Fathom

Strip the models and a recorder remains, which is not what any tier is sold on. Every capability the vendor markets is inference: transcription, chronological and enhanced summaries, more than fifteen expert summary templates built around named sales qualification frameworks, generated action items, drafted follow up emails, a conversational assistant over a single call and across the whole account, search alerts that fire when new meetings match a saved topic, coaching metrics derived from speaking patterns, scorecards, and deal summaries assembled from what was said.

Even the free tier's headline pairs unlimited recording with instant summaries. Recording, storage, clips and playlists are containers for that output rather than the product. This is consistent with the lane, where no record has cleared the removal test, and it is the second such record in this batch. A buyer here is purchasing inference quality and should evaluate summary and action item accuracy directly rather than trusting the category.

Several things happen without a person and the published controls sit above them. The autonomous surfaces are real: a bot joins meetings on its own, summaries and action items generate unattended, follow up emails are drafted, and search alerts fire automatically when a new meeting matches a saved topic or keyword, which means the system is continuously reading everything captured and notifying on matches.

Against that the administrative controls are enumerated but positioned high in the ladder, with organisation wide security controls, single sign on and directory provisioning all at the quoted tier and team level administration from the middle tiers. Two lower controls exist and both cut against disclosure rather than toward oversight, being the ability to rename the bot and to switch off the in meeting banner.

Nothing published bounds what a drafted follow up may say or whether one can send without review. Ask what an administrator can restrict below the quoted tier, and whether drafted emails can send unreviewed.

The integration surface is documented precisely and the model layer is not named at all. On the published side this vendor goes further than most in one specific respect: two external assistant products are named as first class integrations with their own dedicated pages, and a public interface and assistant connection are documented at a developer subdomain, so a buyer can see exactly how meeting context reaches an assistant of their choice.

Summary templates are enumerated by count and by the named frameworks they follow. On the undisclosed side there is nothing: no provider, model family or version for transcription, summarisation, the conversational assistant, the coaching analysis or the drafted emails, and no statement of how accuracy is measured or how a customer corrects a wrong summary that has already synchronised into a customer record. Ask which models produce transcripts and summaries, and what happens to a record field written from an inaccurate one.

A confident commercial offer standing in for evidence. The pricing page carries no case study, no named customer, no logo wall, no outcome metric and no quantified claim of any kind, which is unusual for a product at this level of maturity, and the resource hub and overview pages were not reached on this pass so nothing here asserts they are empty.

What the vendor offers instead is commercial risk transfer, and it is genuinely substantial: a ninety day money back guarantee on every paid plan, free access for affiliated startups of selected investors and accelerators for up to two years, ten free seats for nonprofits, and a switching offer giving the top tier free for the remainder of a competitor's contract with data migration included. Those are real inducements and they are not measurements.

For a product whose value rests on summary and transcription quality, no accuracy figure appears anywhere reached. Ask for measured transcription and summary accuracy, and for a reference customer with numbers attached.

No outreach is conducted here and one generated artifact sits at the edge. The product records and summarises meetings a customer is already holding, so there is no prospecting, no list and no sending, and the obligations this axis examines do not arise for the core product. Graded in the middle of the band with that scope stated rather than at the floor, since the floor asserts an undisciplined sending product and this does not send.

The edge is the drafted follow up email, generated after a call and sent, if at all, from the user's own mail platform under that platform's authentication, so the compliance burden stays with the customer. Nothing published states whether a recipient of such a message is told it was machine drafted, or whether any review step is required before it leaves. Ask whether drafted follow ups can be sent without review and what a recipient can tell about their origin.

The instruments exist, were not read, and the retention control is sold rather than provided. A privacy policy, terms of service and a security and compliance centre on its own subdomain are all published and none was retrievable on this pass, so retention periods, lawful bases and the rights exercise route are unestablished and nothing here asserts they are absent.

What the pricing matrix does establish is the shape of the gap, and it is the same one recorded against the other meeting recorder graded this session. This product stores recordings and transcripts of people who are not its customers, whoever the customer happened to be meeting, and custom data retention policies are listed only against the quoted top tier. A customer on any of the four published plans therefore has no stated ability to set how long other people's recorded speech is kept. Ask what the default retention period is on the published tiers, and how a meeting participant who is not a customer requests deletion.

No purchased corpus, and the provenance question here is consent rather than licence. There is no contact database, no enrichment and nothing brokered: the material processed is the customer's own meetings, so the usual chain of sources, resale permissions and lawful bases does not apply and most of the axis is out of scope. Graded in the middle of the band with that stated. What remains in scope is that a transcript is not solely the customer's material.

It contains the recorded speech of everyone else on the call, routinely people at other companies with no relationship to this vendor, and the basis for capturing and storing their voice rests wholly on the customer having obtained consent. Nothing published describes how the platform supports, prompts for or evidences that consent, and the features described under the disclosure axis make the absence more consequential rather than less. Ask what record exists that participants agreed to be recorded, and whether it is retrievable per meeting.

The capture mechanics run against three companies' platforms and no standing with any of them is stated. The bot joins calls hosted on the major conferencing services, which are neither the vendor's infrastructure nor the customer's, and nothing published names a partner programme, marketplace listing, approved application status or conformance position with any of them, though each sets its own rules about automated participants and recording.

The alternative capture mode sharpens rather than softens the question: it records without a bot joining, which on a desktop means capturing the meeting outside the platform's own participant model altogether, and it is offered on the free tier and marked as in beta. Two things keep this off the floor. The platforms are named plainly, and having two independent capture paths means a customer is not wholly dependent on bot admission continuing. Ask what standing the bot holds with each conferencing platform, and how the bot free mode captures audio.

The question this category turns on is not answered anywhere reached. For a product that ingests recorded conversations, the load bearing commitment is whether customer meetings train the vendor's models or reach a third party model provider that retains them, and no statement on either appears on the pricing page, the feature comparison or the navigation, where a comparable vendor graded this same session places its training position directly on its pricing page.

The security and compliance centre was not retrievable on this pass, so nothing here asserts the commitment is absent, only that it was not found where a buyer would look first. What is published are adjacent controls at the quoted tier: custom data retention policies, organisation wide security controls, a signed agreement for the health information regime, and increased cyber insurance coverage.

Two external assistant integrations also carry meeting content out to systems the vendor does not control. Ask for a written statement that meetings do not train any model, and what a connected assistant may retain.

Every signal that a meeting is being recorded is removable, and two of the removals are sold as features. The default position is reasonable: a bot appears in the participant list and an in meeting banner announces the recording. From there the feature comparison lists a custom bot name and, separately, the ability to disable the in meeting banner, both as paid entitlements, and the free tier already offers a capture mode that runs without any bot joining.

Taken together a paying customer can record a meeting with no named bot, no banner and no participant list entry. The people affected are the other attendees, routinely from other companies, whose consent is a legal precondition in many jurisdictions rather than a courtesy. Nothing reached describes a consent prompt, a jurisdictional consent mode, an announcement that survives the banner being switched off, or any record that consent was obtained.

This is graded at the floor because the disclosure is not merely thin but purchasable away. Ask what an external participant sees when the banner is disabled and bot free capture is used, and what consent record the platform retains.

A named, tiered surface with two external assistant products treated as first class destinations. Six integrations carry their own dedicated pages, covering a work management tool, two assistant products, two customer record platforms and a general automation connector, with a full directory behind them and a messaging integration alongside.

A public interface and an assistant connection are documented together at a developer subdomain, which is a real developer surface rather than a footer pointer, and a partner programme and help centre sit around it. The customer record integration is described at field level, mapping meeting insights into specific named fields rather than pushing a note, which is depth rather than breadth.

It stops below the top band because the developer documentation was not retrieved on this pass so what the interface exposes is unverified, and because the record synchronisation is capped at three users per domain on three of the five tiers, which is a material limit on a headline capability.

Nothing on this question was reachable. No hosting provider, region, processing location, residency election, subprocessor list or cross border transfer mechanism appears in any material retrieved, and the security and compliance centre was not retrievable on this pass, so this reflects what could be reached rather than an established absence of disclosure.

The one adjacent entitlement is custom data retention policies at the quoted tier, which governs how long material is kept rather than where it sits. The question carries weight because the stored material is recorded speech of identifiable people, frequently from other organisations, across whatever countries a customer holds meetings in, and because one capture mode runs locally on a desktop machine while the other runs in the vendor's cloud, which are different data paths with potentially different geographies. Ask in which regions recordings and transcripts are stored, whether a region can be elected, and where bot free captures are processed.

A security centre exists on its own subdomain and would not open on this pass, which is a retrieval limitation rather than a finding. It is linked from the footer as security and compliance and returned nothing readable, and a separate status page is published alongside it, so no certification, audit period, auditor, report request route or control set was established and nothing here asserts any of them is absent.

What is readable comes from the feature comparison, where security appears with named entitlements rather than assurances: single sign on, directory provisioning through a named identity platform, organisation wide security controls, custom contracts with red line support, a signed agreement for the health information regime, and increased cyber insurance coverage, which is an unusual and specific thing to publish.

All of them sit at the quoted top tier, so a buyer on any published plan gets none. Ask for the current examination report, its audit period and auditor, and which controls are available below the quoted tier.

Four tiers carry per user figures at both billing frequencies with a permanent free plan beneath them, seat minimums are stated where they apply, trials start without a conversation, and three named concession programmes are published with their eligibility rules, covering investor backed startups, nonprofits and customers switching from a named competitor.

A ninety day money back guarantee is attached to every plan including the free one, which is a genuine commercial commitment rather than a discount. Two things hold it below the top band and both bear on what a buyer will actually experience.

Several allowances in the comparison are given as limited use rather than as numbers, including the cap on advanced summaries that free users meet first and the caps on the conversational assistant and automation connectors, so the constraint most likely to force an upgrade is unquantified on the page. And the advertised annual saving of twenty five percent or more is accurate only on the highest priced tier, computing to about twenty and twenty one percent on the two below it. Ask for the numeric caps behind limited use, and the rate for the onboarding programme marked as separately priced.

Retrieval routes exist and none is described as a way out. Call downloads and clips are marked unlimited from the free tier upward, which is a real and unusually early entitlement, and a public interface with an assistant connection provides a programmatic path, though its tier availability and what it exposes were not established.

Record synchronisation pushes summaries and mapped fields into a customer's own system continuously, so some of the value accumulates outside the platform by design, which is a genuine structural advantage on this axis. What is absent is the departure question.

Nothing published states what happens to stored recordings and transcripts after cancellation, how long a departing customer has to retrieve them, whether a bulk export exists, or how the coaching metrics, scorecards and deal views built up over years can be taken out. Ask for a documented bulk export of recordings, transcripts and analytics, and the retention window after an account closes.

There is no sending apparatus here to examine. The product operates no mailbox, no sending domain, no warmup, no rotation and no throttling, because it records and summarises meetings rather than contacting anyone, and its output travels as notes, clips and record fields into systems the customer already runs. Graded in the middle of the band with that scope stated rather than at the floor, since the floor asserts a sending product without discipline and this does not send.

The one adjacent surface is the generated follow up email, which is drafted after a call and, if sent, leaves through the user's own mail platform under that platform's authentication and reputation rather than the vendor's, so deliverability consequences sit with the customer's existing provider. Nothing published addresses volume, pacing or safeguards on those drafts. Ask whether follow up emails can be dispatched automatically and what limits apply if so.

The commercial ladder is well built and the coverage description around it is thin. Five tiers run from a permanent free plan for one person through two individual and team levels to a record synchronisation tier and a quoted enterprise arrangement, with seat minimums stated, so every size has a described and mostly priced path, and three named concession programmes address investor backed startups, nonprofits and competitive switchers with explicit eligibility.

Four buying contexts carry their own pages, covering sales, marketing, customer success and teams generally. Beyond that nothing is published on any dimension. No industry is addressed, no company size is characterised, no region, market or support location is stated anywhere reached, and no language coverage is given for transcription, which is a material omission for a product whose entire function is converting speech to text and which a buyer outside English speaking markets would need before purchase. Ask which languages transcription supports and in which markets the vendor sells and supports.