CallSine

The strongest packaging evidence on this axis in the index, because the SKU is an agent rather than a seat and the tiers are named after employment status. Pilot Agent buys one campaign agent, Full Time buys two expandable, and the custom tier buys coordinated multi agent workflows; nothing on the price list is denominated in users.

The product is a pipeline of specialised models, one for research, one for writing, one for deliverability, with a fourth adjusting strategy from prospect response. Remove them and there is no residual product, only a mailbox connection and a contact list.

The vendor also argues the axis explicitly against its competitors, calling legacy platforms glorified templating that force AI onto systems never built for it and dismissing prompt stuffing and if then logic as insufficient, which is a claim the removal test happens to support.

Marketed as autonomous without qualification: agents that research, generate and deploy outreach automatically, outbound agents that work while you sell. No approval step, human sign off, withholding guardrail, escalation path or audit trail of agent actions is described anywhere. Three real constraints exist and none of them is oversight. Throughput is hard capped by tier at 20 and 200 net new contacts per day, which is a rate governor rather than a review.

Pre launch testing appears on the upper two tiers, which is a check before a campaign rather than on any message. And the pipeline is described as quality controlled with a dedicated agent per stage, which is models checking models. The privacy policy then closes the loop in the vendor's favour: the customer assumes all responsibility and liability for any and all content generated through use of the platform. The system deploys on its own and the buyer owns whatever it says. Direct contrast with Cadivra, graded in the same session, where approval is unconditional and nothing sends without a human.

A different shape of disclosure from Cadivra's, and substantial in its own way. The architecture is published rather than gestured at: a multi agent pipeline with named stages for research, writing and deliverability, each handled by a dedicated agent, built on RAG and embeddings over the customer's own library.

The research agent's sources are enumerated on the homepage as website, LinkedIn and Glassdoor, and naming Glassdoor is a disclosure most vendors would quietly omit, because it means employee reviews of the prospect's own employer can inform the pitch.

Most notably, section 12 of the privacy policy publishes an explicit acknowledgement that AI carries a margin of error varying by application, dataset and model, that such systems are not perfect and can produce unexpected, unintended and erroneous results. The reason this is not higher is that the same company's marketing says the opposite: that training on the seller's own materials significantly reduces hallucinations, ensuring that every message is factually accurate. The legal instrument disclaims what the sales page promises. No model provider, family or version is named.

Bottom of band. The company is real and dated, with a formation and platform release announced in trade press in June 2024, a patent pending claim on the application, and a product blog carrying dated releases including the 120M contact database and the AI chat agent. What is absent is any evidence of results: no named customer, no case study, no quantified outcome with a basis, and no presence located on any review platform.

Marketing claims accuracy improvements and engagement lift without a measurement basis. Three shipping care tells recorded as observed rather than concluded, in the same class as Alta's unfilled link placeholder: the footer copyright reads 2025 across every page, the Google Tag Manager container ships to production as the literal placeholder GTM-XXXX, and the meta description that propagates to every page and every social card contains a grammatical error and a stray trailing character.

Four regulated surfaces ship, email, LinkedIn, calling and SMS, and no posture governing the product's outreach was located: no regulation named, no consent framework for the recipients, no suppression list, no do not call handling for the voice and text channels.

The compliance material that does exist concerns CallSine's own marketing rather than the customer's campaigns, and some of it is genuinely well done: EEA, UK and Swiss recipients are contacted only on legitimate interest or consent with a free objection route, and text opt in consent data is committed never to be sold or shared with third parties, which is a real TCPA adjacent statement.

There is also an honest negative disclosure worth crediting, that the company has no mechanism for responding to browser do not track signals. None of it reaches the outreach the platform performs on the buyer's behalf.

A substantial instrument, better than most in this index, with one structural gap that keeps it off A. What is there: an explicit data controller designation for the EEA, UK and Switzerland; legal bases cited by GDPR article rather than asserted, covering 6(1)(a) consent, (b) contract, (c) legal obligation and (f) legitimate interests; EU and UK Standard Contractual Clauses named for onward transfers; a complete CCPA and CPRA section running to right to know, deletion, opt out of sale and sharing, correction, non discrimination and limitation of sensitive data use, with an authorised agent process and an appeal route; a data protection officer contact; a dated data governance policy effective 10 May 2026; and the candid admission on do not track noted above.

The gap is scope, and it is the same shape as Boomerang's. Section 1 limits the entire notice to subscribers, authorised users and people who visit the site, attend events or contact the company. It does not reach the prospects, and the prospects are the point of the product: the people held in the 120M contact database, and the individuals the research agent profiles from their website, LinkedIn profile and Glassdoor. Nothing addresses their rights, and no Article 14 style notification exists.

Two data layers and only one is even partly explained. The research layer is disclosed with unusual specificity, naming website, LinkedIn profile and Glassdoor as the sources the agent scrapes for each prospect, which at least tells a buyer what the resulting message is built on.

The database layer is not disclosed at all: a 120M plus verified business contact database is shipped as a product line, and no supplier, licensing arrangement, collection method, refresh cadence, accuracy measure or data broker registration was located, and no route exists for a person in it to see or remove their record. The privacy policy does state that CallSine will not give, sell, rent or loan personal information to third parties, which addresses onward sale but not acquisition. Apollo holds the A on this axis for naming three sourcing routes, naming the extracting features and publishing a genuine Article 14 discharge, and none of those pieces is present here.

Two distinct exposures and neither is addressed. LinkedIn outreach is included on every tier including the 349 USD entry, and nothing states whether it automates actions against the platform or queues a task for a human, which are very different bands. Separately, the mailbox architecture is the volume infrastructure pattern: one inbox on Pilot, up to ten on Full Time, up to one hundred on the custom tier, alongside a dedicated Email Infrastructure product line.

Connecting a hundred mailboxes to one outbound engine is how cold email at scale is built, and it is the exact practice Cadivra refuses in the same session on the grounds that a buyer should send from the address they already own.

The Gmail connection itself is done properly and is the mitigating half of this grade: OAuth 2.0 with the requested scopes enumerated, revocable by the user at any time through Google account settings, which is the revocable scoped OAuth side of the custody line rather than the credential holding side.

The cross tenant training question is answered in writing for the most sensitive input, and the commitment is unusually flat. Under Limited Use of Google User Data the vendor states that data obtained through Google Workspace APIs is not used to train, improve or develop generalised AI or ML models, is used solely for the core functionality of sending, viewing and managing mail, and is not used for advertising, marketing or analytics.

A separate retention clause states that Google Workspace API data is not retained beyond immediate use, is processed in real time and is not stored, and the Gmail section adds that email content is not stored on CallSine servers and is not shared with third parties. That is stronger than Autotouch's personalised training only clause, which is the nearest comparison in this index.

B and not A because the commitment is scoped to one input and the vendor's own headline differentiator sits outside it: RAG embeds the customer's entire sales and marketing library, pitch decks, case studies, positioning and pricing collateral, which is the most commercially sensitive corpus a company owns, and nothing published addresses tenancy, training or retention for it. The same silence covers the contact database and the LinkedIn, call and SMS channels.

Manufactured effort is stated more explicitly here than by any other vendor graded: messages that feel human written, cold emails your team would actually send, personalise faster not lazier. Four channels carry it, including calling and SMS where the recipient's expectation of a human is strongest, and no Article 50 position, disclosure line or acknowledgement of the obligation exists.

The research inputs sharpen it, because a message can be shaped by what a prospect's own employees wrote about them on Glassdoor, which the recipient has no reason to expect. The genuine mitigation is a first for this index and deserves naming: the vendor states that the system is hardcoded to keep personalisation at a professional level so that messages never cross into feeling invasive.

That is the only published ceiling anywhere in the corpus on HOW personal personalisation may get, and describing it as hardcoded rather than configurable is the stronger form. Held at C because the ceiling governs tone rather than disclosure, and the recipient still has no way to know a machine wrote it.

A CallSine API was announced at launch for integration with CRM and sales enablement systems, and five product surfaces are published, AI Agent, Sales Engagement, Generation, Email Infrastructure and Contacts. Against that, no named connector to any CRM or sequencer was located, no public developer documentation, no marketplace or app directory listing, and no third party verifiable evidence of any integration operating at object level, which is the standard Autotouch met for a higher grade. The packaging detail that matters most here is that data integrations are listed only on the custom tier, so a buyer on either published plan does not get them at all.

A residency position is stated plainly rather than left to inference: personal information is transferred to and maintained in the United States, with the notice telling non US individuals in terms that if they do not consent to US processing they should not use the service.

For EEA, UK and Swiss transfers the safeguard is named rather than asserted, EU or UK Standard Contractual Clauses as applicable, with a commitment to transfer only to recipients providing an adequate level of protection. That is the 9Lenses precedent applied and then improved on, since 9Lenses disclosed US storage clearly but leaned on an invalidated framework, while this names current instruments.

Not A because there is no regional option, no hosting provider or sub processor list, and no statement of where model inference runs, which matters on a platform whose pipeline sends prospect research and the customer's own sales library to models.

Recorded as observed rather than concluded, and held at C for consistency with Boomerang rather than because the evidence is weak. A trust centre is published and linked in the site footer, hosted on Vanta, which is a compliance automation platform whose presence normally indicates an active certification programme; retrieval returned a robots directive disallowing automated access, so its contents could not be read.

No certification, audit report, control set or penetration test is claimed anywhere on the main site, and the privacy policy's security section is the generic clause that reasonable precautions are taken and internet transmission cannot be guaranteed. Two specific commitments do exist and are worth crediting: email content is not stored on CallSine servers, and Gmail data is encrypted and accessed through the API in real time. A certification most likely exists. Re verify directly at the Vanta trust centre before quoting this grade.

Three tiers published with real numbers, Pilot Agent at 349 USD a month and Full Time at 999, plus a custom tier, and the usage economics are published alongside them, which is the rare and genuinely useful part. Throughput is stated as a hard cap: up to 20 net new contacts a day on Pilot and up to 200 on Full Time with no total cap, unlimited on custom. Inbox counts are stated at one, up to ten and up to one hundred.

Agent counts, analytics retention of 14 days against all time, and the availability of pre launch testing, team collaboration and data integrations are all shown per tier. A buyer can therefore model cost per prospect contacted, which almost nothing else in this index enables. Not A for one decisive reason: every tier including the 349 USD entry routes to book a demo, with no self serve checkout anywhere, so the published price is an anchor for a negotiation rather than a purchase path. Recorded as observed: a stray fragment reading one part time agent renders below the footer, suggesting a tier that was removed or has not shipped.

Unusually, a post termination position is actually stated, and what it states is adverse to the customer on both sides at once. On termination or expiry the vendor will retain customer data for the maximum period permitted by state and federal law, and simultaneously has no obligation to maintain that data and the right to delete or destroy all copies. So the buyer can neither compel deletion promptly nor rely on preservation.

A deletion route does exist, purge on request by an authorised member of the customer's organisation, with the honest caveats that latency applies and backups may persist, plus a carve out that information held in common with other users is not deleted from server files. No export path, format, scope or portability commitment for contacts, research, playbooks, campaign history or generated content was located. Above D because governing terms, a deletion route and a stated retention position all exist, which is more than Accent had.

Structurally serious and specifically thin, which is an unusual combination. Deliverability is not an afterthought here: it is a named stage of the multi agent pipeline with its own dedicated agent, and Email Infrastructure is a standalone product line with its own page, so the vendor has organised its architecture and its roadmap around the problem rather than bolting on a warmup toggle.

The published daily caps on net new contacts, 20 and 200 by tier, are a real volume governor of the kind most tools leave entirely to the operator, and pre launch testing on the upper tiers is a genuine pre send check. What is missing is every specific control: no mailbox warmup, SPF, DKIM or DMARC checking, blocklist monitoring, bounce or complaint threshold, inbox placement testing or domain health reporting was located, so the apparatus is asserted at the level of having an agent for it. The multi inbox architecture cuts both ways, distributing volume but also being the pattern that puts a hundred sending identities behind one campaign.

The ladder is internally coherent and pitched above the solo operator: 349 USD a month for a single agent and one inbox, 999 for teams scaling personalised outbound with up to ten inboxes, and a custom tier for organisations running coordinated agent workflows across up to a hundred inboxes with data integrations. An enterprise application plus an API was the launch framing, and published use case material points at software vendors.

What is absent is any evidence the segment has been reached: no customer named at any tier, no geography beyond a US legal address, no vertical concentration demonstrated, and no enterprise apparatus described such as single sign on, role based access or administrative controls. A stated market position rather than a demonstrated one.