Cognlay

The removal test leaves nothing anyone would buy. Strip out the models and what remains is a comma separated file importer, a mail relay over standard protocols and a suppression list. Everything the product is sold on is generated: adaptive email writing, rewriting after silence, reply intent classification, the scoring that decides the next action, and the learning loop that feeds outcomes back into the next draft.

The pricing confirms it in the way this index has come to rely on, because the meter is denominated in the model's work rather than in seats. Both units are AI units: a managed lead is one the platform sources, imports, enriches, scores or writes at least one AI email for, and an AI email credit is consumed by every draft, rewrite, regeneration, follow up or reply draft. There is no seat price anywhere on the page and no tier that ships without the model, including the free one. When a vendor cannot construct a non AI edition of its own product to sell, the removal test has already been answered.

Top of the band on controls that ship at every price point rather than as an enterprise upgrade. Review first workflow controls and pre send quality guardrails are both listed as included on every plan including the free one, which inverts the pattern seen repeatedly here of oversight machinery gated to the top tier. Two further statements are load bearing and unusually direct.

The platform does not send from sequences the operator has not explicitly activated, which the vendor describes as removing the risk of uncontrolled sends. And access can be severed at the source, since the operator can rotate or revoke the mailbox credentials at their own provider, which the vendor states stops all further sending even if its own systems are unreachable. Any sequence can be paused or stopped at any time, and queue controls arrive on the top tier.

What holds it off the highest grade is that the vendor's own framing is honest about the rest: it operates autonomously within the parameters the operator configures. The guardrails are named and their contents are never described, no audit trail of agent actions is published, and nothing escalates by risk.

The best disclosure on this axis in the index so far, and it comes from the smallest vendor in it. Three separate things are published that almost nobody else supplies. The provider family is named rather than gestured at, with the vendor stating it uses large language models from leading providers including Google's Gemini family.

The input scope is enumerated, listing exactly what the model receives for an outbound action, namely the email subject and body, lead behaviour, reply text where present, the sequence configuration and the sender profile, and stating plainly that it is not given unrelated inbox threads for general analysis.

And the retention position is stated for the sub processor as well as for the vendor, with the claim that those providers do not retain customer data for training under their enterprise interface terms. Held off the top of the band on two counts: the naming is partial, since including one family by name leaves the others unnamed, and no model version or inference region is published anywhere.

Recorded as a stage fact rather than as concealment, in the same terms applied to other pre revenue vendors here. There are no named customers, no logos, no case studies, no quantified outcomes, no presence on any software review platform, no analyst coverage, no funding record and no team beyond a single named founder. The vendor also discloses its own beta status on the pricing page, which is more candid than most products at this stage manage.

What sits in the slot where customer proof normally goes is worth describing precisely, because it is a pattern this index will see again: the homepage carries roughly fifteen launch directory badges from product discovery sites, arranged as a trust wall, in place of a single customer. Directory placement is a distribution activity rather than evidence of outcome, and a reader cannot infer anything about whether the product books meetings from the fact that a launch site featured it. Re verify once the vendor accumulates review platform history or names a customer.

The clearest allocation of responsibility between platform and sender anywhere in this index, published by a beta stage vendor with no customers. Four regimes are addressed separately with their own detail rather than named in a list.

Under the United States commercial email rules, unsubscribe handling is built in and opted out leads are automatically suppressed from all future sequences, sends leave from the operator's own domain, the vendor commits that it does not spoof sender identities or forge message headers, and it states in writing that the sender remains responsible for ensuring the list was lawfully obtained.

Under the Canadian regime it goes further than any vendor graded here, stating plainly that express or implied consent is required before commercial email reaches a Canadian recipient, that this is the sender's responsibility, and recommending that a consent record be maintained before those contacts enter a sequence. Under the European regime it designates the customer as controller and itself as processor under the customer's instructions.

Under the Californian regime it states it does not sell residents' personal information and publishes a deletion route. Suppression is architectural rather than advisory, which is the piece that turns a policy into a control.

A real posture, stated in plain language rather than buried in a template. The controller and processor split is designated explicitly, with the customer holding controller responsibility for lead data and the vendor acting as processor under instruction, which is the correct allocation for this product shape and one that several larger vendors here decline to make. Lead data is stated not to be sold, rented or shared with third parties for marketing.

Content passed to a model for follow up rewriting is stated not to be retained beyond that immediate task. Account deletion, including all associated data, is available on request through a named address, and the Californian route is stated separately. A privacy policy, terms of service and partner terms are all published.

Off the top of the band on four counts: the policy itself was not read this pass, no data protection contact or European representative is named, no sub processor list is published even though a model provider is named elsewhere on the site, and no retention schedule is given for lead records as distinct from AI processed content.

The sourcing layer is small, honestly labelled and completely unsourced. Three of the four plans are import only, so the customer supplies the list and the vendor states in writing that the lawfulness of that list remains the sender's responsibility, which is a clean allocation. The top plan adds a monthly allowance of lead finding credits described as provider sourcing and enrichment, letting a user search and source new prospects inside the platform.

That phrasing is candid about the architecture, since calling it provider sourcing concedes the records come from third parties rather than from proprietary research, and it is more honest than the vendors who describe the same arrangement as their own database. What is missing is everything after that.

No provider is named, no licensing or permitted use position is published, no accuracy or coverage figure is given, and there is no notification route, self service lookup or removal path for the individuals whose records the platform sources on the customer's behalf.

There is no third party platform being acted against, which is what earns the band. No social network is automated, nothing scrapes, no browser extension pulls data from profiles, and the only system the product touches is the operator's own mailbox through the standard mail protocols that provider already supports. So no account belonging to the buyer is exposed to restriction by anyone else.

The custody model is disclosed rather than and the vendor explains its reasoning: it deliberately does not request broad platform interface permissions covering mail, files or calendar, does not use delegated authorisation at all, and does not store the provider account password, and the mailbox credentials it does hold are encrypted and can be rotated or revoked by the customer at their own provider at any time.

Off the top of the band because credentials rather than scoped revocable tokens is the weaker custody model by the standard applied elsewhere in this index, and the vendor has chosen it deliberately, and because mail providers' own bulk sending terms are nowhere addressed.

Top of the band, and the substance would carry the highest grade if the instrument matched it. The training question is answered directly and broadly: the vendor states it does not use customer email content, lead data, mailbox threads or sequence replies to train models, and that content processed for follow up rewriting is used only for that specific task.

It then extends the commitment one layer down, stating that the model providers it uses do not retain customer data for training under their enterprise interface terms, which addresses the sub processor gap almost every vendor here leaves open.

Around that sits an eight item list of things the product never does, covering shared sending pools, selling or sharing lead data, browsing unrelated mailbox content, training on email content, requesting broad platform permissions, sending from unconnected domains, storing the provider password, and continuing to send after mailbox access is severed. That is the second published negative capability list in this index and the most operational. Held below the top grade on the established distinction: this is published assurance rather than a contractual term, and the sole holder of the highest grade on this axis earned it in a contract.

Top of the band, on manufactured effort with unusually clean identity handling around it. What is synthetic here is the work rather than the person: every message is model written and rewritten, adapted to what the recipient did, and arrives looking like a human read the last exchange and composed a considered reply.

The sender is a real named operator, sending from their own mailbox and their own domain, and the vendor commits explicitly that it does not spoof sender identities or forge message headers, which is the first anti spoofing commitment recorded on this axis. There is no agent persona, no invented human name, no synthetic voice, no cloned identity and no area code matched to the prospect.

What keeps it in this band rather than above it is that nothing discloses the machine to the person receiving the message, and the entire product proposition is that the recipient cannot tell. Article 50 of the EU AI Act is not addressed, and the marking duty under it rests principally on the provider of the system rather than on the customer sending through it.

Bottom of the band, and the gap is a product limitation rather than only a disclosure one. The integration surface is a mailbox connection over standard mail protocols, comma separated file import, saved lead tables, and calendar booking paths on the upper two tiers. That is the whole of it.

No customer relationship platform is integrated, none is named, and no path is described for the qualified meetings and enriched lead records the product generates to reach a system of record, which matters for a product that positions itself directly against two of the largest sales engagement platforms and expects to sit in the same stack. There is no programmatic interface, no automation platform connector, no marketplace, no agent facing endpoint and no webhook documentation located. Recorded as observed rather than concluded: a documentation section exists on the site and was not read this pass, and it is the most likely home for anything further. Re verify there.

The residency question is unaddressed on every surface read this pass. No hosting region is named, no residency option is offered, no cross border transfer position is stated and no sub processor list is published, although one model provider is named elsewhere on the site.

One architectural fact does genuine work in the buyer's favour and the vendor states it plainly: the emails and replies themselves live in the operator's own mailbox rather than in the vendor's store, so the most sensitive part of the corpus is already resident wherever the customer's mail provider keeps it, and the vendor observes that if the company ceased to exist the mailbox would not disappear.

What the vendor does hold is lead records, sequence configuration and behavioural history, and none of that has a stated location. Recorded as observed rather than concluded: a dedicated security page is published and was not read this pass. Re verify there.

No certification is held or claimed, and the vendor does not pretend otherwise, which is the right posture for a beta product run by a single founder and better than asserting an attestation without naming its type. What is published in its place is a described control set in plain language: mailbox credentials stored encrypted, no storage of the provider account password, no request for broad platform permissions across mail, files or calendar, customer side rotation and revocation of credentials at the provider, and a statement that access ceases entirely once the mailbox connection is severed.

That is a privilege minimisation argument rather than an audit, and it is a coherent one. Two gaps belong on the record. There is no penetration test, vulnerability disclosure policy, trust centre, auditor or enumerated control framework anywhere. And the breach commitment is phrased conditionally, saying the vendor would notify affected users in the event of an incident, with no timeline and no policy behind it. A dedicated security page exists and was not read this pass.

Four plans with real numbers, a free tier requiring no card, and no quote only ceiling anywhere on the page, which for a metered product is rarer than it sounds. Prices run free, sixty nine, one hundred and twenty nine and two hundred and ninety nine dollars a month, against a twelve row comparison matrix that enumerates every limit that decides the real bill: managed leads, AI email credits, lead finding credits, sequence counts, connected sender counts, team seats and which controls arrive at which tier.

The decisive quality is that both meters are defined in plain language rather than left as jargon. A managed lead is a unique lead the platform sources, imports, enriches, scores or writes at least one AI email for in the billing cycle. An AI email credit is consumed by every draft, rewrite, regeneration, follow up or reply draft. Almost no metered vendor in this index supplies that, and it is what lets a buyer model cost per prospect before signing up.

Two disclosures run against the vendor's own interest and both are published: deleting a lead does not restore the allowance, and extra credits cannot be purchased during the current beta, which also discloses the beta status. Cancellation is self service and reverts to the free plan.

Top of the band, and it contains the first direct answer in this index to the question every buyer of a young vendor actually has. Asked what happens if the company is acquired or shuts down, the vendor answers rather than deflecting: the emails and replies live in the customer's own mailbox, lead records can be exported or deleted from the platform, mailbox access can be disconnected or the credentials rotated at any time, and the customer is never locked in at the infrastructure level.

The architecture backs the claim, since the correspondence itself never depended on the vendor to exist. Two further commitments sit alongside: account deletion including all associated data is available on request, and queued sends are delayed rather than lost if the service is unavailable.

Off the highest grade because the deletion runs through an email request rather than a self service mechanism with a defined window, no deletion confirmation artefact is described, and no export format or post termination retention period is stated. Both holders of the top grade on this axis publish a defined window and a defined deletion event.

Earned partly on controls and partly on a refusal, and the refusal is the more unusual half. Asked directly whether it does inbox warmup, the vendor says no, tells the buyer to use a dedicated warmup or sender health tool if the mailbox still needs reputation building, and separately states that the product is built for signal driven outbound rather than mass volume sending and that high volume teams may want to keep dedicated sender operations.

Declining to claim a capability, and naming the category of tool that should be bought instead, is a candour move almost nothing else in this index makes. The controls that are shipped run on every tier including free: sender health monitoring, pre send quality guardrails, suppression of opted out leads, safe send windows, and a hard structural governor in the form of connected sender caps that rise from one to twenty across the plans.

Architecture helps too, with no shared address pools, no shared warmup accounts, and sending on the buyer's own domain and reputation. Off the top of the band because sender authentication guidance, bounce and complaint thresholds, blocklist monitoring and inbox placement testing are all absent, and the guardrails are named without their contents ever being described.

Well specified for a product this young, and the piece that earns the band is the statement of who it is not for. The vendor says plainly that the product is built for signal driven outbound rather than mass volume sending, and that high volume teams may still want dedicated sender operations, which is a vendor pointing a whole buyer segment away from itself.

The positive definition is a named buyer type, founders, lean sales teams and agencies, backed by a partner programme aimed at the third of those and by tier structure that matches the claim, running from a solo seat through three seats to unlimited alongside sender counts that rise in step. Off the top of the band on three counts, all of them stage related rather than evasive. No headcount, revenue, industry or geographic distribution is published.

There are no customers anywhere to corroborate the claimed fit. And the site offers a Japanese language option with no stated market presence, customer or go to market reason behind it, which is an unexplained coverage claim rather than a described one.