Ring.io

Three features carry the label and all three sit on the Growth tier: an AI parallel dialer, call summaries with transcription, and a statistics assistant. Tiering them is honest, because it tells a buyer the model is an upgrade rather than the thing they are buying at entry.

That is also the finding. The product underneath is telephony and it works without any of them. Click to call, the power dialer, voicemail drop, call recording, automatic disposition and write back to the record are the machinery a customer runs on, and that machinery was built long before the model layer arrived. The company was founded in 2009 as a calling service wired into a customer relationship system, and the intelligence has been added on top of a mature dialer rather than the dialer being rebuilt around it.

Each of the three is plausibly conventional under the label. Transcription and summarisation are post call processing, the statistics assistant is a natural language surface over reports that already existed, and the intelligence in the parallel dialer is most likely answer detection and pacing, both of which predate the current wave by decades. None of that is disqualifying, but none of it makes the model central either.

Ask which of the three uses a language model rather than classical signal processing, and what the product does when that layer is unavailable.

The vendor argued for constrained automation in public and then shipped the opposite, and the gap is the finding.

The published position is unusually principled. A blog post explains why a power dialer was chosen over a predictive dialer, on the reasoning that live human interaction retains value that pacing algorithms erode. The compliance article states the system dials one person at a time and that the agent is actively involved before initiating every call. A vendor that limits its own automation and publishes the reasoning has made a commitment rather than described a feature.

The current price list sells an AI parallel dialer on the Growth tier, and the home page states that on Growth it dials several numbers at once so nobody sits through dead rings. Multi line dialing is the mode the vendor's own guidance rules out. Neither surface acknowledges the other, and the compliance article has not been revised. A buyer reading both cannot determine what the product actually does or what the vendor now believes about it.

Supervisory oversight, as distinct from automation limits, is genuinely well provisioned and available from the entry tier: live monitoring, call whisper and call recording, with a paired assist capability at enterprise. Those are real controls over what representatives do, published per tier.

Ask how many lines the parallel dialer opens per agent, what the abandonment rate is in practice, and whether the vendor's published dialing standard still represents its position.

Three labelled features and no description of any of them. Across the product pages, pricing matrix, security page and privacy policy, all retrieved in full, no model, provider, architecture or version is named for the transcription engine, the summarisation layer, the statistics assistant or the intelligence in the parallel dialer.

For transcription the missing facts are exactly the procurement questions. No word error rate, no supported language list, no accent or audio quality caveats, and no statement of whether call audio leaves the vendor's infrastructure to reach a third party model. The vendor names Amazon Web Services and Twilio as its infrastructure and Google for email services, and never says which of them, or which fourth party, processes recorded speech.

The unanswered question with the most consequence is training. Call recordings are among the most sensitive material a revenue organisation holds, containing customer commercial terms, competitive intelligence and personal information disclosed in conversation. Whether those recordings or their transcripts train any model is addressed nowhere. The privacy policy commits that call records are never shared with partners, which is adjacent but not the same commitment, since training is a use rather than a disclosure.

Ask which vendor processes call audio, whether recordings or transcripts are used to train or improve any model, and for the transcription accuracy figure on your own call conditions.

The vendor publishes a live statistics page carrying its own platform telemetry, and it is the strongest artefact of its kind found in this lane. Twenty four individual dated daily observations of the human connection rate were retrieved, running between 9.7 and 13.9 percent across mid July to mid August 2026. Outbound calls per active agent day are published as a distribution rather than a headline: 22 at the median, 70 at the top quartile, 124 at the top decile and 170 at the top 5 percent. A high output agent day is broken out at 134 calls, 66.3 talk minutes and 12.2 conversations lasting over sixty seconds.

Two things lift this above a marketing statistics page. The industry baseline is attributed and linked to three named external sources rather than asserted, so the comparison can be checked. And the numbers are not uniformly flattering. The vendor publishes its own median connection rate at 12.9 percent alongside an industry top quartile of 13.3 percent, which concedes on the vendor's own page that a median team here does not beat a top quartile team elsewhere. A median of 22 outbound calls per agent day is a modest figure for a power dialer, and publishing the median at all when the top decile is six times larger is a choice most competitors do not make.

The instrument has hygiene problems, recorded here rather than discounted. The daily chart is labelled as 14 observations covering 1 to 17 July while listing 24 observations running 16 July to 14 August. The page states it was last updated 15 August while the home page reports the same data as updated 23 August. The headline multiple appears as 2.3 times on one surface and 2.1 on the other, and the median as 12.9 against 13. A page whose entire premise is measurement should reconcile with itself.

The grade stands because the substance is exceptional and the defects are labelling. Ask how a human connection is defined and whether the denominator counts every dial or only completed attempts.

The vendor maintains a dedicated dialer compliance collection in its help centre, which is more than most of this lane offers. The finding is not absence, it is contradiction and staleness.

The lead article, written by the chief executive and dated 22 December 2021, sets out an acceptable dialing standard: one person at a time, abandonment held below three percent, agent involvement before every call. It then names what is unacceptable, listing triple line dialing and predictive dialing explicitly, and concludes that the product is not an auto dialer. The current price list sells an AI parallel dialer, and the home page says it dials several numbers at once. The vendor's own published guidance rules out the mode its own price list sells, and nothing reconciles them.

The cited authority is superseded. The article rests entirely on a declaratory ruling from July 2015. The definition of an automatic telephone dialing system was narrowed by the Supreme Court in April 2021, eight months before the article was written, and is not mentioned. That narrowing probably helps the vendor's conclusion, which makes the omission careless rather than self serving, but a buyer is being pointed at reasoning that no longer holds. The article also carries an explicit disclaimer that it was prepared by a non lawyer.

What the collection never reaches is the exposure that survived the narrowing, which is where enforcement actually moved: national and state do not call scrubbing, internal suppression lists, calling time restrictions, consent capture and timestamped scrub logs. None appears on any retrieved surface across two passes. Compounding this, a testimonial the vendor chose to publish on its own security page praises leaving commercial messages after hours to reach hundreds of potential clients, a practice that runs against calling time restrictions.

Ask how do not call scrubbing is performed and logged, and for the vendor's current written position on parallel dialing.

The privacy policy carries three commitments that are specific enough to be held to, which is what keeps this mid band. Service providers are named individually rather than described as a category: Twilio for toll free services, receiving caller identity and dialed number mapping; Amazon Web Services for hosting, with access to network information; Google for email; and a named payment gateway for card address verification. A published subprocessor list at this level of specificity is rare for a company of this size. The policy states that call records and usage statistics are never shared with partners even under a non disclosure agreement. And the Google user data section carries a proper limited use commitment with verbs attached, undertaking not to transfer message data, not to use it for advertising, and not to permit human reading absent affirmative agreement.

What is missing is scoped to the product. This platform records calls between customers and third parties, and the policy addresses website visitors and account holders. Recording consent, which is a two party requirement in a substantial number of states, is not covered anywhere. Retention of recordings is unstated. The rights of a recorded person, who never agreed to anything, are not described.

The document is also dated in a way that undermines its own security section. Last updated May 2023, it describes sensitive data as protected by the best encryption software in the industry, names a protocol superseded years earlier, and directs the reader to a lock icon at the bottom of Internet Explorer, a browser retired in June 2022. The security page separately claims state of the art encryption. Both cannot be current.

No European lawful basis, data subject rights, transfer mechanism or California section appears. Ask for the recording consent guidance and the retention period for recordings and transcripts.

This vendor does not supply contact data, so the axis is not applicable in the provider sense and is rated accordingly rather than penalised. Lists originate in the customer's own customer relationship system or are uploaded by the customer, and the platform dials, texts and emails what it is given. Provenance liability for the underlying records sits with the buyer and whoever they licensed them from.

What the record can assess is the boundary the vendor creates, and it is drawn cleanly. The Growth tier adds integrations with two named commercial data providers, which is the sanctioned route: the customer holds their own licence with that provider and the platform consumes it under those terms rather than the vendor reselling anything. No data marketplace, no bundled contact credits, no enrichment product and no scraped source appears anywhere on the retrieved surfaces across two passes, and the pricing page prices telecom minutes and messages rather than records.

One consequence follows for the buyer and it is not the vendor's to solve. Because the platform imposes no provenance check on an uploaded list, a file of unknown origin dials exactly as readily as a licensed one, and the vendor publishes no guidance on what a customer should verify before loading a list. For a product whose adjacent axis exposure runs through calling regulation, list hygiene is where the two meet.

Ask what the platform records about the origin of an uploaded list, and whether any provenance or consent field travels with an imported record.

Exposure here is low and the reason is structural rather than promised. This vendor operates inside platforms it has been admitted to, and the admissions are checkable. Listings were located on the Salesforce application exchange, the Zoho marketplace and the Pipedrive marketplace, alongside a Chrome web store extension and published mobile applications on both major app stores. Each of those is a separate third party review against that platform's terms, and passing six of them is an act rather than an assertion.

The browser extension is the surface where this category usually goes wrong, and here it does not. It is a web phone that places and receives calls and offers click to call from a page, which reads and dials a number the user is already looking at. It does not harvest, export or bulk collect, and there is no scraping utility, no social network export tool and no list building product anywhere in the range.

The integrations with two commercial data providers on the Growth tier run the same low risk way, consuming data the customer already licences rather than extracting it.

The residual exposure is regulatory rather than contractual and sits with the carrier layer. Dialing and messaging obligations flow from the named carrier's acceptable use terms into the customer's own conduct, and the vendor does not describe where its responsibility ends and the customer's begins. Ask which carrier terms bind the end customer and who bears a suspension if traffic is flagged.

The published security programme is described in process terms rather than adjectives, and for a company of roughly a dozen people that is more than expected. Encryption at rest and in transit, a security incident and event monitoring system, containerised infrastructure under stated administration policies, a business continuity plan required to be tested at least annually, and a secure development lifecycle standard applied from planning through release. A public status page is maintained, which is an operational commitment a vendor cannot quietly abandon. The named subprocessor list and the undertaking never to share call records with partners both belong here too.

None of it is externally validated, and the stewardship question that matters most is unaddressed. This platform accumulates a corpus of recorded customer conversations and, on the Growth tier, machine generated transcripts and summaries of them. Whether that corpus trains any model, whether tenants are isolated from one another within it, how long recordings persist, and how a customer forces deletion are all absent from every retrieved surface across two passes.

One passage on the security page deserves flagging on its own terms. The vendor writes that its privacy policies and standards are derived from established software companies and that anyone who has reviewed those companies' standards will find its own familiar. Read plainly, that describes inheritance of documents rather than a programme designed for a product that records phone calls, and the dated security language in the privacy policy is consistent with it.

Ask whether recordings or transcripts train any model, what the retention period is, and for a written deletion commitment.

The vendor sells caller identity manipulation as a headline connect rate lever and takes no published position on where it stops being legitimate.

RingLocal dials from a number local to the person being called, with a premium version at enterprise. The reasoning is stated openly on the home page: people answer local numbers, so present one. There is a defensible version of this, where a business with genuine regional presence surfaces its real regional number. There is also a version where a company with one office presents a rotating set of local numbers it has no other connection to, which is the practice that made carriers build spam labelling in the first place. The vendor markets the capability without distinguishing them, and offers a caller identity product branded as bulletproof plus spam monitoring and remediation, which together describe an arms race with the labelling systems rather than a disclosure posture.

Voicemail drop with automatic answer detection delivers a pre recorded message without the representative speaking, at scale. The security page carries a testimonial praising exactly this, used after hours to reach hundreds of potential clients. Whether a recipient is told they received an automated drop rather than a missed live call is not addressed.

Generated content raises the same question from the other side, since transcription and summarisation are sold but no position is taken on notifying the other party that a call is recorded or processed by a model.

Ask what constrains which local numbers can be presented, whether the numbers are registered to the customer, and what the vendor's position is on recipient notification for recorded and machine processed calls.

Depth is described as behaviour and priced honestly. Five customer relationship systems are natively supported and, unusually, the integrations are included free at every tier rather than gated as a connector upgrade. The integration is bidirectional and works where the representative already is: a list view or report screen becomes a calling session, and calls, texts and emails write themselves back to the record without anyone typing up their day. Dispositions, custom fields and custom objects are configurable, and higher tiers allow platform actions to be triggered from native workflow automation, which puts the dialer inside the buyer's own orchestration rather than beside it.

Reach beyond the browser is real: published applications on both mobile app stores, a Chrome extension web phone, and listings on three separate platform marketplaces. Two named commercial data providers connect at the Growth tier.

The ceiling is the developer surface and where it sits commercially. An application interface exists, and it is the route to sending messages and email, triggering sequences and connecting an internal or custom system, but every one of those capabilities is confined to the enterprise tier with a twenty five user minimum. No public interface documentation and no self serve credentials were located across two passes. A buyer below that tier has five integrations and no way to build a sixth, which is a defensible commercial choice but a real constraint.

Ask for the interface documentation and rate limits before assuming a custom integration is feasible at your tier.

Infrastructure providers are named, which is a genuine disclosure and is credited: Amazon Web Services for production hosting and Twilio as the carrier layer, with a payment gateway and a mail provider named separately. Knowing who holds the data is the first question and the vendor answers it.

The second question goes unanswered entirely. No region, no country and no data centre location appears anywhere across two passes. The only statement on the subject is a warning in the privacy policy, set in capitals, that personal information may be processed in the country where it was collected and in other countries including the United States where the laws may be less stringent than the reader's own. That is a disclaimer of residency rather than a description of it, and it tells a buyer only that their data may go somewhere with weaker protection.

There is no European hosting option, no single tenant deployment, no customer managed encryption key, no data processing addendum, no standard contractual clauses and no transfer mechanism of any kind.

The stakes are set by the product rather than by general principle. This platform records telephone conversations, and recordings are among the most jurisdictionally sensitive data a company can hold, because the legality of making them varies by the location of both parties. A vendor that will not say where recordings are stored is asking a buyer in a regulated sector to assume the answer. Named industry pages target financial services and higher education, both of which will be asked this question by their own auditors.

Ask in which region recordings are stored and whether any non United States option exists.

The security page promises certifications twice and names none. Its own page description advertises key compliance certifications, and the body repeats that the vendor maintains multi layered safeguards and key compliance certifications. Not one is identified anywhere on that page or on any other surface retrieved across two dedicated passes. There is no certification to apply the credential test to, because none is named, let alone scoped or dated.

One fact makes the silence harder to read charitably. The enterprise tier includes a phone payments capability, and the feature page for it is titled for the payment card industry standard. Taking card numbers over a recorded telephone line places a vendor squarely inside that standard's scope, and it is the one credential a buyer would expect to be named without asking. Its absence alongside a product that requires it is the sharpest gap on this axis.

The programme description is not nothing and is credited on the stewardship axis rather than here, since this axis measures verifiable credentials. A public status page exists. Infrastructure providers are named.

The internal contradiction runs the same way as elsewhere on this record. The security page claims state of the art encryption standards while the privacy policy, updated May 2023, describes protection by a protocol long superseded and refers the reader to a lock icon in a browser retired in June 2022. Marketing and the legal document disagree about the vendor's own controls.

Ask which attestations exist, who issued them, over what period, and for the assessment covering the payments feature.

Close to a model disclosure for this lane. Three tiers are published at both monthly and annual rates rather than the usual single figure with the other behind a toggle nobody can screenshot: 120, 155 and 300 dollars per user per month billed monthly, against 90, 120 and 250 billed annually. User minimums are stated plainly at 2, 3 and 25. A free plan exists for a single representative and a twenty one day trial is offered without a call. A full feature matrix maps every capability to every tier.

Usage costs are published to the cent, which is what makes the subscription figure meaningful: calling at two cents a minute, video at one cent a minute, messages at two cents and four cents, email at one cent. An answer explains that telecom minutes are a pass through cost from carriers and states there is no cap on call volume.

The strongest element is the one most vendors would suppress. The vendor publishes that its plans carry maximum user limits which force an upgrade, then prints the hostile version of the question in the buyer's own voice, including the accusation that this is a way of charging for unwanted features, and answers it. Publishing an unfavourable commercial term and arguing for it in public is rarer and more useful than publishing a price.

The gaps are minor by comparison and worth naming. The maximum user numbers themselves are not stated, only that they exist. The unlimited North American calling plan is referenced without a price. Contract length, renewal terms and the treatment of the annual commitment on the monthly rate are not addressed.

Ask for the user ceiling on your intended tier and the unlimited calling plan rate before modelling.

No export route, retrieval format, termination assistance period or notice provision was located on any surface across two passes. The privacy policy offers correction, update, deletion and deactivation handled through the support community, which is a personal data mechanism for an account holder rather than a business data export for a departing customer.

What makes this consequential is the asset the product accumulates. Every call is recorded and, on the higher tier, transcribed and summarised. Over a few years a customer builds a corpus of recorded customer conversations that has evidentiary, coaching and commercial value, and in some sectors a retention obligation attached to it. The vendor publishes no retention period for that corpus, no export format, no bulk retrieval mechanism and no statement of what happens to recordings after an account closes. A buyer cannot tell whether their recordings would be returned, retained or destroyed.

The partial answer available is a paid one and it is fragile. The enterprise tier includes interface access that would in principle allow programmatic extraction. That is a capability sold at 250 dollars per user per month with a twenty five user minimum, not a portability commitment, and it is unavailable to every customer below that tier. A capability that disappears on downgrade is not an exit route.

Activity data is comparatively safer, since calls, messages and dispositions are written back into the buyer's own customer relationship system as they happen, so the interaction log survives departure even though the audio may not. That mitigation is real and is why this does not sit lower.

Ask for the retention period for recordings, the export format, and a written commitment on retrieval at termination.

In the voice channel the equivalent of deliverability is caller reputation, and this vendor treats it as a product problem rather than a carrier's problem. Spam monitoring and remediation is a named, tiered feature rather than a background assurance. Local presence dialing is provisioned at two levels. A caller identity product sits at enterprise. A published article argues that a cheap dialer costs more than it saves because flagged numbers destroy connect rates, which frames reputation as the economic centre of the category and commits the vendor to a position.

The decisive point is that the outcome is published rather than promised. A daily human connection rate maintained on a public page is the closest thing the voice channel has to a measured bounce rate, and a vendor whose numbers were being widely blocked could not sustain that page. Twenty four dated observations holding between roughly ten and fourteen percent is evidence of working reputation management in a way no feature list can be.

The mechanics are undisclosed. How remediation works, how quickly a flagged number is rotated out, what the success rate is, and how many numbers a tier permits are all unstated beyond the count of company numbers per plan. Caller authentication at the carrier level is presumably inherited from the named carrier, which handles it as a matter of course, but the vendor never says so and never states its attestation level.

The messaging side is the weaker half. Messages are sold by the unit at every tier and no registration guidance for application to person messaging appears anywhere, despite registration being the gate that determines whether commercial texts arrive at all.

Ask for the attestation level on outbound calls and who completes messaging registration.

Segment is encoded in the price list rather than described in prose, which makes it unusually legible. A free plan serves a single representative, the entry tier begins at two users, the middle at three and the enterprise tier at twenty five. Those minimums state plainly that this is built for small and mid sized calling teams and that the largest tier begins where many competitors' mid market starts.

Industry coverage is specific and the named verticals are not the usual technology monoculture: dedicated pages address technology, financial services, marketing and demand generation, online retail, and non profit and higher education. Named reference logos match that spread, covering medical devices, local marketing services, a state university, recruiting technology, lending and fitness management software, which is a genuinely mixed base rather than a single vertical with decoration.

The scale claims reconcile, which is worth noting because they often do not. The vendor claims more than a thousand customers. Third party sources place headcount in the low teens and annual revenue near 7.7 million dollars. Those figures imply an average customer paying roughly eight thousand dollars a year, which at the published rates is a team of three to seven representatives. The picture is coherent: a small company serving many small teams, and the vendor's own minimums say so rather than posturing at enterprise.

Geography is the gap. An unlimited calling plan is referenced for the continental United States and Canada, implying a North American centre, but no supported country list, no international rate card and no language coverage for the transcription feature was located. Ask which countries are supported for numbers and calling, and which languages transcription handles.