Demodesk

Remove the models and what remains is a video recorder. Transcription in 98 languages, summaries, follow up drafts, scorecard grading, deal risk detection and the analyst that answers questions across the whole corpus are all model output, and the product is packaged as a set of named agents rather than as software with an AI feature bolted on.

Packaging confirms it at the price list, which is the test this index applies before the marketing: the entry tier is named Capture and already ships the assistant and conversational search, the main tier is named Coaching and AI, and no edition exists that a buyer could purchase without the models.

One approval gate is published precisely and placed where the risk is. The CRM concierge proposes field and record updates and the operator approves or rejects before anything is written, described as approve before push and shipped with a stated 99 percent accuracy figure, so the one agent with write access to the system of record cannot act unsupervised. Around it sit audit logging, role based access and group based permissions.

Off the top of the band on the same gap this axis keeps finding: customers can build unlimited custom agents with AI Crew, and no published statement describes what constrains those agents, what they may act on, or whether the approve before push discipline extends to them.

Two disclosures here are rare enough to lift this above the band the category usually sits in. An accuracy figure is published for the highest risk automation, 99 percent for the CRM concierge, where most vendors publish no number at all for anything. And the data path to the model is described rather than assumed: transcription is stated as EU hosted, and names, companies and personal data are stripped from a transcript before it reaches any language model.

Off the top of the band because the models themselves stay unnamed, with no provider, family or version anywhere, and because no method, sample or period sits behind the 99 percent, which on a write path into the system of record is exactly the number a buyer would want substantiated.

Breadth and attribution are both strong. More than a dozen case studies are published and linked individually, each attributed to a named person with their title and employer, among them a vice president of revenue at OMR Reviews, a business development lead at StudySmarter, a vice president of sales at WorkFlex, a vice president of strategy at HeyJobs, a head of sales at Tanso and a head of inside sales at Treatwell, alongside a logo set including Verivox, Forto, Holidu, Pipedrive, AUTO1 and WorkMotion.

A 4.9 rating is displayed from an independent review platform. Off the top of the band because the quotes are qualitative almost throughout: one figure appears, a 50 percent reduction in follow up time, and no method, period or sample accompanies it or anything else.

The axis applies weakly and is graded on what is knowable, as the convention requires for products that do not prospect. Nothing here contacts a stranger: the outbound artefact is a follow up email drafted by the assistant and sent by the rep from their own mailbox to someone they have just met. No outreach regime is named.

The regulated act this vendor does engage is recording rather than sending, and its consent machinery and its works council and data protection officer rollout process are graded on the recipient disclosure and privacy rows, where they belong.

The strongest privacy posture in the index, and what earns it is that almost all of it is shipped rather than written. Consent is a product mechanism: two step consent with a notetaker announcement in every call, plus group based recording permissions administrators control. Data handling is a product mechanism: configurable retention policies, role based access and full audit logging, certified to ISO 27001 in its 2022 revision.

Residency is absolute, EU only with no transatlantic transfer. And the vendor is the only one graded here that addresses the employee side of recording rather than only the customer side, publishing a works council and data protection officer rollout process for a system that monitors its users' own staff, which is a legal requirement in its home market that almost every competitor selling into that market ignores. Separate privacy notices exist for the website and for the platform. Not enough to move it and worth naming: those notices went unread this pass, and a data protection officer contact and sub processor list were not located.

The posture is light because the vendor sells no data. There is no contact database, no enrichment, no third party sourcing and no marketplace of records: everything in the system arrives because a customer's own rep held a conversation that the customer chose to record. Off the top of the band on the counterparty question, which this category cannot escape.

The other person on the call becomes a transcript, a sentiment reading and a set of extracted commitments held in a vendor system, and while consent to record is obtained in the meeting, no notification, access route or removal route is described for that person's copy of the record afterwards.

Exposure is low by architecture. Recording runs through published applications and platform integrations rather than around them, with native recorders for desktop and mobile alongside Zoom, Microsoft Teams and Google Meet, a browser extension, and calendar connections through official interfaces. No credential custody, no scraping, no automated action taken under a user's account on a platform that did not sanction it, and no marketing built on avoiding detection.

Off the top of the band on the standard applied to everyone: a conformance position is never stated, and a recording product depends on meeting platform policies that can change, which nothing published addresses.

The second A on this axis in the index and it is earned architecturally rather than contractually, which makes it a different and useful precedent. Three commitments stack. Customer data never trains the vendor's models, stated plainly and repeated on the pricing page. Data stays on EU servers and never crosses the Atlantic, so the training question and the transfer question are answered together.

And the third is a control no other vendor graded here offers: names, companies and personal data are automatically stripped from a transcript before it reaches any language model, so the inference step happens on anonymised text. That is the difference between promising a model will behave and removing what it could expose. Configurable retention, role based access and full audit logging sit underneath.

The distinction from the contractual A held elsewhere is worth keeping: this one lives on product pages rather than in terms, and a custom data processing addendum is offered only at the enterprise tier.

The strongest position on this axis in the index so far, and it stops just short of the top of the band. The recorder announces itself in every call rather than in supported workflows, consent runs as a two step flow rather than a policy statement, and administrators control group based recording permissions, so the person on the other side is told what is happening by the product itself.

Nothing impersonates anyone, no synthetic voice or invented persona appears, and the follow up email is drafted for a rep who sends it under their own name to someone they just spoke to. Off the top of the band on two counts. Article 50 goes unmentioned by a German vendor selling into European markets, which is the most conspicuous silence available to it. And recording extends to phone, in person and field conversations, where the announcement mechanic that works in a video call is never described.

Built for the agent era and priced that way. An MCP server runs at its own subdomain and is included on every plan including the 25 euro entry tier, marketed with an explicit portability promise: take the conversation record to any language model or data warehouse, no lock in.

Around it sit a public marketplace, a REST API and, unusually, a SQL API at the enterprise tier, CRM sync with Salesforce, HubSpot and Pipedrive, native integrations across Slack, Microsoft Teams, Zoom, Gmail and calendars, recorders published for iOS, Android, Windows and Chrome, a knowledge centre and a public status page. Placing the agent interface on the cheapest paid plan rather than gating it is the detail that separates this from a competent connector list.

An absolute residency posture with the provider and the region both named, stated on every tier rather than sold as an upgrade: EU only data storage on Azure Frankfurt, EU hosted transcription, and an explicit commitment that data never crosses the Atlantic. There is no regional tier to buy, no processing exception and no fallback described.

For a European buyer weighing exposure to United States surveillance law on a corpus of recorded customer conversations, that is a complete answer, and it is the reason the whole product is positioned as built in Germany. The gap that would matter to a global buyer, and it is a small one: no non European option is offered, so a customer needing data held elsewhere has no path.

ISO 27001 is claimed with the revision year stated, 2022, which is a precision almost nothing else here manages, and it sits alongside a dedicated security and compliance portal on its own subdomain, a public status page, audit logging, role based access and configurable retention, all enumerated in the plan comparison rather than asserted in prose. Off the top of the band on two counts.

SOC 2 appears as a line item in the enterprise tier alongside the custom processing addendum, which presents an attestation as a purchasable feature rather than a property of the service, and leaves a buyer on the published plans unable to tell whether the report exists for them. And no auditor, audit period or route to any report was located, with the trust portal itself unread this pass.

Two real prices, both cycles, and a page that behaves like a price list rather than a lead form. Capture at 25 euros per user per month and Coaching and AI at 49, with a monthly and annual toggle stating the saving at 17 percent, a complete tick and cross comparison across five feature groups, and free viewer seats on every plan so the people who only watch recordings cost nothing.

The piece almost nobody publishes is the boundary: team size is printed as up to 30 licences on both paid plans with enterprise required above that, so a buyer knows in advance where the quote based conversation starts instead of discovering it in a sales call. A 14 day trial runs with no card and the vendor states there is no book a demo gate. One gap named and not enough to move it: the AI Crew agent builder is an add on described only as usage based pricing applies, with no rate published.

Portability is stated as a design principle and given a named mechanism, which is rarer than either alone. The conversation record can be taken to any language model or data warehouse over the MCP server included on every plan, the vendor writes that the record stays the customer's and describes the platform as open orchestration with no lock in, and a REST API plus a SQL API at the enterprise tier give conventional routes alongside it.

Configurable retention gives the customer control of the corpus while they are still a customer. Off the top of the band because the contractual half is absent from the pages read: no post termination retention period, no deletion timeline, no deletion artefact and no return commitment, with the terms of use unread this pass.

The axis applies weakly to a product whose only outbound artefact is a follow up email drafted for a rep and sent from that rep's own mailbox through their existing provider, so the sending reputation at stake belongs to the buyer and the volume is one message per meeting rather than a campaign. What is knowable is therefore mostly architectural and favourable. What is absent is everything the axis normally measures: authentication guidance, bounce or complaint handling, suppression, and any statement of whose infrastructure carries the send.

The market is defined by geography and by role, and both are supported. Six role specific solution pages cover revenue leaders, sales managers, reps, revenue operations, customer success and marketing, and the geographic claim is explicit and unusual for its directness: built in Germany for the German speaking region and the European Union, with the works council process and EU residency as the reasons a European buyer would choose it.

The logo set corroborates, being almost entirely European scale ups. The licence boundary at 30 seats functions as a published segment line between the self serve product and the enterprise motion. Off the top of the band because no headcount band, customer count or size distribution appears anywhere, and because the positioning claims one person startups through Fortune 500 enterprises on the same page.