Avoma

The base product is the model, which distinguishes this from vendors that layer AI onto working software. Strip transcription, summarisation, topic detection and scoring and what remains is a video recorder and a calendar tool, not a diminished version of the same product.

The entry tier is named for the assistant and consists almost entirely of model output: transcription across a claimed seventy plus languages, generated summary notes, extracted action items and generated follow up emails. That is a genuine removal test failure at the level a buyer actually purchases. Held below the top band by two things.

The platform has accreted conventional components, with a scheduler, round robin routing and a lead router that are ordinary software sold in the same subscription. And the differentiating intelligence, meaning call scoring, deal risk and forecasting, is sold as separately priced add ons, which is the vendor itself demonstrating those layers detach from the base. Ask which capabilities remain if the add on modules lapse, and what the base tier delivers without them.

The system writes into the customer record system on its own and the boundary is not published. Two way field updates from call content are a named capability of the deal intelligence module, meaning the platform reads what was said in a meeting and writes values into opportunity and contact records that forecasting and reporting then depend on. That is autonomous action on the system of record rather than a suggestion surfaced for a human.

Nothing published describes which fields it may write, whether it overwrites values a person entered, whether any review step exists before a write, or what an administrator can constrain by field or by role. The base plans are safer by construction, since they log meetings one way rather than updating fields, so the exposure arrives with the add on rather than with the platform. Generated follow up emails are drafted for a human to send. Ask which fields the two way update can write, whether it overwrites human entered values, and what approval or rollback exists.

Model output drives decisions about people and nothing about the models is published. No provider, family or version is named, no model card exists, and no accuracy figure accompanies any of the three claims that matter most. Transcription is offered across a claimed seventy plus languages with no word error rate published for any of them, which is the foundational measurement for a product whose every downstream feature reads the transcript.

Call scoring against named methodologies produces a number a manager uses to coach an individual, with no published accuracy or inter rater agreement against human scoring. Deal risk scoring drives forecast decisions with no published validation against realised outcomes. Independent review coverage separately reports recurring transcription and recording reliability complaints, which is user reported evidence in the space where vendor measurement is absent. Ask for word error rate by language, scoring agreement against human reviewers, and deal risk accuracy against closed outcomes.

Four quantified outcome claims on the homepage, none carrying a population, a period or a method. The vendor states that customers save more than four hours a week, schedule twice as many qualified meetings, increase win rates by forty percent through automated scoring and coaching, and increase quota attainment by thirty percent through deal risk alerts and win loss analysis.

Those are large numbers presented without a baseline, a cohort or a measurement window, and the two revenue claims in particular assert causation from a coaching tool to quota performance with nothing published behind them. The genuine evidence runs the other way and comes from outside: a review base of roughly thirteen hundred and sixty ratings at four point six out of five is substantial third party corroboration of satisfaction, and the same corpus surfaces a consistent reliability complaint about the recording bot failing to join or capture meetings, which is an operational finding the vendor does not acknowledge. Ask how the win rate and quota attainment figures were measured, across what population and period, and what the recording success rate is.

The platform sends in two narrow ways and neither is described in compliance terms. Generated follow up emails go to meeting participants after a call, which is correspondence within an established conversation rather than cold outreach, and scheduling links go to invitees. A lead router add on distributes inbound leads to representatives, which is internal assignment rather than contact.

So the volume and the consent exposure are both small by comparison with a campaign product, and the axis applies narrowly rather than being penalised. What is unaddressed is the boundary: nothing published states whether an external participant who asks not to be contacted is suppressed from generated follow ups, whether the scheduler honours any preference state, or what governs repeated automated follow up after a meeting. Ask whether suppression or opt out state applies to generated follow up emails, and what limits repeated automated follow up to an external participant.

Avoma's privacy policy, last updated January 2, 2026, describes data subject rights, retention for as long as the service is used, and EU-US Data Privacy Framework certification as the transfer mechanism, and its trust center lists GDPR and CPRA compliance and six named subprocessors, all in the United States.

The product carries consent controls for the people it records: a meeting reminder to participants with a recording consent disclaimer, which Avoma recommends switching on for external participants, and recording permissions that differ for internal and external meetings.

The gap is the instrument: mutually signed DPAs come only with the Enterprise plan, so a customer on Startup or Organization records external participants without a signed agreement governing that processing, and the consent notice is a setting rather than the default. Free viewers and collaborators can open recordings without taking a seat, so access to the corpus is wider than the license count.

Nothing is purchased or resold: there is no contact database, intent feed or third party corpus behind the product, so provenance turns on what Avoma does with recorded conversations, which include the speech of external participants who are not its customers. The only published training statement covers data obtained through Google Workspace APIs, which the privacy policy excludes from developing, improving or training generalized AI or machine learning models.

Whether recordings or transcripts improve transcription, summaries or scoring for other customers, whether that use can be switched off, and what basis covers an external participant's voice are not addressed.

Integration runs through sanctioned mechanisms across two distinct estates. On the meeting side the recorder joins calls as a participant through the conferencing platforms own guest mechanisms rather than by intercepting streams, which is the sanctioned route for this category.

On the data side bidirectional synchronisation runs with six named customer record and support platforms through their own published integration surfaces, alongside dialer integrations and, at the middle tier and above, an interface and webhooks. Nothing resembling credential storage or unsanctioned capture appears anywhere. Two things hold it below the top band.

No stated conformance position against any specific platform's terms was located, which is what separates the bands on this axis. And the recorder's dependence on conferencing platforms is a real exposure the vendor does not address: independent review coverage reports recurring bot join and recording failures, which is a symptom of operating as a guest on infrastructure the vendor does not control, and nothing published describes what happens if a platform changes its guest policy. Ask what the fallback is when the recorder cannot join, and whether any named platform conformance commitments exist.

Avoma holds recorded audio and transcripts of a customer's meetings with its own customers and prospects, together with derived scoring about individual representatives, and its generative features read across all of it. Its trust center names the processors involved: Deepgram and Rev for transcription, OpenAI for artificial intelligence and AWS for hosting, all in the United States, and Avoma states ISO/IEC 42001:2023 certification for its AI management system.

The training question is answered for one data source only: the privacy policy says data obtained through Google Workspace APIs is not used to develop, improve or train generalized AI or machine learning models. Nothing published says whether recordings and transcripts train Avoma's own models, or whether its AI suppliers are barred from training on them.

Recordings are kept without a time limit on every plan, custom retention policies come with Enterprise, and the terms commit to returning or deleting customer personal data within 30 days of termination.

Avoma records through a bot that joins as a named participant or through botless native recording, and the bot's name can be customized, so a visible recorder is no longer guaranteed. Its consent controls are a meeting reminder to participants with a recording consent disclaimer, which Avoma recommends switching on for external participants wherever they are, and recording permissions that differ for internal and external meetings.

Generated follow up emails go out from the host's own mailbox under the host's name, and nothing states whether recipients are told the summary and message were machine written, or what position Avoma takes on the Article 50 transparency obligations that applied from August 2026.

Bidirectional depth across an unusually wide set of customer record platforms for a vendor of this size. Two way integration is named with six systems spanning the major platforms plus mid market and support side alternatives, which is broader coverage than most conversation products manage and matters because it lets smaller buyers running less common systems get the same behaviour.

Meetings, recordings, transcripts and generated summaries log automatically to contact and deal records on every tier, with field level write back arriving through the deal intelligence add on. Conferencing platform coverage spans the major services, dialer integrations connect the calling side, and an interface with webhooks is available from the middle tier.

Held below the top band on gating and verification: interface access requires the middle tier so the entry plan is integration limited to prebuilt connectors, no developer documentation or rate limits were reached on this pass, and independent reviewers report intermittent synchronisation failures which is a reliability qualification on the depth rather than on its existence. Ask for the interface documentation and which integrations are available at the tier quoted.

Cloud only with no residency answer and an explicit boundary reported from outside. No hosting provider or region is named, no European or United Kingdom residency election is described, no tenancy model is stated, and no recovery objective is published.

Independent analysis states directly that the platform is not built for organisations requiring on premises deployment or compliance scope beyond health information handling, which is a clearer statement of the limit than the vendor makes and is useful precisely because it comes from a reviewer rather than a sales page.

The residency gap matters in proportion to the corpus, since recorded voice is personal data of external participants who never contracted with either party, and a European customer recording European counterparties needs to know where that audio is stored before deploying. Health information handling is available but gated to the top tier. Ask which regions recordings and transcripts are stored in, whether regional residency is available, and what the recovery objectives are.

Avoma states SOC 2 Type II and ISO/IEC 42001:2023 certifications, and its trust center lists SOC 2 and ISO 42001 as certified alongside HIPAA, GDPR, EU-US Data Privacy Framework and CPRA, with audit reports, policies and security documents behind an access request; the security page says the compliance report is shared with customers as required.

The same page describes hosting on AWS in the United States inside Avoma's own virtual private cloud, encryption at rest and in transit, third party penetration testing at least annually, mandatory code review, separate development and production environments, and a vulnerability reporting address with a 24 hour response policy. No audit period or auditor is shown publicly.

SSO through SAML or OIDC, SCIM provisioning and custom retention policies come with the Enterprise plan, and granular access controls with audit logs start at Organization, so a Startup customer records customer conversations without SSO.

Avoma publishes every rate: Startup at $19 per recorder seat a month billed annually or $29 monthly for up to 25 seats, Organization at $29 or $39 for up to 100, and Enterprise at $39 billed annually only with a 10 seat minimum. Conversation Intelligence and Revenue Intelligence cost $29 each per seat billed annually or $35 monthly, Lead Router $19 or $25, and add ons follow the base plan's billing cycle.

Viewers and collaborators are free, a 14 day Organization trial with every add on needs no card, and there is no setup or platform fee; expert onboarding is an optional $1,000. The advertised $19 covers recording, notes, follow ups and scheduling, while call scoring, coaching and deal forecasting sit in the add ons, so a sales seat with both runs $77 a month on Startup billed annually, and signed DPAs, SSO and retention policies require the Enterprise plan whatever the team size. Bulk import of existing recordings may carry a volume based fee with no rate shown.

Avoma's terms, last updated January 6, 2026, set out the end of a contract: the customer may export its data during the 30 days after termination and Avoma deletes it once the export is complete, customer personal data is returned or deleted within 30 days of termination with written confirmation on request, and a customer can terminate for convenience on 30 days notice, effective at the end of the current term with no refund.

Recordings can be downloaded and are stored without a time limit while the subscription runs, and after a trial an account becomes a viewer that keeps access to past recordings and notes. The terms do not state the scope or format of a bulk export: whether recordings, transcripts, scorecards and playlists leave together, and in what form.

Sending is incidental to the product and none of it is described. Generated follow up emails after meetings and scheduling invitations are the only outbound traffic, both directed at people already in a conversation with the sender, so there is no campaign infrastructure, no pooled sending estate and no shared reputation for the vendor to manage.

Nothing published states whether that mail leaves from the user's own connected mailbox or from vendor infrastructure, who configures authentication records, or what happens to bounces on a scheduling invitation. The distinction matters mildly here because follow up volume scales with meeting volume rather than with campaign size, so the risk profile is low, but a buyer whose scheduling links fail to arrive has no published guidance on why. Ask which domain sends follow ups and scheduling invitations, and who configures authentication.

The served segment is stated numerically rather than by adjective, which is rare and genuinely useful. Seat caps and minimums define each tier explicitly, at twenty five seats for the entry plan, one hundred for the middle plan and a ten seat floor for the top plan, so a buyer knows immediately which tier they belong in and when they will be forced to move.

The positioning is equally direct: independent reviewers describe the platform as the consolidated alternative to running four separate point contracts, aimed at smaller and mid market revenue teams rather than at large enterprises, and the review base of roughly thirteen hundred and sixty ratings at four point six out of five supports genuine adoption in that segment. Transcription across a claimed seventy plus languages supports international use. Two limits keep it below the top band.

The compliance gating means a regulated buyer of any size is pushed to the top tier, so the stated segment ladder and the purchasable ladder diverge for that group. And independent analysis states plainly that the platform is not built for on premises deployment or for compliance scope beyond health information, which the vendor does not say itself. Ask what the largest reference deployment is, and what the options are for a small regulated team.