Instantly

The removal test is settled by the vendor's own packaging and this is now the clearest instance in the index. The agents run on a separate subscription entirely: the pricing page states that they draw on an Instantly Credits plan rather than a per plan allowance, so a buyer can subscribe to Outreach at 47 dollars a month and receive a complete product with no model access at all.

Strip the models and what remains is unlimited mailbox connection, a warmup network, a sequencing engine, a unified inbox, email verification, inbox placement testing, sending infrastructure, a website visitor tool, a pipeline view and a stated 450 million contact database. That is a large saleable business.

A company named for its artificial intelligence has made that intelligence an optional line item, which is stronger packaging evidence than the Bigin free tier or the Apollo entry tier. Graded C rather than lower because the bottom band is reserved for a marketed claim that fails removal, not for a competent product with genuine model features layered on.

Two opposite postures in one product and the split runs by channel. On email the agreement contains a deemed approval clause: the buyer may configure the agent to generate, schedule and send without reviewing any message, and by electing not to review is deemed to have reviewed and approved the resulting campaigns. That is the exact inverse of the unconditional review Cadivra publishes, and no guardrail, threshold, withholding behaviour or audit trail is described for the email path.

On voice the architecture is the strongest in this index after Agentforce, and the controls are ones the customer cannot switch off: calling windows enforced by the platform for the recipient's location and in no case outside eight in the morning to nine at night local time, a block list from which the buyer may not remove a number without documented renewed consent, a one call at a time throttle, an express prohibition on scripting or chaining automations to simulate bulk calling, and a reserved right to suspend on call patterns that look anomalous against the buyer's own consent records. Held at B rather than A because the channel carrying almost all of this vendor's volume is the one with no oversight mechanism at all.

Model providers are named, which remains rare in this corpus. The pricing page sells access to five major language models and names OpenAI and Anthropic on the plan card, the privacy notice names ElevenLabs as the voice technology sub processor with retention periods attached, and a buyer may supply their own provider key.

The agreement carries an explicit accuracy acknowledgement stating that generated content may be inaccurate, incomplete or inappropriate for the use case and that the buyer must review output before use. Off A on the numbers that would let a buyer judge quality: no model version, no inference location, no accuracy or error rate for the enrichment and matching that decides which person a rep contacts, and no evaluation method.

The agreement also states plainly that the vendor is not responsible for how the third party providers handle inputs and outputs, so the disclosure names the providers and disclaims the relationship with them in the same breath.

Scale claims are large and unattributed. The homepage claims more than 50,000 sales teams while the vendor's own review platform profile says more than 20,000 customers, and the two are never reconciled. A logo strip names HP, Sony, Stripe, Ramp, Revolut, Flexport, Linear and Lovable with no case study, contract scope or named contact behind any of them, and on a self serve product a logo may represent a single seat.

Three testimonials do carry real names, titles and companies, which is more than most vendors offer. The defect that decides the grade: an identical statistics block reading four times the reply rate, twice as fast to start and thirty percent higher inbox placement is attached to all three different testimonials, so the same three numbers are presented as three separate customer results with no measurement basis behind any of them.

Recorded as observed, one named testimonial is attributed to a co founder of one company on the homepage and to a head of growth at a different company on the pricing page. The vendor also publishes an annual cold email benchmark report, which is vendor run research and should be read as such.

The deepest regulatory drafting in this index and it clears the bar Aloware set. The agent terms name the Telephone Consumer Protection Act and cite its implementing rule at 47 C.F.R. section 64.1200, the Telemarketing Sales Rule, federal and state do not call registry requirements, individual state telemarketing statutes by name including the Florida and Oklahoma telephone solicitation acts, state call recording and all party consent laws, state artificial intelligence disclosure laws with California Public Utilities Code section 2874 cited specifically, biometric privacy laws, United States commercial email law, the Canadian anti spam law, GDPR, UK GDPR and the European privacy and electronic communications rules.

A separate anti spam sending policy is incorporated into the agreement with a stated order of precedence, and breach of it is grounds for immediate suspension and account closure. Two provisions go well beyond naming. First, the voice agent may only call a person from whom prior express written consent meeting the federal standard has already been obtained, and calling any number taken from a purchased, rented, scraped or third party enriched list is prohibited outright, which is a vendor drawing a consent line straight across its own contact database.

Second, a dedicated section restricts what a buyer may do with European and British contact data to three permitted uses and supplies a worked legitimate interest example. The honest counterweight: nearly every clause transfers liability to the buyer rather than assuming it, and the European artificial intelligence regulation is the one regime this estate never names.

The most complete privacy notice in this index and it is dated twelve days before grading. Legal bases are enumerated by category with worked examples. Transfer mechanisms are current rather than stale: adequacy decisions, standard contractual clauses, the transatlantic data privacy framework and the British data bridge, with no reliance on the invalidated predecessor framework that still appears in policies graded elsewhere here.

European and British representatives are appointed and named with postal addresses. A privacy officer is designated with postal address, telephone and email. Rights carry a toll free line with a service code, an authorised agent route and a named appeal channel. The global privacy control signal is honoured and explained.

Session replay inside the customer facing application is disclosed, including capture of clicks, navigation and form interactions, with sensitive fields masked where feasible. The controller and processor split is worked correctly, and unlike the scope carve outs that held Boomerang and CallSine down, the individuals in the vendor's own marketing database sit squarely inside the main notice rather than outside it.

Deidentified data carries an express commitment not to attempt reidentification. Gaps for the note rather than the grade: a single long document rather than a layered notice, and a residency answer with no options in it.

The most complete data governance record in this index, and unusual because the vendor documents its own broker status rather than obscuring it. The privacy notice describes a data cooperative and enumerates source classes: data compilers and consumer data resellers, publicly available websites, the vendor's own customers, and government sources including the postal service and the census bureau.

A statutory table states affirmatively that identifiers, protected classifications, employment information and inferences were sold or shared in the preceding twelve months, which almost nobody says out loud. A live removal page carries a fifteen business day commitment backed by a telephone route and an authorised agent process.

The piece nothing else here publishes is the outcome data: request statistics for the 2025 calendar year showing 253 requests to know, all complied with at a three day median, and 5,229 opt out and deletion requests, all complied with at a one day median. For the website visitor product the supplier is named outright, GetEmails, LLC trading as Retention.com and also as RB2B, whose customer restrictions are incorporated and enforceable by that supplier directly against the buyer.

Downstream control runs strong in the other direction too: buyers are comprehensively barred from reselling or redistributing anything obtained here, or using it to build a broker database, a look alike audience or a training corpus, with breach treated as non curable. The gap that stops this being unqualified: no individual supplier is named for the 450 million record database itself, and no notice reaches an individual at the point their record enters it.

Better positioned than the marketing suggests, and the reason is who holds the account when something goes wrong. The vendor provisions, registers, owns and operates the sold domains and mailboxes under its own name and billing, so enforcement action against that infrastructure lands on the vendor rather than on the buyer's primary sending identity. That is the opposite of the account rotation vendors where the buyer holds the credential that gets restricted.

There is no social platform automation here at all: no profile scraping, no connection requests, no automated commenting. The vendor runs genuine first party sending infrastructure, requires in writing that buyers not breach the terms of any connected mailbox account, and states a hard ceiling of one hundred connected accounts per workspace.

Two real exposures keep it off A. Reselling provisioned mailbox provider accounts and pre warmed identities for bulk cold sending sits in contested territory with those providers, and no conformance position with any of them is stated. And the sharding system swaps flagged sending addresses out immediately, which treats a reputation penalty as something to route around rather than something to correct. Recorded as observed and worth a buyer's attention: the pricing page advertises unlimited email accounts while the agreement caps connected accounts at one hundred per workspace.

Candid about mechanisms that are adverse to the buyer, which is why it holds C rather than falling below. The agreement states plainly that the vendor is not responsible for how the third party model providers handle inputs or outputs, including their use in model training. That is the cross tenant training question raised and then expressly left with the provider rather than answered.

The vendor takes a perpetual and irrevocable licence to deidentified aggregated data derived from customer data, retains all rights in it, and reserves the ability to make it publicly available. Performance data, defined broadly enough to cover session recordings, mouse movement and keystrokes outside form fields, belongs to the vendor for any lawful purpose.

The warmup network deserves separate attention: the notice states that a participant's name, address and signature contact details are shared with other participants, so the buyer's own identity becomes visible inside a pooled network of strangers, and pooled outreach success data is used to give other customers benchmarking.

The genuine mitigation, and one almost nothing else in this index offers, is that a buyer may supply their own model provider key and route inference under their own terms. There is no zero retention commitment, no statement that customer content is withheld from training, and no tenancy position.

This vendor holds the single strongest disclosure mechanism in the index and the narrowest coverage of it. On the voice channel every call must disclose the business identity, a contact telephone number, that the voice is artificially generated, and that the call is recorded where recording is enabled. The artificial identity line is inserted by the platform and cannot be overridden, and the buyer is expressly forbidden from removing, suppressing or instructing the agent to contradict it.

Nothing else graded here implements disclosure as a control the customer cannot switch off, and it is precisely the shape the European marking obligation contemplates, resting on the provider rather than the deploying buyer. The terms go further: no holding the agent out as a human, no presenting generated output as solely human written, no automated decisions producing legal or similarly significant effects without human supervision, no biometric identification or voiceprint creation from call audio, and no cloning or simulating a real person's voice without that person's documented consent, which answers directly the synthetic likeness problem flagged elsewhere in this index.

Held at B because the channel carrying essentially all of this vendor's traffic is email, where the recipient receives a machine written message engineered to read as personal correspondence and no equivalent disclosure position exists anywhere. The same reasoning held Airspeed at B: excellent on the channel that is not the product.

A documented developer interface on its own subdomain, native webhooks, export into the major systems of record and outreach tools, a partner marketplace, and an unusually open model layer where five providers are selectable and a buyer may bring their own key.

The distinctive item is a machine readable pricing document maintained specifically for language models, linked from the human pricing page under a line addressing artificial agents directly, and carrying an instruction in its own source to keep it synchronised with live pricing, plan limits and add on eligibility. That is a vendor building for retrieval rather than only for browsers, and it is the sharpest instance of a pattern this index has been tracking across the corpus. Off A because no model context protocol server was located, which four vendors graded here already ship, and because webhooks are gated to the second tier rather than available at entry.

A residency posture with no options in it, disclosed plainly, which beats silence. The notice states that information used in the services is generally stored in the United States and warns users outside it that their data will be processed under a different privacy standard than their own. Cross border transfer mechanisms are enumerated and current.

For the voice product the sub processor is named with per artefact retention periods attached: ninety days for recordings, ninety for transcripts, summaries and analyses, thirty for technical traces, with compliance records deliberately kept longer.

Off A on three counts: no region choice anywhere, no data centre or hosting provider named for the core service, and no general sub processor register located, only the one voice supplier and the one visitor data supplier disclosed in their own contexts. A data processing addendum and a technical parameters addendum are both published and were not read this pass.

The gap in an otherwise unusually complete legal estate, and that is what makes it notable. Eight published instruments were located covering terms, agent terms, privacy, cookies, data usage, a processing addendum, a technical addendum and a supplier restriction document, and not one of them is a security page. No certification is claimed on any vendor surface read, no trust centre was located, and there is no audit period, auditor, penetration test or report route.

What is published is a single paragraph naming firewalls, encryption, hashing or truncation and access controls, plus unique non shareable credentials per user and a reserved right to suspend accounts on suspected unauthorised access. Held at C rather than lower because a named control set and a published processing addendum are more than an absence.

The reason a buyer should press hard here: this platform holds authenticated access to as many as one hundred mailboxes per workspace across a claimed fifty thousand teams, provisions sending identities in its own name, and records session replay capturing keystrokes outside form fields.

A third party vendor profile lists a long certification checklist including one federal programme no cold email tool would plausibly hold, which reads as aggregator template rather than evidence and was not relied on. Re verify at the published processing addendum and the technical parameters addendum.

The new benchmark on this axis alongside Bigin, and earned differently. Three product lines are separately and fully priced: outreach at 47, 97 and 358 dollars a month with the monthly email and contact allowance printed against each, credits from a free tier through 9, then 47 to 77, then 97 to 147, then 197 and up with credit volumes attached, and bundles at 94, 194 and 555 with a ten percent annual discount shown against the monthly figure.

The pieces almost nobody publishes are all present. Add on rates are priced rather than quoted, at 87 dollars a month for extra volume and 425 for extra credits. Per action consumption is published, at five credits per lead generated, five per reply generated and roughly half a credit per enrichment row. Sold infrastructure is priced, at 15 dollars a year for a domain and 4 to 10 dollars a month per mailbox.

A no overage policy is stated, so a buyer knows they will be asked to upgrade rather than billed a surprise. And credit expiry is disclosed, at two months on monthly plans and a year on annual, which is adverse to the vendor and published anyway. A parallel machine readable version is maintained for retrieval.

Counterweights recorded as observed rather than as grade movers: the plan comparison table and the plan card give different monthly email allowances for the second tier, one hundred thousand against one hundred and twenty five thousand; published reviews complain that individually useful functions sit behind tier gates met only after purchase; and all fees are non refundable and non cancellable with no proration of any kind.

The sharpest exit position in this index, and it sits in the bottom band for an architectural reason rather than merely a contractual one. On termination the vendor may delete customer data at its sole discretion, is under no obligation to retain, recover or make it available, and may but need not offer a limited download window. All fees are non refundable and non cancellable including credits, and unused credits are forfeited on termination.

The decisive fact is the sending infrastructure. Domains and mailboxes sold under labels suggesting purchase are provisioned, registered, owned and operated by the vendor under its own name and billing, and the agreement states in terms that the buyer acquires no ownership, no equitable interest and no right to transfer, port, migrate or take control of any of it, with no administrative access. On exit the vendor may disable, reclaim, repurpose or retire it.

In cold email a warmed sending domain and the reputation attached to it are the asset that takes weeks to build, and they are precisely what a departing customer cannot take. Held above nothing at all because in life export into the major systems of record is on the price list and the buyer retains ownership of its own uploaded data. This is a different shape from the Accent D, which was silence: here the adverse position is drafted, deliberate and enforceable.

The most extensive named infrastructure in this index paired with the most aggressive architecture, and a buyer should weigh both. Real and specific: a sharding and rotation system assigning dedicated address blocks and swapping flagged ones, reputation protection and bounce detection on every plan, a global block list, a spam wording checker, a dedicated deliverability agent from the second tier, a standalone inbox placement testing product, email verification that blocks invalid addresses before send, configurable pacing, and an annual benchmark report.

The vendor also disclaims in capital letters that it cannot guarantee inbox rather than spam placement, which is more honest than the guarantees several competitors imply. Against that sit three things this axis exists to surface. The warmup network manufactures engagement between participating mailboxes, which is signal creation rather than sending discipline.

Pre warmed domains and mailboxes are sold as a product, which is the secondary domain playbook Cadivra explicitly refuses and holds its A for refusing. And swapping a flagged address out immediately routes around a reputation penalty rather than correcting the sending behaviour that earned it. B on the breadth and specificity of the named controls, with the architecture recorded as the reason it is not higher.

Segmentation runs by sending volume rather than headcount, and to the vendor's credit the volume bands are real numbers rather than adjectives: five thousand, one hundred and twenty five thousand and five hundred thousand emails a month with matching contact ceilings. Four distinct buyer pages exist for agencies, freelancers, founders and enterprise, and the agency case is the most developed, which matches a product whose economics reward many mailboxes across many clients.

Off A because no headcount band, revenue band or qualifying condition is published anywhere, and because the enterprise claim rests on a logo strip of very large companies while the architecture, the hundred account workspace ceiling and the self serve checkout all point at small teams and agencies. A vendor that names its own ceiling earns the top band on this axis, and this one does not name one.