Clodura.AI

The removal test leaves the entire original business standing. Strip out every model and what remains is a 600 million record contact database with basic and advanced search filters, org charts, technology install base data, a waterfall enrichment engine that is API orchestration rather than inference, rule based email verification through syntax, domain and SMTP checks, automated email sequences, a dialer with recording and dispositions, mailbox warmup and inbox placement testing.

That is a complete product and it is what the company sold for most of the decade since its founding in 2015. The intelligence layer is real and growing, with an agentic sequence writer and model generated company intelligence reports, but it sits on top of a data business rather than constituting one.

Worth recording as an observation rather than a criticism: the company now brands itself a GenAI go to market platform in its page titles and its own credit table prices an agent written email at the same rate as a single contact lookup.

The second A on this axis in the index, and it arrives from the opposite direction to the first. Where the other holder documents controls that constrain an autonomous agent, this vendor has built its entire agent positioning on refusing autonomy, publishing a dedicated page titled Approval Before Scale and comparison pages set explicitly against autonomous AI sales agents and named rivals.

The approval model is published as a sequence rather than asserted: the agent proposes a cadence from context and reasoning, humans review structure and intent, messaging and tone are edited if needed, explicit approval is given, and only then does execution begin. It is placed at a named position in a published six stage workflow between creation and execution, and it is unconditional rather than a setting a customer can switch off.

The decisive piece, and the first of its kind in this corpus, is a published negative capability list stating what the agent will never do: send emails autonomously, override approval settings, bypass compliance rules, or learn or change behaviour without human oversight. That last commitment addresses model drift, which nothing else graded here has touched.

The honest caveat: this is a design principle published in marketing rather than a mechanism documented in product material, no audit artefact is described despite auditability being named as a benefit for revenue operations buyers, and no risk based escalation or withholding behaviour is specified.

Bottom of the band, earned on mechanism rather than on model identity. The vendor publishes a six stage agent workflow covering input, reasoning, cadence design, approval, launch and learning, maintains separate pages titled How Atlas Thinks and How Atlas Works, and names a third party supplier by name as its default email verification engine, which is more supply chain candour than most data vendors offer about any component.

What is absent is the model itself: no provider, no family, no version, no inference location, no retention position for prompts or generated content. The single claim that comes closest is a homepage line describing a real email powered language learning model, which is both garbled and unexplained, and which raises rather than answers the question of whose real emails trained it. Neither of the two Atlas explanation pages was read this pass, and they are the most likely place for anything further. Re verify there before this row is quoted.

Real corporate credibility undercut by systematically unsubstantiated numbers. On the credit side the customer logos are substantial and named, including Bosch, HCL, Hexaware, Wipro, Monster, Quess, NoBroker and Leadec, the crowd review standing is genuine with a spread of category badges across sales intelligence, lead intelligence, sales engagement and market intelligence, and there is a disclosed funding round with named institutional investors.

Against that, every testimonial on the site is anonymised to an initial, and several carry no name at all, only a role such as SaaS Founder or VP Sales, which is a choice given the logos on the same page. And the quantified claims come in volume with no method behind any of them: 95 percent or better accuracy, 30 percent fewer bounces, three times more live connects, up to 20 percent higher inbox placement, doubled open rates, roughly eight hours saved per representative per week, 40 percent lower tool costs.

One of them, a claimed 100 percent connectivity rate on the phone product, is not credible on its face. Scale figures also fail to reconcile, with the site claiming more than 20,000 go to market teams while the vendor's own review platform profile claims more than 73,000 sales professionals.

The data protection side is addressed and the outbound side is not. On the credit side there is a stated position covering the General Data Protection Regulation and the California regime, a dedicated page for the former, a Do Not Sell My Info route, a statement that only business data is provided, and a claim of built in opt out and Do Not Contact support.

That last item is asserted in a single badge line and nowhere described, so a buyer cannot tell whether it means registry screening, a suppression list, or something else. Nothing published names a single outbound regime. There is no reference to United States commercial email law, no telemarketing rules, no calling hours, no registry scrubbing method and no call recording consent position, on a platform that ships a dialer with recording enabled and local presence numbers.

The company sells globally from India, so the applicable set is broad and unaddressed. Recorded as observed rather than concluded: terms, a privacy policy and a data processing addendum are all published and unread.

One of the more complete published legal surfaces for a data vendor in this index, graded on its existence rather than its contents. Five documents are linked from the footer of every page: a privacy policy, a dedicated General Data Protection Regulation page, terms and conditions, a published data processing addendum, and a Do Not Sell My Info route.

A processing addendum published as a page rather than supplied on request is uncommon, and a standing Do Not Sell route is the mechanism the California regime actually turns on. The vendor also states plainly that it provides business data only, which is the relevant scoping claim for a contact database. Off the top of the band on four counts. None of those documents was read this pass.

No notification route to the individuals in the 600 million record database was located, and no self service lookup by which a person could see whether they are held. No data protection officer or European representative is named on any surface read. And no sub processor list was located, which matters unusually here because more than fifty external data providers sit in the supply chain by the vendor's own account.

The vendor discloses the architecture of its supply chain, which is more than most data businesses here say about where a record came from. Email discovery is published as a waterfall querying more than fifty external finders and verifiers in sequence until a verified address returns, phone discovery runs the same pattern across more than twenty providers, and the default verification engine is named as a specific third party product.

That tells a buyer the records are brokered rather than researched, which is the honest shape of the business. Two things hold it off the top of the band and the second is the sharper. The fifty providers are counted and not named, so a buyer cannot assess the licensing chain of anything they export.

And the vendor's own pricing reveals that it tracks provenance per record while declining to surface it: the credit estimator footnotes that its figures assume the lowest cost data source and that actual cost may vary, and the pricing questions state that a phone number costs ten credits varying by provider. So the platform knows which supplier answered for each contact, prices accordingly, and shows the buyer a number rather than a source. No notification to the individuals and no removal route was located.

A straight application of the established band. The vendor ships a Chrome extension, published on the official browser store, marketed as pulling contact information and company data from any website or professional social media profile, and named on the pricing page specifically for capture from a professional network.

No method is stated, no conformance position with that network's terms is published anywhere, and nothing describes what happens to the buyer's account if the extraction is detected. Held above the bottom of the band because there is no rotation, no multiple account support, no evasion marketing and nothing that treats another platform's limits as an obstacle to be routed around, which is what separates this from the worst cases graded here. The rest of the integration surface is clean: sending runs through the customer's own mailbox by ordinary connection, and the six CRM integrations are official and individually documented.

Top of the band on a genuinely unusual commitment, held there by a question the vendor's own copy raises. The creditable piece is on the agent page, where the published list of things the agent will never do includes learning or changing behaviour without human oversight. That is a drift commitment and nothing else in this index has published one.

What is unanswered is the tenancy question, and the marketing sharpens it rather than settling it: the homepage describes a real email powered language learning model, on a platform that connects to customers' own Gmail, Outlook and generic mail accounts, runs their sequences and holds their replies. Whose real emails, and scoped to which tenant, is never addressed.

Nothing states whether customer mail content, sequence performance or reply text is used to train or improve models serving other accounts, how long prompts and generated content are retained, or which sub processor sees them. Vendors far smaller than this one have answered exactly that in writing.

Two of the established shapes are present and the worst ones are absent. Manufactured effort is the first: the agent writes deeply researched personalised cold emails that leave from the representative's own mailbox under their own name, so the recipient receives the signature of individual attention they did not receive.

Manufactured location is the second and it is sold as a feature: local presence numbers displaying an area code matched to the prospect rather than to the caller, offered for the United States, Canada and on request elsewhere, on a dialer that also records. What is absent matters and pulls the grade up within the band.

Nothing here impersonates a person, because the agent is architecturally forbidden from sending on its own, so no synthetic correspondent ever conducts an exchange under a human name, and there is no voice cloning anywhere. Article 50 of the EU AI Act is not addressed, which is worth noting for a vendor selling a generative outbound agent into European markets, and no call recording notification position is published.

Solid and specific rather than broad. Six customer relationship platforms are integrated and each carries its own documented page rather than appearing as a logo in a grid: Salesforce, HubSpot, Pipedrive, Zoho, Freshsales and Microsoft Dynamics 365. Two purpose built interfaces extend beyond that, a website form enrichment interface and a CRM cleanup interface, both aimed at data flowing in rather than out.

A browser extension is published on the official store, a public knowledge base is maintained on its own subdomain, and sending connects to any mail provider through generic protocols. The detail that earns the band is placement: programmatic interface access is listed as available on the free tier rather than gated to an enterprise plan.

Off the top of the band because no connector count, marketplace, partner application directory, webhook documentation or event model was located, and because a company describing itself as a generative go to market platform in 2026 ships no agent facing endpoint at all, while four other vendors in this index publish one.

Nothing published. No data centre region is named, no residency option is offered, no cross border transfer position is stated on any read surface and no sub processor list exists. That gap is larger here than it would be for most vendors, because this is an Indian company holding 600 million business contact records assembled from more than fifty external providers and sold to customers across Europe, North America and Asia, so personal data from several regimes necessarily crosses the platform and its resting place is unstated.

An information security certification is claimed, which usually implies a defined hosting environment, and no location is attached to it. Recorded as observed rather than concluded: this is a failure to locate documentation rather than a finding that none exists, and a privacy policy, a General Data Protection Regulation page and a data processing addendum are all published and were not read this pass. Re verify in the addendum, which is where residency and sub processors normally sit.

Top of the band and held there by a specific omission this index has now seen several times. Four compliance badges appear in the footer of every page and again in a dedicated security block: data protection, the California regime, information security certification under ISO 27001, and SOC 2. The report is offered on request in plain terms rather than routed through a sales conversation, which is better than several larger vendors manage.

But the SOC 2 claim states no type, so a buyer cannot tell whether it reflects a point in time design assessment or an operating effectiveness period, and no audit window, auditor, certificate number or issuing body is named for either attestation. There is no trust centre, no status page, no penetration testing attribution, no vulnerability disclosure policy and no enumerated control set anywhere on the read surfaces. Asserting certifications without enumerating them is weaker than it looks, and that is the whole distance between this grade and the next.

The most granular pricing disclosure of any metered vendor in this index. Three purchase models are published in full. A permanently free tier with a hundred credits. A subscription tier with twelve published credit bands running from 4,500 credits a month at ninety nine dollars monthly or eighty nine annually up to 250,000 credits at 4,969 monthly, each band showing its own volume discount from one to nine percent, in two currencies with a monthly and annual toggle.

And a pay as you go ladder with eleven published bands, plus a top up rate published as a unit price per credit. The piece that makes it exceptional is the conversion table: a published rate for every action, one credit per email found, ten per phone number, three per call minute, two hundred per warmup session, fifty per inbox placement test, and fractional rates for standard and deep verification, wrapped in a live estimator that converts a credit budget into how many of each a team can actually run.

Almost nothing else in this index enables a buyer to model cost per prospect that precisely. Every plan carries unlimited users on one shared credit pool with no per seat fee, credits roll over two months, plans can be changed or cancelled at any time, and no credit is deducted when a lookup returns nothing.

The defect that belongs on the record: the credit table and the frequently asked questions on that same page contradict each other on two of the heaviest items, giving a call minute as three credits in one place and six in the other, and an agent written email as one credit in one place and fifty in the other. A fifty fold discrepancy on the rate for the vendor's own headline feature is a serious flaw inside an otherwise best in class disclosure. Two smaller observations: the estimator footnotes that its figures assume the lowest cost data source, and a call to action at the foot of the page points at a local development hostname.

The commercial exit is clean and the data exit is unstated. A customer can upgrade, downgrade or cancel at any time from the account dashboard and keeps access to the end of the billing cycle, unused credits roll over for two months on monthly plans, and annual credits are issued upfront and valid for the full term. None of that is a data right.

Nothing published states an export route or format, a post termination retention period, a deletion timeline, a deletion confirmation artefact or a data return mechanism, and no export feature appears in the plan feature lists despite comma separated file enrichment being a documented import path.

Two things point the right way without settling it: a programmatic interface is available from the free tier upward and is a real extraction route, and credits are consumed at the point a contact is unlocked, which implies the unlocked record belongs to the buyer, though no document says so. Terms and conditions and a data processing addendum are both published and unread and are the likely home for all of it. Re verify there.

Top of the band on a deep, named control set, held off the highest grade by how those controls are sold. What is published is substantial: mailbox warmup with gradual volume scaling, human like inbox interactions and daily performance reporting; spam scoring across more than thirty filters with identification of risky keywords, headers and links and remediation guidance before send; inbox placement testing across major providers with delivery tracking, geographic analysis and provider specific insight; and a four layer verification engine applied to every contact unlocked in the platform and available for external lists too, covering syntax, domain validation, a mailbox ping and risk filtering for catch all domains, spam traps, toxic and disposable addresses.

Sending runs through the customer's own mail account, so the buyer carries their own reputation and the controls protect it. The finding that keeps it off the top is economic rather than technical, and it is a new shape for this axis: the hygiene is metered against the same credit pool as the data, at two hundred credits per warmup session and fifty per placement test, so a team under budget pressure rations the deliverability controls in order to buy more contacts.

The comparison holder on this axis includes the same tooling on every tier at no marginal cost. Also missing: no bounce or complaint threshold, no blocklist monitoring detail and no suppression mechanics, and a claimed thirty percent bounce reduction carries no basis.

Well segmented by role and by size, less coherent on geography. The vendor publishes four separate use case pages addressed to founders, sales development teams, revenue operations and enterprise go to market, which is genuine role level positioning rather than a single audience claim.

Size tiering is explicit and thresholded rather than vague: dedicated onboarding, an assigned account manager and plan customisation all unlock at fifty thousand credits a month and the page says so, and there are named programmes for startups, agencies, enterprises and venture firms. Geographic honesty is partial in a useful way, since outbound calling is stated as covering the United States and Canada with everywhere else routed to sales. Off the top of the band for three reasons.

The named customer base is heavily concentrated in Indian technology services while the marketing claims global reach, and the one testimonial attached to a large western brand is anonymised to an initial. No customer size or industry distribution is published. And the startup discount is stated as fifty percent on the pricing banner and twenty five percent in the questions below it on the same page.