ENAI

Remove the models and nothing remains to sell. The unit of work is an instruction stated as an outcome, after which the system plans the steps, reads market signals, assembles an account brief, maps the operators, economic buyers, evaluators and procurement path around a decision, prepares an approved first touch, qualifies the reply and hands the account over.

There is no tier, module or entry plan that ships without this, no mechanical product underneath it, and no seat based alternative. The vendor's own framing puts the model at the centre and then argues about its constraints rather than its presence: the stated bottleneck in enterprise adoption is not generation but trust. This is the clearest application of the removal test in the affirmative direction that the index has graded.

The published claim set is the most complete on this axis outside Agentforce, and it is worth listing because the specificity is unusual. Authority is customer defined over what the system may research, prepare, approve, send or escalate. Approval rules cover voice, claims, volume and exclusions. An audit trail is claimed at the level of the individual message, tracing back to the signal, the source and the rule that permitted it.

Sensitive accounts, unsupported claims and exceptions route to a person before action. Actions are described as constrained, logged and reversible. And a negative list is published under the heading of what the system will never send, naming messages outside approved voice and claims, outreach to excluded accounts or segments, and volume beyond the customer's pacing rules, which no other vendor here has written down.

Held at the top of the band rather than at A because none of it is evidenced. No configuration surface is documented, no rule syntax is shown, no audit artefact is published, and the security page places the control review inside the deployment conversation rather than in public. The vendor offers recorded product runs as the evidence and those were not inspected this pass.

For a vendor whose central promise is explainability, the model itself is undisclosed. No provider, family, version or hosting arrangement is named on any page read, no accuracy or error rate is published for the research, the committee mapping or the reply qualification, and no fallback behaviour is described for a signal the system reads wrongly.

What is published is a transparency claim about outputs rather than about the system: every message is said to trace back to its source context and to the rule that allowed it, and an architecture diagram sets out ontology, governance, toolchain, application and agent layers. Tracing an output to its evidence is a real and useful form of disclosure, and it is a different question from what produced the output. The distinction matters most here because the account brief and the buying committee map are inferences presented to a seller as fact.

No named customer, no logo, no case study, no quantified outcome and no independent review was located across the homepage, the company page, the security page and the privacy policy. Two things distinguish this from the other vendors at this grade and belong in any comparison.

First, the vendor substitutes a different kind of proof rather than filling the slot with borrowed statistics: unedited recordings of the product working a live workflow, offered under a heading about evidence before promise, with the explicit statement that the runs rely on no deck, no staged dashboard and no outcome claim.

Second, the previous version of this site carried exactly the claims now absent, being fifty percent more qualified leads, prospecting time cut by seventy percent and a tenfold sales team, and those still surface in third party search indexes. This vendor removed its own unsourced multipliers. The grade reflects that customer outcome evidence is absent, which it is, and the note records that the absence is deliberate rather than concealed. Recognition on file is selection for the London agentic showcase at Web Summit Qatar 2026.

The control vocabulary is stronger than most of this index, and it is aimed at the wrong target for this axis. Exclusion lists, pacing rules, approved claims and an escalation path are all published as customer configurable, which is a real answer to how a buyer keeps outbound inside its own policy. What none of it does is name an obligation.

No consent standard, no unsubscribe mechanic, no suppression duty, no regional filtering and no sending statute appears anywhere, and the omission that bites hardest is domestic: a London company selling outbound execution into European markets never mentions the Privacy and Electronic Communications Regulations, the instrument that actually governs its customers' unsolicited approaches. Data protection appears once, on the assurance roadmap, framed as contract review rather than as a duty transferred to the sender.

A current and competently drafted policy, effective 10 July 2026, naming the legal entity and a London address, splitting the controller and processor roles correctly, enumerating rights including portability and the right to complain, addressing the Google interfaces specifically with Limited Use adherence, and stating that customer content is not sold. What holds it here is a set of missing specifics that a buyer in a regulated market asks for first.

No data protection officer or representative is named, no supervisory authority is identified, no sub processor is named beyond the categories, no transfer mechanism is identified for international processing, and no retention period is stated. The structural gap is the same one this index keeps finding: the policy processes business information obtained from public and commercial sources, and the people in that information are addressed only by routing their requests back to the customer. Where the vendor sources data itself it is closer to a controller than a processor, and the policy never occupies that role.

One provenance sentence exists and it is a class rather than a source: business information provided by customers or obtained from permitted public and commercial sources. That is more than several data handling vendors here manage, and the word doing the work is permitted, which is asserted rather than evidenced. Two things keep it at the top of the band.

Much of the raw material is the customer's own licensed data flowing in from named systems, being Salesforce, HubSpot, Gong, 6sense, ZoomInfo, Clay and Apollo, so this vendor is not selling a standing inventory of records about people who never contracted with anyone. And the market signal layer works on company level change rather than on personal records.

Against it, no supplier, licence or lawful basis is named for the research the agent performs on individuals, and a person mapped into a buying committee has no notice and no removal route from either the vendor or the policy.

The exposure this axis usually measures is absent by design. There is no social automation, no browser extension, no credential rotation, no proxying, no scraping described and no evasion language anywhere on the site. Connections run through official interfaces into systems of record and engagement platforms the customer already licenses, and the Google interfaces carry a stated Limited Use commitment in the privacy policy.

Held off the top by an exposure the index has not graded before and should watch for again as execution layers spread: this product reads from ZoomInfo, Apollo, Clay, 6sense and Gong and acts on what it finds, and the data licences behind several of those platforms restrict export and derivative use by a third party system. Whether a customer may pipe licensed contact and conversation data into an outside execution layer is a question about their contracts, and nothing here addresses it.

The training commitment appears in two places and one of them is the privacy policy rather than a marketing page, which is the distinction that matters on this axis: customer data is not used to train generalized artificial intelligence or machine learning models. Tenant separation is claimed alongside it, with workspace access and tenant controls described as keeping customer context separated and permissioned, and encryption in transit and at rest is stated in both documents.

Held at the bottom of the band for three reasons. The qualifier generalized carries the same limitation the index has flagged before, ruling out shared model training and leaving the vendor's own models unaddressed. Tenant boundaries are described as designed to separate rather than as enforced. And retention and deletion terms are stated to be agreed as part of deployment, so the stewardship terms a buyer would actually rely on are the ones nobody outside a sales process can read.

The message that reaches a person is researched, drafted and paced by a system, and arrives as a first touch from a company under its own brand and seller. Nothing simulates a human identity: there is no persona, no invented correspondent, no synthetic voice and no manufactured location.

The mitigation is unusual and belongs on the record, because it governs the content rather than the disclosure: approved claims and approved voice are customer set constraints on what the recipient can be told, and the never send list rules out messages outside them. That is closer to accuracy governance of a cold approach than anything else graded here.

Held at C because the effort is manufactured in the sense this axis measures, the recipient learns nothing about how the message was produced, and Article 50 of the European artificial intelligence regulation, in force since August 2026, goes unmentioned by a vendor selling generated outbound into European and Gulf markets.

Ten integrations are named and grouped by the job they do, which is a clearer presentation than most: Salesforce and HubSpot as systems of record, Gong, 6sense and ZoomInfo for signals and context, Clay, Apollo and Slack for workflow and data, Outreach and Salesloft as engagement channels. A public status page runs on its own subdomain.

The reason this sits at the top of a middle band rather than higher is that integration depth is the load bearing claim of the whole architecture, since the product's pitch is that a customer keeps every existing system and adds an execution layer above them, and none of the mechanics are described. No sync direction, no object mapping, no authentication model, no rate behaviour, no public interface documentation, no webhooks and no agent endpoint were located. A dedicated integrations page exists and was not read this pass.

No hosting country, region, cloud provider or residency option is stated anywhere across the four pages read. The privacy policy addresses transfers only in the abstract, saying information may be processed in countries other than the one where it was collected and that contractual and organizational safeguards are used where required, without naming a mechanism, and the security page defers deployment specific requirements to documentation agreed with the customer.

One point deserves care rather than credit: the homepage carries a region list reading London as headquarters, then North America, Europe, the Gulf, Asia Pacific and Australia, under a heading about cross border execution. Read closely it describes the markets the product is configured for rather than where data is processed, and a buyer scanning for a residency answer could easily take it for one. For a British vendor selling into regulated European buyers, the strongest available argument is the one left unmade.

This is the most candid version of this grade in the index and it deserves to be recorded as such. The security page publishes an assurance roadmap headed by the principle that there should be no badge before the evidence, and then states three times what the vendor does not hold: a readiness programme for the audited service organisation report with certification not currently claimed, control mapping toward the information security standard with certification not currently claimed, and data protection work described as regulatory alignment rather than a certification badge.

Set against the pattern this index keeps finding, being certifications asserted without a type, a host's certificate presented as the vendor's own, and an attestation advertised in a page description while the body evidences nothing, a plain negative disclosure is the honest posture. The grade cannot move on candour alone. No certification, auditor, report, audit period or trust portal exists, and the controls themselves are stated at the level of principle rather than enumerated. A public status page is availability rather than security.

No price, unit, tier, range, floor, pricing model or pricing page exists. The navigation and the footer carry no pricing entry across four pages read, a targeted search returned nothing for this vendor, and every call to action on the site resolves to the same scheduling link or a personal mailbox.

A buyer cannot learn whether this is priced per seat, per agent, per account worked, per meeting produced or on an annual platform fee, which is a more complete absence than the vendors here that publish a pricing page and withhold the contents. The gap is sharpened by the vendor's own standard.

This site argues that buyers should be able to inspect evidence before a pilot and that certifications should not be claimed before they are earned, and the single number a buyer needs in order to decide whether to spend an hour on an evaluation is the one thing routed through a sales conversation.

Governing terms are published and a return or delete commitment is stated, which is what holds this above the bottom of the scale: customer data is returned or deleted according to the applicable agreement and legal obligations, retention runs only as long as needed to provide the service and meet legal duties, and portability is enumerated among the rights a person may exercise. Every one of those resolves in a document nobody outside a deployment can read.

No retention period, no post termination window, no deletion timeline, no deletion artefact and no export function is named on any page read. The architecture supplies a partial answer the vendor never makes explicit, in that the systems of record stay with the customer and this layer sits above them, so the account history is not held hostage in the way it is with a platform that replaces the pipeline. What the customer would lose is the accumulated evidence, rules and mapping, and nothing describes whether those leave with them.

One genuine control sits at the top of this band and it is stated as a rule rather than a setting: volume beyond the pacing the customer sets appears on the published list of what the system will never send. Pacing as an authority boundary rather than a throttle is a better framing than most vendors manage. Everything else that constitutes sending discipline is absent.

No warmup, ramp schedule, authentication guidance, per mailbox rate, bounce or complaint threshold, blocklist monitoring or placement testing is described, and no sending infrastructure is identified at all. The integration list names two engagement platforms as channels, which suggests mail may leave through systems the customer already operates, and if that is so then the sending discipline belongs to those platforms and not to this one. The vendor never states which path applies.

The buyer is described with more precision than most: complex business to business markets with long procurement cycles, multi stakeholder buying committees and approval paths, named down to the operator, economic buyer, technical evaluator and procurement route the product expects to find. Eleven verticals are published and an initial focus of three is stated outright, being logistics and transport, healthcare, and wholesale and business to business retail.

The detail that earns the grade is a disclosure discipline this index has seen fail elsewhere: the vertical list marks three entries as having a demo available and eight as workflow models, so a reader can tell what is built from what is designed. A vendor in an earlier session published an integration strip captioned as available and planned with no way to tell which was which, and this is the inverse of that defect. Held off the top because no headcount band, deal size, seat count or ceiling is named, and eleven verticals with a stated three market focus is a description of ambition as much as of coverage.