Project Blue

Artificial intelligence appears in three places and none of them carries the product. Call outcomes are derived automatically from transcripts rather than tagged by a representative, an intelligent detection layer chooses between the two messaging channels when a recipient is not on an Apple device, and voice memos can be generated from text with an optional clone of the sender's own voice.

Strip all three and the product still stands: a dedicated messaging line, native delivery through the consumer messaging service, synchronisation into the buyer's system of record, a dialer with recording, and an analytics dashboard. The claim made in the marketing is proportionate to what was built, which is a smaller claim than most vendors in this category make. The convention reserves the bottom band for a marketed claim that fails the removal test, not for a competent product with modest assistance added to it.

This is an automation channel rather than an agent, and the automation is real: messages fire from workflows inside the buyer's system of record, synthetic voice memos can be dispatched to hundreds of contacts without anyone recording anything, and call outcomes are written automatically from transcripts with the vendor stating explicitly that no tagging, dropdowns or representative judgement are involved.

Against that machinery the published oversight consists of two sentences of acceptable use posture: excessive or abusive use may be reviewed to maintain platform integrity, and mass outbound behaviour on the consumer messaging channel is described as against the vendor's guidelines. Neither is mechanised. A setup line item named configure message intent appears on the price card with no explanation of what it configures. Approval steps, role permissions, sending ceilings, audit records and escalation thresholds are addressed nowhere.

Several distinct model capabilities are marketed and not one provider, model family, version or hosting arrangement is identified for any of them. Text to speech generates voice memos, a separate capability clones the sender's own voice, transcription runs on every call, outcome classification runs on every transcript, and routing between the two messaging channels is described as intelligent and real time.

Voice cloning in particular is served by a small and named market of providers, so the omission is conspicuous rather than incidental. Where the training samples for a cloned voice come from, how many are required and where the resulting model is held are all unaddressed.

Fourteen customers appear with full name, employer and photograph, which is more named attribution than most vendors of this size publish, and the employers span pet retail, fitness, lighting, property, wellness, peptides, automotive and mortgage technology. One testimonial carries a quantified result attributed to a named person at a named company: appointment show rates rising from twenty five to thirty percent up to fifty to sixty percent.

A second names volume, describing thousands of messages a week. Holding it off the top band, the aggregate customer claim contradicts itself between two pages of the same site, appearing as three thousand customers generating more than eight million in revenue on the homepage and three thousand customers generating more than one million on the pricing page, a factor of eight apart inside the same sentence and the same design element.

A separate counter labelled companies working with us renders as zero to a fetcher. The twice repeated claim of double the response rate of ordinary text messaging carries no sample, period, denominator or method, and there is no case study library behind any of it.

The absence of compliance obligation is the product's headline rather than a footnote to it. The homepage promises skipping registration delays and carries two badges reading zero application to person compliance steps and no application to person opt out required, the closing line offers all the power of the consumer messaging service and none of the red tape, and the mobile application listing states no registration, no carrier filters, no compliance headaches.

Advertising that an opt out requirement does not apply is materially different from failing to mention one, because opt out is the load bearing protection in commercial messaging and it attaches to the sender under telemarketing law regardless of which transport carries the message. Across every page read, no statute is named, no consent standard is stated, and no suppression list, calling window, do not call handling or complaint route is described.

The genuine counterweight, and it belongs on the record, is that the pricing page publishes an explicit refusal to serve buyers who want to mass spam or cold blast thousands of contacts. That refusal sits on the pricing page while the escape from the opt out obligation sits on the homepage.

A privacy policy and a terms of service exist at their own routes in the footer and neither was opened, so this row rests on what the remainder of the site establishes and is flagged for re verification. What the surface does show is worth recording: the mobile application's store listing declares that the developer collects no data from the application at all, which for a platform whose stated function is synchronising message content, call transcripts and contact records into a third party system of record is a declaration to examine rather than to credit. A processing addendum appears nowhere in the footer or navigation, at a vendor handling the message content and call recordings of every conversation its customers have with their prospects.

The vendor sells no contact data and supplies none, which is the clean baseline for this axis. Records arrive from the buyer's own system of record and the outreach targets are the buyer's own contacts, so the provenance question that dominates this category largely does not arise.

One element sits outside that and is unsourced: the marketing offers industry specific health scores for every message sent, which implies a benchmark corpus assembled from somewhere, and no source, sample, industry definition or method is stated for it. Graded at the middle rather than higher because the clean position is a consequence of the business model rather than a disclosure the vendor chose to make.

This is the axis the product is built on, and every element of the finding comes from the vendor's own marketing rather than from inference. Commercial outreach at volume is routed through a consumer messaging service, and the bypass of the carrier registration regime is sold as the reason to buy. Three further tells sit on the same pages.

The vendor offers device line recovery support, explaining that if a number is ever flagged or restricted it will help resolve the problem quickly and restore the messaging flow, which concedes that restriction is a foreseeable outcome of ordinary use. Setup is performed remotely on the vendor's own infrastructure so that a customer needs no handset, which locates the sending identity somewhere other than the buyer's device.

And the frequently asked questions concede that mass outbound behaviour on this channel is against the vendor's own guidelines, an admission that the channel was not built for the volume the platform exists to push through it. The account that gets restricted belongs to the buyer.

The platform holds message content, call recordings, full transcripts and, for any customer who takes up the offer, a synthetic model of a named individual's voice. Whether any of that is retained, for how long, whether it trains anything, and whether it is isolated between customers are questions the site does not raise anywhere.

The voice model deserves separate mention: a cloned voice is a biometric artefact of an identified person, several jurisdictions regulate biometric identifiers under dedicated statute, and the offer to create one appears inside a frequently asked question with no accompanying word on consent, storage, deletion, or who may authorise the cloning of whose voice. Graded at the middle rather than lower because the gap is silence rather than misdirection, and because the terms and privacy documents were not read.

The vendor describes its own feature as producing a false impression of human authorship, in its own words. The frequently asked questions explain that personalised voice memos can be sent to hundreds of contacts without anyone actually recording them, then add that the sender's voice can be cloned, making it seem like it came from you. A recipient hearing what sounds like a specific person speaking to them has no way to know that nobody spoke.

The surrounding design points the same way: the mobile listing states that every message arrives like it came from a real person's phone because it did, branded contact cards let the sender shape how they appear in the recipient's inbox, and the channel itself is chosen because it reads as ordinary personal messaging rather than commercial traffic.

Against all of that there is no artificial content marking, no disclosure at first contact, no position on the European marking obligation for synthetic audio, and no mechanism by which a recipient learns any of it. This is the band reserved for marketing that celebrates the recipient being unable to tell.

Three named systems of record each carry a dedicated integration page rather than a logo, a documented programmatic interface has a page of its own, developer documentation sits on its own subdomain, and a public status page sits on another. A native mobile application is published in the official store for one platform.

One of the three integrations reports an installed base of more than four thousand three hundred accounts inside that vendor's own marketplace, which is independent placement rather than a self declared count. Number porting to the vendor's carrier partner is offered, which matters in a category where the number is the asset.

Holding it off the top band, webhooks are not documented on any page read, there is no marketplace of the vendor's own, federated sign on appears nowhere, and a fourth mark sitting on the integration row links to a dead anchor.

Hosting provider, country, region, data centre and residency option are absent from every page read, and no corporate address or jurisdiction is published beyond the entity name in the copyright line. Two indirect signals establish the market rather than the location: the publishing entity is a United States limited liability company, and the regulatory regime the marketing is built around is the United States carrier registration framework, with a United States toll free number as the only published telephone contact. Where message content, call recordings and transcripts physically sit is not addressed, and there is no security page or trust centre where the answer would normally live.

A public status page and developer documentation each sit on their own subdomain, which is more operational transparency than several vendors in this category publish, though a status page reports availability rather than security. Past those two, no certification, audit report, attestation, penetration test, control set, encryption statement, trust centre or vulnerability disclosure route was located on any page read.

Set against what the platform holds, which is the full message and call content of its customers' commercial conversations plus a cloned voice model for any customer who requests one, an unevidenced position is a weak one. The terms of service and privacy policy were not opened and either could carry a control statement.

Both purchase paths carry real numbers. Monthly is three hundred dollars for one line and the platform, with two hundred dollars a month for each additional line. Annual is one hundred and sixty six dollars a month billed as one thousand nine hundred and ninety eight paid in full, with one thousand eight hundred and ninety eight a year for each additional line. Each path states no contract.

The setup fee is disclosed rather than buried, at five hundred dollars on the monthly path and waived on the annual one, both shown against a struck through one thousand. That is a complete enough picture for a buyer to compute a multiple line bill in advance. Two things hold it off the top band.

A demonstration call gates the entire purchase, with the vendor stating that it first confirms the buyer's stack fits and matches them to a specialist, so there is no route from the published price to a purchase without a sales conversation. And for a product advertised as truly unlimited texting, fair use is left undefined while excessive use may be reviewed, so the one allowance that actually governs the service carries no number.

The structural position is reasonable and it is the reason this row is not lower. Logs, transcripts, notes and outcomes synchronise automatically into the buyer's own system of record, so the accumulating record of every conversation lands somewhere the buyer already governs rather than only inside the vendor.

Against that, the only porting language published runs the wrong way: customers are told they may port a line in to the vendor's carrier partner, and nothing anywhere describes taking a number out. Message history held on the platform, call recordings, the analytics record and any cloned voice model have no stated export function, format, post termination right, deletion timeline or retention period. Both legal documents were unread and are flagged.

Real discipline is published and deserves crediting. Lines are dedicated rather than shared so sending reputation is never pooled with another customer, delivery falls back automatically to the alternate channel when the primary is unavailable, uptime is stated at ninety nine point nine percent with a public status page behind the claim, and the pricing page refuses buyers whose stated intent is high volume low quality outreach.

Against that, the entire proposition is escaping the registration regime that governs volume on the alternate channel, truly unlimited texting is advertised with fair use left undefined, and the vendor offers help when a customer's line is flagged or restricted, which concedes that restriction happens. Warmup, ramp schedule, daily ceiling, complaint threshold, bounce handling and volume governance appear nowhere.

A published statement of who the product is not for is rare enough in this index to carry the grade on its own, and this one is specific rather than decorative. Six lines name buyers the vendor does not want: those looking to mass spam or cold blast thousands of contacts, teams focused on high volume low quality outreach, companies pushing low ticket offers where only volume matters, spammy offers and questionable products, anyone wanting a cheaper generic messaging alternative, and anyone trying to replace a full dialer system.

A matching statement of who it is for sits beside it. Fourteen named customers span eight unrelated sectors and a stated three thousand customers gives a scale figure. Three gaps hold it off the top band. That customer figure contradicts itself between two pages, seat bands, regions, languages and industry coverage are stated nowhere, and the named customer roster is overwhelmingly businesses selling to consumers while the positioning addresses sales teams generally, so the segment served and the segment marketed to are not the same.