Sybill

Remove the models and nothing remains that anyone would buy. Every capability the product is sold on is inference: summaries structured around sales qualification frameworks, opportunity fields written into the record system from what was said, follow up emails drafted in the representative's voice, a query layer answering questions across every deal, and the behavioural analysis that reads facial expression and body language from the video feed.

What would survive a model removal is a recorder and a transcript, which is not this product and is available free elsewhere. The vendor's own framing matches: it positions against an established incumbent on the quality of the intelligence rather than on capture, coverage or price, and its customer quotations are about the artificial intelligence specifically.

The behavioural layer settles it beyond argument, since inferring an emotional state from a face has no deterministic implementation at all. This is the shape the top grade on this axis exists to record, and it is worth noting the contrast with the established platforms in this category, where stripping the model leaves a complete recording and analytics product behind. Ask which model or models produce the behavioural inference and the summaries, since here that is a question about the product rather than about a feature.

One genuinely good verification control, and no gate before the writing happens. The control worth crediting is specific and rare: when the product populates a field in the customer's record system, the written value carries a link back to the exact moment in the call it was derived from, so a representative reviewing a populated opportunity can check any individual value against the sentence that produced it.

That is the correct remedy for the actual risk, which is silently wrong pipeline data, and most products that write to a record system offer nothing equivalent. Follow up emails are drafted rather than sent, keeping a person in the loop on outbound. Against that, the writing itself is unattended by design and marketed as such, with fields populating within minutes of a call ending and the vendor describing the record system as writing itself.

Nothing published describes an approval step before a value is committed, a confidence threshold below which a field is left alone, a way to review a batch before it lands, or what happens when the call classification error reviewers report causes an internal meeting to be processed as a customer conversation. Ask whether field writes can be held for review, and what happens to committed values when a summary is later corrected.

The most contestable output in the product carries no accuracy figure at all. No model provider, family or version is named for any capability, no model card exists, and no evaluation, benchmark or error rate is published for the summaries, the field extraction or the behavioural inference. The gap matters most on the last of those.

Inferring that a participant was excited at a pricing moment, or labelling someone a silent champion from their body language, is a claim about a real person's internal state derived from a video feed, and it is acted on commercially. Nothing published states how often that inference is right, what it is validated against, or what its failure modes look like across different demographics, camera setups or cultural contexts, all of which are known to affect this class of inference.

Independent accounts supply the limitation the vendor does not, noting that the analysis degrades with video quality and camera position and produces nothing at all when a camera is off, which means the signal is systematically absent for some participants and present for others without the output indicating which. The credit accounting is published in detail, which is operational transparency rather than model transparency. Ask what the behavioural inference is validated against, what its accuracy is, and how the output distinguishes a low confidence reading from an absent one.

A headline time saving with no method, and a comparative study the vendor hosts about itself. The published claim is that the product saves a representative fourteen hours a week, which is a specific and checkable sounding figure carrying no population, period, baseline or definition of the work being replaced.

Customer quotations are stronger, being attributed and pointed, including teams stating they moved entirely from the established incumbent, and independent review aggregation is favourable at a high rating across roughly a hundred and forty reviews weighted to smaller companies. One item needs recording plainly. The vendor publishes a comparison of the leading products in this category on its own blog, in which it appears favourably.

A comparative study hosted on the domain of one of the products compared is not independent evidence whatever its contents, and a reader encountering it through a search result would have no particular reason to notice where it sits. Reviewers also report a specific accuracy limitation the vendor does not: the product misclassifies call types, confusing a sales call with an internal meeting or a coaching session, which degrades the features that depend on knowing which is which. Ask how the fourteen hour figure was measured and across what population, and what the call classification accuracy is.

The product composes rather than sends, so this axis applies narrowly and the grade records that. Follow up emails are drafted for a representative to review and send from their own mailbox, and no campaign tool, sequencer, contact database or bulk dispatch exists anywhere in the product, so the obligations that attach to volume outbound do not arise.

The correspondence goes to someone who has just been on a call with the sender, which is about as clear a prior relationship as this index encounters. What sits adjacent and unaddressed is the consent question belonging to the capture rather than the sending. The vendor publishes a support article on recording consent, so the topic is treated, and its contents were not retrieved on this pass.

Recording a conversation is regulated separately from contacting someone, several jurisdictions require all parties to agree, and the obligation falls on the customer running the tool rather than on the vendor. Nothing established here describes what the product does to help a customer meet that, whether any announcement is made automatically, or how consent is evidenced afterwards. Ask what the recording consent article commits to, whether consent is captured automatically, and how it is evidenced.

A real privacy surface exists and the most sensitive processing is not on it. The vendor publishes a dedicated privacy and security collection of eleven articles covering recording consent, its stated privacy posture, who can access call recordings, deletion of meeting reports and recordings, role based access, and removing or hiding specific meetings.

Those are user facing controls rather than policy statements, and deletion and access control being documented for the person whose calls are recorded is better than most products in this category manage. The contents of those articles were not retrieved on this pass, so this records what is addressed rather than what is committed to. The gap is what the collection does not cover.

None of the eleven articles addresses the behavioural analysis: not its legal basis, not what is retained from a video feed, not whether inferences about a person's emotional state persist after the recording is deleted, and not whether a participant can decline that specific processing while still permitting the call to be recorded. Those are different consents.

A person agreeing to a recording is not thereby agreeing to have their facial expressions scored and their engagement ranked, and the material that would tell them otherwise is absent from the place a reader would look. Ask what legal basis supports the behavioural analysis, what is retained from video, and whether a participant can decline it separately from recording.

No third party data enters the product and the provenance question runs the other way. The vendor supplies no contact database, no firmographic corpus, no intent feed and no enrichment, so there is no purchased dataset whose origin a buyer needs to trace, and the exposure this axis normally captures does not arise. Everything processed arrives from the customer's own systems: their calls, their electronic mail, their messaging, their record system and their deals.

What is unestablished is the direction of travel out of that tenancy. Nothing published states whether recordings, transcripts or behavioural inferences from one customer's calls contribute to models or benchmarks used for another, which is the provenance question that matters for a product whose competitive claim is the quality of its inference.

It is a sharper question here than for a text product, because the material is video of identifiable people making expressions, and because the vendor's central differentiator is a behavioural model that must have been trained on something. Nothing published describes what. Ask whether customer recordings or derived behavioural data train or tune any model, and whether anything crosses a tenant boundary.

One capture mode is designed not to appear, and no conformance position is stated with the platforms it captures from. Integration into record systems, messaging and electronic mail runs through ordinary documented routes and raises nothing unusual.

The meeting platforms are different, because they set their own rules about how a recording is made and what participants must be shown, and the product offers capture both as a visible participant and through a separate mode the vendor's own help centre presents under its own heading as an invisible recorder. A further support article addresses whether prospects can see the notetaker on a call.

Taken together those establish that whether the other participants can observe the capture is a setting rather than a constant. Nothing published states which platforms the invisible mode operates on, how it captures without joining, whether it relies on a local client rather than a meeting participant, or what position the vendor takes on those platforms' own recording indicator requirements.

A capture method whose value is that it does not show up is exactly the pattern this axis treats cautiously, and it is unaddressed. Ask how the invisible recorder captures, on which platforms, and what conformance position the vendor holds with each.

Video of identifiable third parties is processed to infer their internal states, and no stewardship position accompanies it. The material at issue is not a transcript. It is the face and body of a person who joined a sales call, analysed for expression, engagement and attention, converted into a claim about how they felt at particular moments, and stored against a deal record for a seller to act on.

Nothing published states whether that video or the inferences drawn from it train or tune any model, which provider processes it, what retention applies to the derived behavioural data as distinct from the recording, or whether a deletion of the recording removes the inferences made from it. That last question is the practical one, because the vendor documents deletion of meeting reports and recordings without stating what happens to the analysis.

No governance document, evaluation record, red teaming artifact or independently audited management standard for artificial intelligence was located, and a compliance assertion is made in a support article rather than through a trust surface. Ask whether video or behavioural inferences train any model, what retention applies to the inferences separately from the recording, and whether deleting a recording deletes the analysis derived from it.

Consent is sought for one thing and something materially more intrusive is done. The people analysed are the customer's prospects, present on a call to discuss a purchase, and what happens to them is that their video feed is read for facial expression, body language, nods, smiles and note taking, an emotional journey is inferred for each of them, and they are labelled or not labelled a silent champion according to how they appeared to react at moments such as pricing.

Two real mitigations exist and are why this is not lower. The vendor publishes an article on recording consent, and another scripting what a seller should say when a participant asks what the notetaker on the call is, which is a vendor supplying disclosure language rather than leaving its customers to improvise. Three things sit against them.

The vendor's own help centre carries an invisible recorder as its own category and a separate article on whether prospects can see the notetaker, so the visibility of the capture is configurable. None of the eleven articles in the privacy collection addresses the behavioural analysis, so the consent the vendor helps its customers obtain is consent to be recorded, which is not consent to be scored.

And an independent reviewer based in a jurisdiction treating this class of data strictly declined to test the feature at all, on the stated ground that running emotion analysis on participants without their explicit informed consent was legally uncertain. Ask what participants are told about the behavioural analysis specifically, and whether they can decline it while permitting the recording.

The right connections for the job and nothing beyond them. Published integrations cover record systems, messaging, electronic mail, dialers and the major meeting platforms, with a dedicated help centre section running to fourteen articles, and the deeper record system connection that lets the query layer reason across deals, calls, electronic mail and messaging together is reserved for the second paid tier.

That is a coherent set for a product whose job is to sit across the places a deal is discussed, and the tier gating is disclosed rather than discovered. What is absent is any route for a customer to build something the vendor has not. No public developer documentation or programmatic interface was located on this pass, and an independent assessment of the vendor's public surface reports finding none either, which is a second look at the same question reaching the same answer.

For a product positioning itself as a context layer that other tools should draw on, the absence of a documented interface is a meaningful limit rather than a cosmetic one, since everything must flow through connections the vendor chose to build. No marketplace or partner directory was found. Ask whether a documented interface exists and at which tier, and for the full connector list.

Nothing on hosting, region, tenancy or recovery was established on this pass. No provider is named, no region is stated, no residency election is offered or ruled out, no tenancy or isolation model is described, and no recovery objective appears.

One published support article establishes that a local component exists, since it addresses whether the product will function on a work computer where the employer restricts which applications may be installed and which sites may be reached, which implies a desktop client rather than a purely server side service and raises its own questions about what that client can access. The residency gap has particular weight for this product because of what is stored.

Video recordings of identifiable people, and inferences about their emotional states drawn from those recordings, are among the more sensitive categories a business tool holds, and several jurisdictions regulate them specifically. A customer selling into or from those jurisdictions has no published answer about where any of it lives. Ask which provider and regions host recordings and derived behavioural data, whether any regional option exists, and what the local client installs and can access.

One certification asserted in a support article, and nothing a reviewer can verify. The vendor states compliance with an audited control report of the continuous type, published as a help centre article rather than through a security page or trust surface, and publishes role based access as a documented product control alongside articles on who can reach call recordings and how meetings can be removed or hidden. Those access controls are real and user facing and belong on the record.

What is absent is the verifiable layer entirely: no audit period, audit date, auditor, certificate number or report request route was located, no penetration testing statement or vulnerability disclosure route was found, and no trust portal was reached on this pass. Placing a compliance claim in a support article is a choice worth noting, because a procurement reviewer looks for a security page and a help centre is not where that search ends.

The corpus behind the question is unusually sensitive: video recordings of identifiable individuals, transcripts of commercial negotiations, and inferred behavioural data about people who are not the vendor's customers. Ask for the current report with its audit period, scope and auditor, a penetration test summary, and whether a trust surface exists outside the help centre.

Per seat rates published at both billing cycles, and a consumption model explained better than almost any credit system in this corpus. Four tiers are published: a free tier that permits querying, viewing summaries and watching recordings but not recording; a tier at thirty dollars per seat monthly on annual billing or thirty six billed monthly; a tier at ninety dollars annually or one hundred and eight monthly that adds the record system context layer; and a top tier for unlimited consumption whose price is not stated.

The annual saving computes consistently at roughly seventeen percent across both priced tiers, which was checked rather than assumed. The consumption disclosure is the strongest part. The vendor states plainly which actions draw on credits and which do not, and the answer is favourable and specific: automated summaries, pre meeting briefs and follow up emails consume nothing, and recording another meeting consumes nothing, with credits drawn only for work requested on demand.

It further publishes that credits reset weekly, do not carry over, are pooled across the whole team, and grow with seat count, and gives example actions with their credit costs. Most credit models leave exactly those questions unanswered. Two things hold it below the top band. The top tier is quoted, and the price of the add on credit bundles was not established.

And the vendor's own listing on a major review platform, which that platform attributes to the vendor, carries a retired plan name and a materially lower entry price than the current structure, so a buyer arriving that way sees figures the vendor itself has superseded. Ask the price of the top tier and of a credit bundle.

Deletion is documented as a user control, which is more than most, and everything about leaving is not. The vendor publishes support articles covering how to delete a meeting report or a call recording and how to remove or hide specific meetings, so a customer can act on individual records without raising a support request, and that is a meaningful control for a product holding video of third parties. Nothing published addresses departure.

No bulk export scope or format was located for recordings, transcripts, summaries, the accumulated deal context or the behavioural analysis, no post termination access window is stated, and no retention or deletion timeline applies at the account level rather than the record level.

The asset that would be hardest to reconstruct is the one least addressed: the product's value grows from a continuously updated context layer assembled across calls, electronic mail, messaging and record system data over time, and nothing states whether any of that reasoning layer leaves with the customer or whether only the raw recordings would.

One open question follows from the deletion controls themselves, which is whether deleting a recording also removes the inferences drawn from it. Ask what exports at termination and in what format, whether the deal context layer is included, and what deletion at account level covers.

The product operates no sending infrastructure and the grade records the scope. Follow up messages are drafted and left for the representative to send from their own mailbox through their own provider, so no vendor domain, sending reputation, authentication configuration or bounce handling enters the picture, and there is no campaign or sequencing capability anywhere in the product.

Volume is one message per meeting rather than per prospect list, which is the lightest sending profile this axis encounters. Two smaller points sit adjacent. Drafting in the representative's own voice means the message reaches a recipient looking exactly like something the sender wrote, and while that is a composition question rather than a delivery one, a buyer should know that the volume of machine drafted correspondence leaving their domain rises with adoption even though the vendor sends none of it.

And the product connects to electronic mail as a data source for its context layer, which is a read relationship with the mailbox rather than a send one and carries its own access questions. Ask what mailbox permissions the electronic mail connection requires, and whether anything is ever sent by the platform rather than by the user.

A clearly drawn market with a gap between who is addressed and who is served. The positioning is unambiguous and it is what keeps this product distinct from the general meeting assistants it is often listed beside: summaries built around sales qualification frameworks, opportunity field population in the record system, deal progression analysis and cross deal querying are all specific to selling, and a general note taker offers none of them.

The commercial ladder is broad, running from a free tier through a rate an individual can justify to a team tier and a quoted top tier, and independent review data places roughly two thirds of its users in smaller companies, so the stated market and the observed one agree. The vendor positions explicitly against the established enterprise incumbent on price and quality together.

What holds it below the top band is a gap an independent reviewer sets out in detail: the marketing addresses founders, freelancers and small agencies, while the capability those readers are drawn by requires a work address, the higher paid tier and a connected record system at the same time, so the smallest buyers are invited toward something they cannot reach. That reviewer is a competitor and the observation is checkable against the published tier structure, where it holds. No geographic or language coverage statement was located. Ask which capabilities require a connected record system, and what coverage exists outside English.