is reader-supported. Some links may be affiliate links: if you buy through them we may earn a commission, at no extra cost to you. It never affects our verdicts. How we make money.

No affiliate links on this page yet. If we add any, this page will be labelled Ad.

Disclosure: Sellside Stack may earn a commission if you buy through some links on this page, at no extra cost to you. Commissions never affect our rankings or verdicts, and vendors cannot pay to be included. Our research is independent. We have not hands-on tested the products in this article. This review is based on vendor documentation, trust centres, published pricing, court records and third-party reviews, all checked in November 2026 and listed under Sources. Nothing here is legal advice.

The best AI note taker for investment banking, judged on risk, not features

By Sellside Stack Editorial · Prices as of November 2026

Most note-taker reviews rank tools on summary quality and integrations. For a deal team, those are second-order. The first-order questions are: Who consented to the recording? Where does the transcript live, for how long, and who else can read it? And what happens when the call contains inside information?

A management meeting on a live sell-side process, a wall-crossing call ahead of a placement, an IPO roadshow one-on-one, an earnings-prep session with the CFO: every one of these can contain material non-public information (MNPI). An AI bot that auto-joins, records, and emails a summary to every attendee, including the counterparty, is not a productivity tool in that setting. It is an incident.

The short answer

Comparison table

Granola Fireflies Fathom Read.ai tl;dv Otter
Joins as a bot? No (device audio) Yes (bot) Bot, or bot-free on Mac (beta) Yes, visible notetaker Bot by default; desktop app option Yes (Otter Notetaker)
Audio stored? No; deleted after transcription Yes, unless Enterprise transcript-only mode Yes Yes (playback on Enterprise) Yes Yes
SOC 2 Type II Yes Yes (plus ISO 42001) Yes Yes Yes Yes
Custom retention / auto-delete Per user; workspace-wide on Enterprise Enterprise Enterprise Enterprise+ Listed on paid plans Enterprise controls
Training on your data No third-party training; internal anonymised training on by default on Basic/Business, off on Enterprise No training; zero-retention terms with AI vendors AI sub-processors not permitted to train (reported) Not used to train shared models No training Reported to train on de-identified data
Data location US (AWS) US; private storage option on Enterprise US Not confirmed EU US
Price, annual billing (as of Nov 2026) Business $14; Enterprise $35 /user/mo Pro $10; Business $19; Enterprise $39 /seat/mo Premium $16; Team $15; Business $25 /user/mo; free tier Pro $15; Enterprise $22.50; Enterprise+ $29.75 /user/mo Pro $18; Business $29 /seat/mo Pro $8.33; Business $19.99 /user/mo

How we evaluate

For confidential deal use we weight:

  1. Consent and visibility (25%). Does the tool make it obvious to everyone that they are being recorded, and can admins enforce that?
  2. Data minimisation (25%). Is audio retained? Can transcripts auto-delete on a schedule set by compliance?
  3. Access control and distribution (20%). Who receives summaries by default? Can external sharing be switched off?
  4. Assurance (15%). SOC 2 Type II, model-training terms, data location, DPA availability.
  5. Usefulness (15%). Summary quality, search, CRM integration.

We have not used these tools on live calls. Where a point comes from a third party rather than the vendor, we say so.

The rules that apply regardless of tool

Recording consent. Several US states, including California, require the consent of every party to record a confidential conversation. Other jurisdictions have their own rules. The safe practice on any deal call is an explicit spoken notice and agreement at the start, recorded in the transcript. That risk is not theoretical: in In re Otter.AI Privacy Litigation (N.D. Cal.), the court on 13 August 2026 granted Otter's motion to dismiss only in part. Federal wiretap, California privacy and Illinois biometric claims were allowed to proceed. That is not a finding of liability, but it shows the exposure.

MNPI and inside information. In the UK and EU, the Market Abuse Regulation prohibits unlawful disclosure of inside information and requires insider lists. In the US, Regulation FD governs selective disclosure by issuers. An AI tool that auto-distributes a management-call summary to all invitees, or stores it in a searchable workspace shared across the firm, can move inside information across an information barrier. Configure distribution to "host only", keep deal calls out of shared workspaces, and make sure anyone who can access transcripts is on the relevant insider or wall-crossed list.

Retention. Your firm's record-keeping policy, not the vendor's default, decides how long a transcript lives. If compliance requires deletion after 30 days, or retention for years for regulated communications, the tool must support it. Otherwise, do not use it for that category of call.

The reviews

1. Granola: the lowest-footprint option

Visit Granola →

Granola captures audio from your own device rather than sending a bot into the call, and does not store meeting audio. It transcribes in real time on Mac and Windows and keeps only the transcript and your notes. It holds SOC 2 Type II and stores data on AWS in the United States. Third-party AI providers are contractually barred from training on your data. Granola does use anonymised data for internal model improvement by default on Basic and Business, which Enterprise switches off workspace-wide. Business is $14/user/month. Enterprise starts at $35 and adds SSO, workspace-wide auto-deletion, and admin-enforced transparency notices (chat notification, video watermark) on Zoom, Meet and Teams.

Workflow fit: Management presentations and buy-side diligence calls where you want your own notes enhanced, not a recording of the counterparty.

Who it's for: Bankers, PE deal teams, corporate development. Skip it if: You need an audio record for compliance or playback.

Cons: - No bot means no automatic signal to other participants unless you turn on the transparency features. Consent is on you. - Internal training on by default below Enterprise. - US-only storage.

2. Fireflies: governed recording at scale

Visit Fireflies →

Fireflies has the most complete compliance stack on paper: SOC 2 Type II, ISO/IEC 42001, GDPR with a DPA, HIPAA with a BAA on Enterprise with private storage, a no-training commitment, and zero-day retention agreements with AI vendors. The Enterprise plan ($39/seat/month, annual only) adds custom data retention, a transcript-and-summary-only mode that discards audio and video, private storage, super-admin controls, SSO and SCIM. Pro is $10 and Business $19 per seat per month on annual billing.

Who it's for: Firms that want a recorded, searchable archive and have compliance resources to configure it. Skip it if: You are on Pro or Business and handling MNPI. The controls that matter are Enterprise-only.

Cons: - Bot-based; visible but intrusive on sensitive calls. - AI credits are not included on Enterprise and are sold separately. - Easy to over-share if auto-join and auto-email defaults are left on.

3. Fathom: the best free tier, for the right calls

Visit Fathom →

Fathom's free plan offers unlimited recordings and transcription, which makes it the default for individuals. It is SOC 2 Type II audited, with HIPAA support. Its pricing page lists a standard BAA on Team. Bot-free capture is in beta on Mac. Paid plans are Premium $16, Team $15 and Business $25 per user per month (annual). A third-party security review notes all data is stored in the US and that Fathom does not hold ISO 27001.

Who it's for: Founders and operators recording sales, hiring and internal calls. Skip it if: You need custom retention. Third-party breakdowns place it on Enterprise only.

Cons: - Custom retention, SSO and SCIM are reported as Enterprise-only. - US-only data storage. - Bot capture by default outside the Mac beta.

4. Read.ai: visible by design

Visit Read.ai →

Read.ai joins as a visible notetaker and says conversations are not used to train shared models. It holds SOC 2 Type II, with a DPA and SCCs available. Pricing (annual): Pro $15, Enterprise $22.50, Enterprise+ $29.75 per user per month. Enterprise+ requires a minimum of five licences (one Read.ai page says 10) and is where custom retention, SAML/SCIM, domain capture and HIPAA live.

Who it's for: Teams that want analytics on meeting effectiveness as well as notes. Skip it if: You need governance below Enterprise+.

Cons: - Governance features gated at the top tier. - Meeting "coaching" metrics may be unwelcome in client meetings. - Default distribution settings need tightening for deal use.

5. tl;dv: the EU-hosted option

Visit tl;dv →

tl;dv is Germany-based and hosts data in Europe. It holds SOC 2 Type II, is GDPR compliant, and states it never trains on customer recordings. Free-plan recordings are deleted after three months. Pro is $18 and Business $29 per seat per month (annual). It makes no HIPAA claim.

Who it's for: European teams that prefer EU data residency. Skip it if: Your procurement needs SSO, SCIM and published audit controls below Enterprise.

Cons: - Bot by default. - Enterprise controls (SSO, SCIM, logs) are not on the lower tiers. - Lighter on deal-specific integrations.

6. Otter: capable, but read the litigation first

Visit Otter →

Otter is SOC 2 Type 2 and inexpensive (Pro $8.33, Business $19.99 per user per month annual). Enterprise adds SSO, SCIM and domain capture. But the consolidated class action alleging that Otter Notetaker joins meetings and records without all-party consent survived a motion to dismiss in part in August 2026, and third-party analyses report that Otter trains on de-identified audio and transcripts.

Who it's for: Individuals transcribing their own non-confidential meetings and lectures. Skip it if: You are on any call involving clients, counterparties or MNPI.

Cons: - Open litigation over consent practices. - Reported training on de-identified data. - Auto-join behaviour requires careful configuration.

A deal-team configuration checklist

  1. Disable auto-join; start recording manually per meeting.
  2. Announce recording and get spoken agreement at the start.
  3. Set summary distribution to host only. Never auto-send to external attendees.
  4. Keep deal calls out of shared team workspaces; restrict to the deal team.
  5. Set retention to match your policy, and turn off audio storage where possible.
  6. Disable AI training at workspace level.
  7. Never use a personal free account for firm business.

Verdict

For confidential deal calls, Granola Enterprise is the AI note taker we would put in front of an investment banking compliance team first: no bot, no stored audio, and workspace-level retention and training controls. If you need a recorded archive, Fireflies Enterprise has the strongest documented governance. Fathom is excellent and free for non-confidential work. Treat Otter as unsuitable for client or MNPI calls until its litigation is resolved. Whichever you choose, the configuration matters more than the logo.

Sources