11x

There is no product without the models and the company has never had one. Founded in 2022 as an autonomous outbound company, every capability sold is inference: prospect research, message composition, reply interpretation and response, meeting booking, and telephone conversation in the voice product. Remove the model layer and nothing remains, not a diminished tool but an empty account, because the vendor does not sell a sequencer, a database or a dialer that a human would operate.

The commercial framing makes the same point from the other direction, since the product is priced against a salary rather than against software and is described as a worker rather than an application, which is a positioning only available to a company whose entire deliverable is model output. The announced re-platforming onto a newer agentic architecture is a rebuild of the model layer itself rather than a feature addition to something else.

This is the second record in this index to take the top band here and both are agent native companies founded after 2022, which is the vintage split behaving as the convention describes. Ask what a deployment produces before any configuration.

High autonomy across two channels, with controls that exist and that users report as laborious rather than sufficient. The outbound agent prospects, writes, sends, reads replies and books meetings without a person in the loop, and the voice agent conducts calls, so both operate unattended by design.

Real configuration controls are described in user accounts rather than vendor documentation: customers report building prompts, defining rules and maintaining blacklists of names and domains, which confirms that suppression and constraint mechanisms exist and are operator managed.

The revealing detail is what users say they had to blacklist, namely their own existing customers, which indicates the agent will contact anyone in scope unless explicitly excluded rather than defaulting to a bounded target set. That is an oversight model placing the entire burden of exclusion on the operator, and multiple accounts describe the effort as extensive without producing acceptable output.

Nothing published describes approval gates, volume ceilings, or a stop control for work in flight. Ask what the agent will contact by default, what must be excluded manually, and whether outreach can be held for review before sending.

An entirely model driven product with nothing published about the models. No provider, family or version is named, no model card exists, and no accuracy, quality or evaluation figure appears for research, message generation, reply interpretation or voice conversation.

The company publicly announced a full re-platforming of its outbound agent onto a newer agentic architecture, with the incoming chief executive citing substantial product improvements, and that announcement carries no technical detail, no comparison against the prior version and no published measurement, so a buyer cannot establish what changed or whether it addressed the personalisation complaints that prompted it. That gap is the consequential one here.

Where user communities report output quality as the core failure, an evaluation figure is the only thing that could distinguish a genuine rebuild from a repositioning, and none is offered. Ask which models power research, generation and voice, what evaluation covers output before it reaches a prospect, and what measured improvement the re-platforming delivered against the prior version.

The published evidence base has been contested by the parties named in it and has not been replaced with verifiable substitutes. This grade addresses the state of the evidence rather than the merits of any allegation, and the dispute is recorded rather than adjudicated.

A March 2025 investigation by a technology publication reported that companies displayed as customers disputed being customers, with one of them confirming publicly that its relationship had been a one month trial and the other denying a client relationship, and cited former employees on revenue recognition practices and on early churn running between seventy and eighty percent in the first three months. The chief executive at the time called the claims inaccurate.

That denial is on record and is not resolved here. What follows for this axis regardless of who is right is that a buyer verifying references against the published logo set encountered contradiction from the named companies themselves, and no replacement evidence with stated population, period or method has since been published.

Independent hands on testing does exist, with one reviewer running two hundred leads over two weeks, and it reports the same personalisation gap that user communities describe. Most other third party analysis is written by direct competitors and was weighted accordingly. Ask for three current customers you source independently rather than from a reference list, and for retention measured across a stated cohort.

Three regulated channels operated autonomously and no compliance apparatus published. The outbound agent sends email and social messages and the voice agent places telephone calls, which brings messaging law, platform rules and telephone preference regimes all into scope simultaneously.

Nothing published describes consent basis, suppression list handling, unsubscribe mechanics, preference register screening for the voice product, calling window enforcement, or what obligations the vendor considers to transfer to the customer.

The one mechanism evidenced is user reported rather than documented: customers describe maintaining their own blacklists of names and domains, which places suppression entirely in the operator's hands and, by the accounts given, requires them to exclude their own existing customers manually.

A product supplying a set number of contacts as part of the subscription is also supplying the target list, which makes the consent basis for contacting those individuals a direct question for the vendor rather than a downstream one. Ask what suppression the agents apply before contacting anyone, what preference register screening applies to the voice product, and the lawful basis for the supplied contacts.

No privacy documentation was reached on the routes taken this pass. No privacy policy contents, processing agreement, subprocessor list, transfer mechanism, retention period or data protection officer was located. This records what a buyer could establish before contacting sales rather than a finding of absence.

The corpus makes the questions substantial for a company of this size: the platform holds supplied contact records and research on individuals at prospect organisations who never contacted the customer, the full text of outbound correspondence, replies those individuals sent, and in the voice product recordings or transcripts of telephone conversations. All of that concerns people whose relationship is with neither the vendor nor its customer at the point of collection.

The company operates a European origin story and sells internationally, which raises transfer questions nothing addresses. Ask for the processing agreement, the subprocessor list, the retention period for prospect research, correspondence and call records, the transfer mechanism, and how a contacted individual would exercise rights over data held about them.

The subscription includes contact data and the source is not named. Third party reporting on the commercial terms states that a set number of contacts is included in the annual price, which means the vendor is supplying prospect records rather than only acting on a list the customer brings, and that makes provenance a direct question rather than a peripheral one.

Nothing published identifies the data provider, describes whether records are licensed or gathered, states a lawful basis for holding personal data on individuals who have no relationship with either party, or offers any indemnification position. The research capability compounds it, since the agent is described as researching prospects before writing to them, which implies gathering further information about named individuals from sources the vendor does not disclose.

For a product whose entire function is contacting people, the absence of any published account of where those people's details come from is the substantive gap on this axis. Ask which providers supply the included contacts, on what basis, what the research layer draws on, and whether provenance is indemnified.

Outreach runs across a professional social network and no method is described anywhere. The outbound agent is stated to operate across email and social channels, and nothing published states whether social messaging runs through that platform's sanctioned interfaces, through an automation layer, or by another route, nor whether any conformance position is held.

The grading convention treats undescribed automated activity against a professional network cautiously by default, and that is where this sits. The voice product adds a second platform dimension, since automated telephone origination depends on carrier and numbering arrangements that nothing published describes.

A third exposure follows from the subscription including a set number of contacts, which means the vendor supplies prospect data from a source it does not name, so a buyer inherits whatever collection basis that source operates under. Nothing suggests improper conduct; the point is that three distinct platform dependencies exist and none is described. Ask how social outreach is executed and under whose terms, what carrier arrangements underpin the voice product, and where the supplied contacts originate.

An entirely model driven product and no stewardship position located. Nothing published states whether customer data, supplied contact records, prospect research, sent correspondence, received replies or call recordings train or tune any model, which providers process that material, what retention applies to prompts and generated output, or whether anything crosses a tenant boundary.

No governance document, evaluation record, red teaming artifact or independently audited management standard for artificial intelligence was located. The re-platforming announcement is the natural place such commitments would appear alongside an architectural change and it carries none.

One question is specific to this product shape and worth asking directly: because the agents learn from reply handling in order to respond, a buyer should establish whether replies received from their prospects, which are the prospects' own words, are used to improve models serving other customers. Ask whether correspondence, replies or call recordings train any model, which providers process them, what retention applies, and what governance documentation exists.

The product is designed and marketed as a person, which makes this the most direct case of undisclosed machine authorship in this index. The agents carry human first names, are sold as workers a buyer hires rather than software they license, and are positioned explicitly against employing a human in the same role.

The outbound agent composes personalised email and social messages and handles replies, so a prospect can hold a multi turn exchange and reach a booked meeting without a person having participated. The voice agent conducts telephone conversations directly. Nothing published states whether recipients are told they are corresponding with or speaking to a system, whether disclosure can be enabled by the operator, or what either agent says if asked.

The naming convention is not incidental: presenting an automated system under a human name to a recipient who has no way to know otherwise is the specific practice this axis exists to record, and here it is the product's central design rather than a side effect. Two channels compound it, since a synthetic voice on a telephone call raises the question more sharply than email does. Ask whether the agents disclose that they are automated in email, social messages and on calls, and what each says when asked directly.

Little integration detail was reached on the routes taken this pass. Customer record integration is implied by the product's function, since an outbound agent booking meetings and tracking prospects must write somewhere, and independent analysis notes that the research layer depends on well defined target profiles and high quality customer record data to ground it, which indicates a real dependency on connected systems.

Beyond that inference nothing was established: no integration directory, connector list, developer documentation, interface reference or authentication model was located, and no marketplace or partner programme was found. This is a retrieval limit rather than a finding that integrations are absent, and a platform of this funding level almost certainly maintains connectors.

Independent analysis does add one relevant observation, that a competing research layer from a general purpose enrichment platform erodes this vendor's differentiation, which speaks to the ecosystem competitively rather than to its depth here. Ask for the integration list, the interface documentation, and which systems the agent writes to.

No deployment or residency information was reached on the routes taken this pass. 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 appears.

The gap carries weight in proportion to the corpus rather than the company size: the platform holds personal data on prospects across whatever territories a customer sells into, gathered and processed without those individuals' involvement, plus the correspondence and in the voice product the conversations that follow. A European prospect contacted by an agent has their details, the research conducted about them and their reply held somewhere the buyer cannot identify.

For a product sold on the premise of operating around the clock across every time zone, where the processing happens is a question the international positioning invites. Ask which regions host prospect data, research and correspondence, whether regional residency is available, and what the tenancy model is.

No certification, attestation or trust surface was reached on the routes taken this pass. No trust centre, service organisation control report, international information security certification, penetration testing statement, vulnerability disclosure route or enumerated control page was located.

This records verifiability before a sales conversation rather than a judgement about the underlying programme, and a company holding this level of venture funding and selling twelve month enterprise contracts would ordinarily need documentation to clear procurement.

The point worth making for a buyer is procedural rather than technical: given that this vendor's published customer evidence has been publicly disputed, the case for obtaining independently verifiable security artifacts, meaning a report with a named auditor and a stated audit period rather than a summary or an assurance, is stronger here than it would be for a vendor whose published claims have not been contested. Ask for the current report with its audit period, auditor and scope, and verify certification claims against the certifying body's own register rather than the vendor's description.

No published figure, a twelve month commitment signed before deployment, and no way to try the product first. Third party reporting converges on roughly five thousand dollars monthly or sixty thousand dollars annually, including a set number of contacts, with a full annual contract required upfront and no free version, and access to any trial requiring a sales conversation. Nothing published states the rate, the contact allowance, overage terms or what happens if the agents underperform.

The structure matters more than the missing number here because of what a buyer is committing to: a full year paid for a product whose output quality independent reviewers and user communities consistently report as inconsistent, purchased before any hands on evaluation.

One term from the reporting is directly useful to a buyer and worth carrying: former employees described most early customers exercising contract break clauses to cancel, which indicates such clauses exist and can be invoked, so a buyer should negotiate one explicitly rather than assume the twelve month term is absolute. Ask for the rate and contact allowance, a paid pilot shorter than twelve months, a written break clause tied to output quality, and the overage rate.

The commercial exit is better documented than the data exit, which is an unusual inversion. Third party reporting establishes a twelve month contract signed upfront and, more usefully, that former employees described most early customers exercising contract break clauses to cancel, which tells a buyer that such clauses exist in this vendor's agreements and have been invoked at scale.

That is genuinely actionable, because it means the twelve month term is negotiable in practice and a break clause is worth requiring in writing. On the data side nothing is published: no statement of export scope or format, nothing on whether prospect research, correspondence history, reply threads, call recordings and configured prompts and blacklists leave with a departing customer, and no retention or deletion timeline.

The blacklists deserve particular attention, since user accounts describe them as extensive manual work, and a customer who rebuilt their exclusion list over months would be reconstructing it from scratch elsewhere. Ask for a written break clause, and what exports at termination including correspondence history and configured exclusions.

High volume autonomous sending with no published discipline, against a specific and widely reported quality complaint. Independent hands on testing recorded eight hundred and forty seven automated messages generated from two hundred leads over two weeks, which is sequence volume rather than spray, but nothing published describes the infrastructure behind it: whether messages leave from the customer's own connected mailboxes or vendor systems, who configures authentication records, whether warmup applies to a new sending identity, what bounce or complaint thresholds trigger intervention, or how one customer's sending is isolated from another's.

The reputational exposure is the finding rather than the volume. Competitive and user analysis converges on generic output as the consistent complaint, and generic output at scale from a domain a buyer owns is precisely how sender reputation degrades, which is the comparison one analysis draws directly against a human representative who would not burn the domain.

A buyer is therefore committing their own sending reputation to a system whose output quality they cannot evaluate before signing. Ask which infrastructure sends, whether warmup applies, what bounce thresholds stop the agent, and whether sending can be capped per day.

The fit statement that exists is written by third parties rather than the vendor, and it is narrower than the marketing. Independent analysis places the product with enterprise organisations that have budget to pilot, tolerance for reliability variance, a well defined target profile and high quality customer record data to ground the research layer, and that are willing to dedicate a revenue operations resource to configuring and monitoring agent behaviour.

It states equally plainly who should not buy, naming smaller companies requiring predictable guaranteed results. That last condition is the substantive one, because a product sold as replacing a hire that in practice requires an operations specialist to supervise is being bought for different economics than the pitch describes.

The vendor's own positioning has moved, with its blog acknowledging the debate between automated and human representatives and softening the earlier replacement framing. No industry pages, localisation or stated minimum deployment were located from the vendor. Ask what internal resource a deployment assumes, and what the smallest successful reference deployment looks like.