Kixie

A thirteen year old telephony company with a recent artificial intelligence programme layered over it, and the price list settles the removal test cleanly. Strip the models and what remains is a complete and saleable product: a business phone service, a power dialer, text messaging with shared inboxes, voicemail drop, call recording, auto attendants, ring groups, queues, rule based routing and bi directional synchronisation with the system of record, all of which ship on every plan.

The model dependent pieces sit outside that core and the vendor's own packaging says so. Answering machine detection is marked as a premium add on against the plan that carries it, and conversation intelligence is sold as a separate premium add on rather than included anywhere. This is the standing rule applied exactly as written, that an established platform with a loud artificial intelligence programme lands here because the removal test leaves a complete product behind.

The architecture places a person on every conversation, which is the strongest thing that can be said for it. Multi line dialing places several calls at once and answering machine detection filters recordings and menu systems, so a representative is connected only when a live human answers. Nothing here composes or sends a message unattended and no autonomous agent contacts anyone on its own initiative. The gap is the governance of the mechanism itself.

Dialing multiple lines per representative is precisely the configuration that produces abandoned calls when more recipients answer than there are people to take them, and nothing published names an abandonment rate, a ceiling, a pacing algorithm, a monitoring measure or a remedy. Calling session management is listed as a feature on two plans and is nowhere described. An architecture that keeps a human present is not the same as an oversight model, and only the first is evidenced.

Four model dependent capabilities are marketed by name and none is described beyond its label: answering machine detection, local presence selection, conversation intelligence with sentiment analysis, and a recently launched reporting and analytics layer. No provider, no model family, no version, no inference location and no data handling position is published for any of them. The accuracy omission that matters most on a dialer is the detection layer.

If the classifier is wrong in one direction a representative is dropped into a voicemail greeting, and if it is wrong in the other a real person answers to silence while the system decides, which is the experience recipients report as a nuisance call and which regulators treat as an abandoned call. No accuracy rate, false positive rate, validation method or confidence measure is published for a component whose error profile is felt directly by people who never chose to be contacted.

Corroboration of adoption is real and independent, which is more than many vendors here manage. The product carries listings in three separate first party marketplaces, on the largest system of record vendor's exchange since September 2021 and on two others, each of which applies its own review and carries public reviewer feedback. The company has been operating since 2013 and runs a public service status page with incident history. What is missing is the vendor's own outcome evidence.

No named customer case study, no quantified result with a stated measurement basis, and no named reference were located on the surface read this pass. Recorded honestly rather than held against the vendor: the homepage body did not render on retrieval, so customer material may well be published there and this is a failure to read it rather than a finding that none exists. A third party review site reports a rating of 4.8 out of 5, which was not verified at source and is noted rather than relied on.

Genuinely mixed, and one packaging decision decides the grade. On the credit side the terms of use name three statutes explicitly, the Telephone Consumer Protection Act, the Telemarketing Sales Rule and the federal commercial email statute, and require the buyer to remain in full compliance with them and with all laws concerning privacy, telemarketing and internet marketing.

Published text messaging terms carry opt in consent, stop and help keyword handling, a message frequency statement and carrier disclaimers, which is the standard the carriers themselves require for campaign registration. Phone number reputation management is offered against spam labelling. Against all of that sits the finding: registry scrubbing is not part of the platform.

The plans page lists it among three add ons and labels it a compliance add on, so automated compliance with do not call registries is a separately purchased item rather than a control the product ships with. The comparable dialer graded in this index earned the top band in part by shipping registry management as standard, and it also named the federal abandoned call limit, cited the governing appellate authority on automatic dialing systems, and published its call authentication attestation and campaign registration support. None of those four appear here. Naming statutes in a warranty transfers the obligation to the buyer; selling the scrubbing separately monetises it.

The policy is dated 3 June 2026 and it describes a different business. Its text repeatedly addresses a consumer offers and personal finance product: it asks for a zip code and a forum username at registration, refers to fax numbers and identity theft, describes third party account credentials used to reach account data held at other institutions, sends registered customers weekly account summaries, governs bulletin boards, blogs and chat rooms, covers contests and giveaways, contemplates co brand partners reached through a co branded address, and repeatedly discusses serving and frequency capping something it calls an offer, including disclosure to advertisers and to organisations researching business spending.

Nowhere in seventeen sections does it address call recordings, dialled numbers, message content or customer relationship records, which together are the entire personal data surface of the product actually sold. There is no section on European or state privacy law, no data subject rights of any kind, no legal basis, no transfer mechanism, no retention period, no data protection contact and no sub processor list, on a platform that records conversations and markets itself to financial services and health services buyers by name.

Two things are real and are the reason this is not worse. The commitment not to sell or rent personal information is unambiguous and is repeated in the page metadata. And the two closing sections on text messaging are specific, current and clearly written for this product, which points to what happened: the recent revision added exactly the sections a carrier requires for campaign registration and left the inherited template around them untouched.

Structurally cleaner than most of this index because there is no data product. The vendor sells no contact database and acquires no third party person records, so the lists dialled and messaged are the buyer's own and provenance is the buyer's responsibility rather than a licence chain the vendor must explain. That is a real advantage and it is why this sits at the middle band rather than below it. Two things stop it going higher.

The privacy policy reserves the right to disclose anonymised and aggregated information to advertisers for their marketing purposes and to organisations conducting research into business spending, which is an onward data flow that is described in a single clause and never explained, and which sits oddly against a business that has no advertising product. And the inventory of real telephone numbers across the North American numbering plan that powers local presence selection carries no sourcing, ownership or rotation disclosure at all.

A comparatively low exposure profile and it comes from architecture rather than assurances. The vendor operates its own telephony infrastructure and carrier relationships rather than automating a platform it does not control, so there is no third party account whose suspension would take the buyer's operation with it.

Distribution runs through official first party marketplace listings on three major systems of record, each of which applies its own technical and security review before publication, and a documented developer interface is available on every plan rather than reserved for the top tier. There is no scraping, no social platform automation, no credential rotation and no multi account mechanism anywhere in the product.

The browser extension reads telephone numbers on pages the user is already viewing and adds a dialing control, which is a mild surface. Off the top band on two counts: no conformance position is stated with any carrier or marketplace, and progressive rotation of the displayed caller identity sits in tension with the direction carrier authentication rules have been moving, which the vendor never addresses.

Conversation intelligence processes recorded human speech and produces transcription and sentiment analysis, which is among the most sensitive inputs any vendor in this index handles, and the cross tenant question is never reached. Nothing published states whether recordings or transcripts are used to train, tune or evaluate models, whether processing is tenant isolated, or which providers touch the audio.

The retention position is the sharper problem and it comes from the marketing rather than a policy: lifetime call recording is presented as a selling point, which describes indefinite retention of recorded conversations between the buyer's staff and third parties who were not party to the subscription, with no stated period, deletion trigger or customer control.

The privacy policy's aggregate disclosure clause is the only adjacent statement and it addresses a different kind of data entirely. Held at the middle band because no cross customer data asset is marketed and no pooled inventory is claimed, so the concern here is unanswered rather than evidenced.

Three mechanisms in this product shape what the person on the other end believes, and none carries a disclosure position. Local presence selection places a caller identity matched to the recipient's own area code drawn from real numbers across the whole North American numbering plan, with progressive rotation of the number shown, so the geography the recipient reads off their screen is manufactured and changes between attempts.

Voicemail drop leaves a pre recorded message that is indistinguishable from one the representative had just spoken. Answering machine detection means a person who answers may hear a pause while the system classifies them before a human arrives. No position is taken on artificial intelligence disclosure and the European marking obligation is not addressed.

Held at the middle band rather than lower because the numbers used are real numbers the vendor controls rather than spoofed identifiers, because reputation management is framed as reducing spam labelling rather than as evading it, and because there is no synthetic voice agent here presenting itself as a person. The comparable dialer in this index holds the same middle band on the same tension, and it at least paired local presence with branded calling that shows the recipient the company's real name.

Distribution and interface depth are both real. The product is listed in three first party marketplaces belonging to the major systems of record, with a further two named connectors and a general automation platform, plus a published integration catalogue. Synchronisation is bi directional rather than one way logging, and calls, messages, dispositions and recordings write back automatically, which is the difference between an integration and a data island.

The detail worth naming is that the developer interface appears on every plan including the entry tier rather than being reserved for a custom contract, which is the inverse of the pattern this index sees repeatedly and materially changes what a small team can build. A public status page and a dedicated developer page complete the surface. Off the top band because no model context protocol server was located, no webhook surface is described, and the depth of each connector was not verified this pass.

No residency position is published on the current surface, and the vendor says so deliberately. The security page states in terms that it does not claim a hosting architecture that has not been confirmed for publication, which is an honest refusal rather than an oversight, but it leaves a buyer with nothing: no region, no data centre, no hosting provider, no processing location, no sub processor register and no transfer mechanism.

The only architectural detail located anywhere is a support centre article from 2019, seven years stale, which names a major cloud provider, a virtual private cloud and identity and access management roles, and which also asserts that the carriers and servers rather than the company itself meet European data protection standards, an attribution error this index has flagged on other vendors.

Unlimited minutes are scoped to the United States and Canada, which implies a North American footprint without stating one. The gap matters more than usual because the data at issue is recorded human speech.

The best drafted middle band on this axis in the index, and the reasoning is worth stating because it runs against the pattern here. This vendor publishes a security page whose explicit purpose is to refuse to overclaim. In its own words it routes evaluators to current information without making certification claims that are not published there, and it states that it intentionally does not claim a certification, audit result, hosting architecture, retention period or control that has not been confirmed for publication.

This index has repeatedly marked vendors down for asserting attestations without enumerating them, and this is the same principle applied by a vendor to itself. What it does publish is procedural and genuine: a named vulnerability reporting address with handling instructions telling reporters not to include credentials or customer records in a first submission, a public service status page with incident history, an explicit rule that the linked policy and contract control where a summary differs, and a review date of 3 August 2026.

What it does not publish is any attestation, audit, penetration test or named control, with evaluation materials available only by asking a sales representative. It cannot rise above the middle band on discipline alone, because a buyer still learns nothing about the posture.

One contradiction is recorded as observed: while the security page declines to claim controls or architecture, the privacy policy asserts secure facilities with biometric entry, security personnel around the clock, encrypted backup drives and tapes and physical access controls, which reads as inherited template text and is precisely the kind of unverifiable claim the security page was written to avoid.

The page publishes the hard half of pricing transparency and withholds the easy half. Three named plans are set out with a complete feature comparison table across roughly twenty rows, every inclusion and exclusion marked, the dialing line count stated per tier, and three add ons identified by name.

There is no figure anywhere: no rate, no unit, no currency, no range, no minimum and no price for any of the three add ons, while the page instructs the buyer to choose a plan based on calling volume without publishing a volume band. That is unusual in shape, because most vendors that withhold price also withhold the feature matrix, and this one has done the reverse. Held at the middle band rather than lower for two reasons.

A seven day trial requires no card and no sales contact, so a buyer reaches a real number themselves rather than through a form. And the possibility that values are behind a billing toggle that did not resolve on retrieval cannot be excluded, so this is recorded as observed with a re verification rather than asserted. Third party sources continue to cite per seat figures in the region of thirty five to ninety five dollars a month that no longer appear on the vendor's own page.

The architecture does more for a departing customer than the paperwork does. Bi directional synchronisation writes calls, messages, dispositions, outcomes and recordings into the buyer's own system of record continuously, so the activity history accumulates somewhere the buyer already controls and does not have to be extracted at the end. That is a genuine structural advantage and it is why this is not lower. What is absent is every contractual element.

No post termination retention window, no deletion timeline, no deletion confirmation, no described export path for recordings or message history, and no statement of what happens to telephone numbers the buyer has been using when the subscription ends, which for a phone system is the most consequential portability question of all.

A support centre article from 2019 offers complete deletion on request under the European erasure right, which is a commitment of a kind but is seven years old and sits on a help site rather than in the contract. Lifetime call recording is marketed as a benefit and implies the recordings themselves remain with the vendor.

Read for a voice and messaging product, the relevant discipline is reaching the recipient without being labelled or blocked, and there is real material here. Telephone number reputation management is offered against spam labelling, caller identity is rotated progressively to spread usage across numbers, and the published message terms carry stop and help keyword handling and a frequency statement, which is what carriers require before a business messaging campaign may run.

The reservation is the same one this index has applied to sending infrastructure elsewhere: rotating away from a number that has picked up a spam label treats the label as an obstacle to route around rather than a signal about the calling behaviour that earned it, and no complaint threshold, connect rate floor, pacing rule or remediation process is published alongside it. Campaign registration support for business messaging is not named anywhere on the surface read, which is a notable omission for a product whose messaging feature ships on every tier.

One of the more developed segment statements in the index and it works on two independent dimensions at once. By buyer it addresses five teams and four roles with separate pages for inside sales, business development, marketing, revenue operations, customer support, sales development, account executives, marketing teams and sales managers. By vertical it names five industries specifically, including two regulated ones it takes on directly.

And the plan structure itself is a segment statement rather than a feature ladder, since the three tiers separate an individual professional from a single line dialing team from a high volume multi line operation, which maps to calling intensity, the variable that actually determines who this product suits.

Off the top band because the framing claims the entire range from individual professionals to enterprise teams, and because no headcount band, seat count or upper limit is published anywhere. Vendors that reach the top band on this axis name a ceiling as well as a floor.