Salesken

There is nothing here that is not model output. Every one of the six product lines exists only because a model can interpret recorded speech: transcription, sentiment, objection detection, competitor mention detection, live cue generation mid call, automated scoring of every conversation, summarisation, and pipeline forecasting derived from what was said rather than from what a representative typed into a system of record.

The distinction from most records in this index is that there is no deterministic substrate underneath. This product does not dial, does not send, does not store contacts and does not run sequences. It attaches to conversations conducted through the customer's existing telephony and meeting tools and returns interpretation. Remove the models and there is no residual product at all, not even a utility.

The metering matches the architecture. Licences are priced per recorded user, so the vendor's revenue scales with the volume of conversation the models process, and the quality assurance line offers a choice between analysing everything and analysing by the hour, which is explicitly a choice about how much inference to buy.

The forecasting claim is the strongest form of the position: predicting deal outcomes from conversational signal is a claim that the model sees something a structured record does not.

Ask what accuracy the transcription and scoring achieve in your languages and call conditions.

The system observes and advises without acting, which is the right posture for this product class, and two design choices go further than the category norm.

Complete coverage is the first. Quality assurance audits and scores every conversation rather than a sample. That is usually presented as a productivity claim and it is more interesting as a governance one, because sampling introduces selection bias into how representatives are assessed, and analysing everything removes it. Real time incident and compliance flagging extends the same principle to the moment rather than the review afterwards.

The second is methodological transparency. Closure probability is computed against two named published qualification frameworks rather than an unexplained proprietary score, which means a manager disputing a forecast can interrogate the reasoning against a public standard instead of accepting a number.

Nothing in the product contacts anyone or commits anything, so the autonomy risk is interpretive rather than operational.

What is undescribed sits on the other side of the relationship. Automated scoring of every call a representative makes has employment consequences, and two passes located nothing on whether representatives are notified they are scored continuously, what they can see of their own assessments, whether a disputed score can be appealed, or what audit record exists of how a score was reached.

Ask what a representative sees of their own scores and whether an assessment can be contested.

Two passes located no model, no provider, no version, no architecture and no evaluation anywhere.

The gap is unusually consequential on this record because inference is the entire product. Every output a customer buys, a transcript, a sentiment reading, an objection flag, a compliance breach alert, a score, a forecast, is a model judgement about recorded speech, and nothing states what produces any of it.

No accuracy figure of any kind is published. No word error rate for transcription, no precision or recall on compliance flagging, and no calibration evidence for closure probability, which is a prediction that can be scored against outcomes and therefore could be evidenced if the vendor chose to.

Multiple language analysis is marketed without naming a single supported language or giving per language quality, and speech recognition quality varies enormously by language and accent, so the claim is unfalsifiable as stated.

The processing question is the one a buyer should press hardest. Recorded sales conversations contain pricing, contract terms and competitive intelligence, and nothing states whether that audio or its transcripts reach a third party model provider, under whose terms, or whether any of it trains anything.

The contrast with the framework disclosure on forecasting is notable: the vendor names its qualification methodology and not its model.

Ask which provider processes call audio and whether recordings train any model.

One named customer carries the evidence and does so reasonably well. A substantial education company is identified by logo alongside four specific figures: twenty seven percent revenue increase, sixty three percent reduction in ramp time, complete visibility into all calls, and thirty four percent reduction in lead cost. A named organisation attached to quantified outcomes is materially better than anonymous claims, and ramp time reduction is a credible metric for a coaching product because faster ramp is what real time cues would plausibly cause.

What is missing behind it is the case study. No period, no baseline, no sample, no methodology and no linked write up supports any of the four numbers.

The single displayed testimonial is weaker than it first appears. It is attributed to a person with no job title and no employer, and it is phrased in the third person as a description of what the product does rather than as an account of using it, which is the shape of an excerpt lifted from a review directory rather than a customer statement given to the vendor.

Third party review volume is thin across the major platforms, so the independent corroboration that carries several other records in this index is not available here.

Presence in a large competitor's partner marketplace is a genuine external signal, though of integration rather than outcome.

Ask for the case study behind the four figures, with the measurement period.

Mostly out of scope in the sending sense and not entirely, which is why this is rated rather than dismissed.

The vendor sends nothing. There is no dialer, no email, no messaging, no sequence engine and no supplied contact data. Conversations analysed are ones the customer's own team conducted through its own systems, so none of the consent, suppression or registration questions that dominate this axis for outbound vendors apply here.

What brings the axis back into play is that compliance detection is a marketed capability rather than an incidental feature. The quality assurance line advertises real time incident and compliance flagging, and a compliance use case appears in the site navigation. A vendor selling compliance detection is making a claim about regulatory knowledge.

Two passes located nothing defining that claim. No regulatory regime is named, no rule library is described, nothing states whether detection covers financial services conduct rules, healthcare confidentiality, consumer credit and collections requirements or recording consent, and nothing describes how rules are configured, by whom, or how they are maintained as regulation changes.

For the contact centre and collections use cases the vendor addresses directly, those regimes are the reason a buyer would want the product.

Ask which specific regulations the flagging detects against and who maintains the rule set.

A data processing addendum is published as its own linked document in the site footer alongside the privacy policy and terms. That is a genuine artefact and it puts this vendor ahead of most of this index, where the common pattern is either no processing terms at all or an invitation to request them during procurement. Publishing the addendum lets a buyer's legal reviewer read the processing terms before the first call rather than negotiating for the right to see them.

The operating base is India and the customer base is international.

What two passes could not locate is everything downstream of the addendum. No subprocessor list, no retention period for recordings or transcripts, no processing location, and no described route for a data subject request from someone whose call was recorded.

The holdings make retention the sharpest of those. This platform accumulates recorded conversations, transcripts and derived analysis across an entire sales organisation, including verticals the vendor cites in finance and education, and the second voice on every recording belongs to a person who never contracted with anyone here.

The published addendum is the right foundation and the absence of a retention schedule beneath it is the gap a reviewer will find first.

Ask for the retention period on recordings and transcripts, the subprocessor list, and the deletion route for a recorded third party.

Not applicable in the supplier sense and rated accordingly. The vendor sells no contact data, operates no database, performs no enrichment and sources nothing from third parties. Every record analysed originates in a conversation the customer's own team conducted, which removes the upstream licence and consent questions that dominate this axis for data vendors.

A different provenance question applies instead, and it is specific to conversation analysis. Every recording contains two parties, and only one of them is the customer's employee. The prospect or customer on the other end of the call has their speech captured, transcribed, analysed for sentiment and objection, and retained as a record inside a third party platform they have never heard of.

That person is a data subject whose words became the raw material of the product, and nothing published describes on what basis their contribution is held, how long it persists, or what they could ask for.

The field sales line extends this to conversations recorded in person at a customer's premises, where the recording is even less likely to be evident to the person being recorded.

Derived analysis compounds it, since a sentiment reading or an objection classification is new personal data created about that individual rather than merely stored.

Ask on what basis the recorded third party's speech and derived analysis are retained.

The classic exposure in this index is absent. Two passes located no scraping utility, no social network extraction, no contact harvesting and no browser extension of that kind. The product connects to systems the customer already licenses, telephony providers, meeting platforms and customer relationship systems, through their own published interfaces, which is the sanctioned pattern.

Where exposure genuinely sits is in meeting capture, and the mechanism is undescribed. Recording a conference call can be done through a platform's native recording interface, which triggers that platform's own participant notification, or by joining a participant that records, which may not. The two routes carry materially different obligations and different standing under the platforms' terms, and nothing published states which this product uses.

The consequence falls on the customer, since it is their meeting, their account and their obligation to notify participants.

The field sales line raises a parallel question outside any platform's terms, since recording an in person conversation at a customer's premises is governed by local law rather than by a conferencing provider.

Customer relationship enrichment writes back automatically to a connected system, which is a sanctioned interface use but is undescribed as to scope.

Ask whether the note taker uses native platform recording or joins as a participant, and what notification each triggers.

The published processing addendum is the substantive artefact and is credited on the privacy axis; the framework disclosure on forecasting is a genuine interpretability contribution and is credited on oversight. Beyond those two, the stewardship surface is empty.

Two passes located no security page, no encryption statement in transit or at rest, no access control description, no tenant isolation, no retention schedule, no incident history, no responsible disclosure route and no status page.

The corpus makes the training question the central one. This platform holds the recorded sales conversations of many companies, which collectively contain pricing negotiations, discount authority, competitive positioning and customer objections, and is among the most commercially sensitive bodies of material any vendor in this index accumulates. One customer's conversations would be valuable to another. Nothing states whether any of it trains models, whether tenants are isolated, or who at the vendor can listen to a recording.

One finding compounds this rather than sitting apart from it. The site navigation advertises a status page, a change log and interface documentation, and each of those links resolves to an unrelated use case page. Operational transparency is claimed in the menu and does not exist behind it, which is a poor signal on precisely the dimension this axis measures.

Ask whether recordings or transcripts train any model, and how tenants are isolated.

Squarely applicable, and here the question is recording rather than impersonation.

There is no synthetic voice, no automated agent and no message written in someone else's name, so nothing in this product misrepresents who is speaking. What it does instead is capture the other party. Every conversation is recorded, transcribed, analysed for sentiment and objection, scored, and retained, and the person on the other end is a prospect or customer who never contracted with this vendor.

A substantial number of jurisdictions require the consent of both parties before a telephone conversation may be recorded, and several require notification at the outset of the call. Two passes located no recording notification capability, no consent capture mechanism, no default announcement, and no guidance to customers on which rule applies where they operate.

The field sales line is the sharper case. Capturing on site customer conversations means recording someone in a physical meeting, where there is no dial tone, no platform banner and no conventional moment at which notification would be expected, and nothing addresses how consent is obtained there.

Derived analysis extends the exposure beyond the recording itself, since a sentiment score is a judgement about a person recorded without their knowing.

The obligation lands on the customer, and the vendor sells the capability without flagging that the duty exists.

Ask what recording notice is available and how consent is captured in field settings.

One commercial commitment stands out and is credited. Integration with a customer's existing stack is stated to be free, published on the pricing page rather than negotiated. In a category where connector fees and paid professional services are routine, removing the charge entirely is a real term.

Coverage is adequate for the product's purpose, spanning major customer relationship systems, the common meeting platforms and several telephony providers, with automatic write back of enrichment into the connected record. Presence in a large competitor's partner marketplace is external evidence that at least one integration is built, listed and maintained by a third party's standards rather than only claimed.

The developer surface is where this falls, and it fails in an unusual way. The site's resources menu advertises documentation, interface documentation, a change log and a status page as four distinct entries. Every one of them resolves to an unrelated use case page or to nothing at all. The same menu still carries unremoved template placeholder text in several positions.

So the position is not that a developer surface is absent, which would be ordinary, but that four surfaces are advertised and none exists. A technical evaluator following those links learns something about the vendor.

Ask whether a documented interface exists, and for the integration list in writing.

Two passes located no hosting provider, no region, no country, no residency option, no single tenant deployment, no customer managed encryption keys and no subprocessor list.

The vendor operates from India and sells internationally, so a transfer question exists for every customer outside that jurisdiction and has no published answer.

What makes this heavier than a generic residency gap is the nature of the data at rest. Recorded telephone conversations are not metadata about a business relationship, they are the relationship itself: the actual words exchanged during a negotiation, retained indefinitely, in a jurisdiction the customer cannot identify. Finance is a vertical this vendor cites explicitly, and financial services firms in most markets face specific requirements on where recorded client communications may be stored and who may access them. Education is another, where recorded conversations may involve students or their families.

A published processing addendum exists and is credited on the privacy axis, but an addendum without a stated processing location leaves the transfer analysis incomplete.

The absence of a subprocessor list compounds it, since speech recognition at this scale is very often performed by a third party whose identity would determine where audio travels.

Ask in which country recordings and transcripts are stored and processed, and whether any residency option exists.

Six badge images appear on the pricing page and in the site footer with no accompanying name, no alternative text, no assessor identified and no linked report. That is the unnamed badge row in its purest form: a buyer can see that the vendor wishes to convey certification and cannot determine what is being claimed, by whom it was assessed, over what period, or against which criteria. Under the credential test there is no verb and no scope, only imagery.

Two dedicated passes located no security page, no trust centre, no named attestation, no auditor, no assessment period, no penetration test summary, no completed questionnaire, no subprocessor list, no status page and no responsible disclosure route.

The published processing addendum is a privacy artefact and is credited on that axis rather than counted here.

The gap is material because of who this product is sold to. Contact centres, financial services and compliance functions are named segments, and every one of them runs a formal vendor security review as a precondition. A customer in those sectors has nothing to submit and would have to begin by asking the vendor what its own badges mean.

The advertised but non existent status page belongs to the same pattern of claimed operational transparency.

Ask what each badge represents, who issued it, over what period, and request the underlying report.

The structure is disclosed and the figures are not, which is a middle position rather than the usual blank.

What the vendor does publish is genuinely useful. Licences are priced per recorded user. A separate platform fee applies, determined by the number of users the platform supports. Integration with the customer's existing stack is included at no charge. A buyer therefore knows the shape of the bill, knows a second recurring fee exists before it appears in a quote, and knows one common surprise charge has been removed. Most quote only vendors in this index disclose none of that.

The quality assurance line adds a real choice, offering pricing based on analysing every conversation or on an hourly analysis basis, so coverage can be traded against cost.

A bundle discount across the product lines is stated without a figure.

Against that, no number appears anywhere. Six separate calls to action all resolve to a request pricing form. The platform fee is disclosed in existence and entirely unquantified, including whether it scales in steps or continuously, so the buyer cannot model the total. An asterisk on the quality assurance pricing line has no corresponding footnote on the page.

And the page headline promises no fees while introducing a second fee whose size is undisclosed, which is defensible in letter and awkward in spirit.

Ask for the platform fee schedule and a worked total at your headcount.

Two passes located no export route, no file format, no termination provision, no notice period, no post termination retention position and no deletion commitment.

The commercial side is equally undocumented. Because pricing is quoted rather than published, no contract term, renewal mechanism or cancellation right appears anywhere, so a buyer cannot know before a sales conversation whether they are entering an annual commitment, a multi year one, or an auto renewing arrangement.

What accumulates is the difficulty. A customer of two years holds recorded conversations across their entire sales organisation, full transcripts, per representative scoring history, coaching records, compliance flags and a forecasting record. That corpus is the institutional memory of how the company sells, it cannot be reconstructed after the fact because the conversations are gone, and nothing states whether any of it can be retrieved on departure or in what form.

A published processing addendum exists and processing terms are not an export mechanism.

The field sales and contact centre deployments raise a further question, since regulated firms may have their own retention obligations over recorded client communications that survive the vendor relationship and require the records back.

Ask what an export contains, in what format, and what is deleted and when after termination.

Out of scope in the sending sense and rated as scoped rather than penalised.

This vendor sends nothing. There is no email engine, no dialer, no messaging channel, no sequence builder and no supplied contact list. Conversations reach the platform because the customer's own team conducted them using the customer's own telephony and meeting tools, so there is no sending domain, no mailbox reputation, no authentication record, no inbox placement question and no carrier relationship anywhere in the architecture.

A buyer evaluating this record should read the grade as not applicable rather than as a weakness, and should not compare it against a sending vendor's score on this axis.

The nearest adjacent concern is capture reliability rather than delivery: whether every intended conversation is actually recorded and analysed, since a coaching and compliance product whose coverage silently gapped would fail in a way that resembles a deliverability failure. The vendor claims complete visibility into all calls as a customer outcome, which speaks to coverage, and publishes no independent measure of capture rate, failed recordings or transcription drop off.

No status page exists to show whether capture was interrupted, and the one advertised in the navigation does not resolve.

Ask what capture rate is achieved and how a failed or partial recording is surfaced.

Segment definition is specific and internally consistent, which is more than most records manage.

Four buyer functions are addressed with their own material: revenue operations for forecasting accuracy and pipeline health, sales management for coaching and performance, compliance as a distinct use case, and contact centres including inbound support and collections. Field sales is served by a separate product line rather than by asserting the main product covers it, which is an honest acknowledgement that recording a doorstep conversation is a different problem from recording a call.

The verticals cited, education, finance and software, are corroborated by the one named customer, which sits in education.

Scale is legible from the commercial design. Pricing per recorded user plus a platform fee scaled to supported users, together with an enquiry form offering team size bands running past a thousand, indicates the buyer is a large call volume organisation rather than a small team. Quality assurance across every conversation only pays for itself at volume.

Multiple language analysis is claimed, which supports international positioning.

The gaps are all specification. No supported language list, no country list, no residency options and no stated minimum team size, so a buyer cannot confirm fit without a call.

Ask which languages are supported at production quality and what the minimum viable deployment size is.