ChurnZero

A decade old platform that renamed itself around artificial intelligence in the last two years. What the product does at its core is computation over data the customer already owns: usage telemetry, support tickets, contract terms and engagement events are combined into a health score, thresholds on that score trigger a play, a journey advances a sequence of touches on a timetable, and a forecast rolls up renewal values.

Health scoring in this category is a weighted model the customer configures rather than an inferred one, which is why buyers ask to tune it. Remove every model and the platform that reviewers describe adopting remains complete. Above it sit a customer success assistant and an agent product launched in June 2026 that the vendor describes as delivering execution rather than only recommendation, which is a genuine model dependency and is ten weeks old at the time of this record.

The vendor now describes itself as the artificial intelligence platform and partner for customer growth, where two years ago it described itself as the platform and partner for customer success. That is a positioning change on a product that predates it by a decade. Ask which capabilities stop working without the assistant and the agent layer, and how health scores are computed.

An agent product that executes, launched ten weeks before this record, with no published account of what constrains it. The vendor states the new agent capability delivers execution rather than only intelligence, which means software acting on accounts rather than suggesting actions to a person, and the platform beneath it already runs automated plays and multi step journeys that send messages and create tasks without a person approving each one.

Nothing published describes what the agent may do unattended, whether an action can be held for approval, what volume ceiling applies, what escalates to a person, or what halts a journey already running against a live account. The stakes are specific to this category: an automated intervention against a customer who is already unhappy can accelerate the churn it was meant to prevent, and the account on the other end is a paying customer rather than a prospect.

Two real controls exist at the platform level and are worth crediting: session expiry is configurable by the customer, and access follows a least privilege model with reviews every ninety days, though both govern staff access rather than agent behaviour. Ask what the agent is permitted to do without approval, what stops an action mid flight, and whether agent actions are logged and reversible.

Two named capabilities, one superlative claim, and nothing underneath either. The vendor names a customer success assistant and, from June 2026, an agent product it describes as the only customer success artificial intelligence delivering execution, intelligence and reach in one system. A claim to be the only product doing something is unusually strong and is published without the comparison that would support it.

No model provider, family or version is named for either capability, no model card exists, and no accuracy, quality or evaluation figure is published for the assistant, the agent or the scoring that determines which accounts they act on. The scoring gap matters most.

Health scores decide which customers receive intervention and which are left alone, so an unexplained or drifting score silently reallocates attention across a customer base, and nothing published describes how the score is computed, how often it is recalibrated, or how a customer would detect drift. A free standalone assistant is offered separately, which gives a prospective buyer a way to see output without a contract. Ask which models power the assistant and the agent, what evidence supports the only in market claim, and how health scores are computed and validated.

A long running research programme about the profession, and nothing measured about the product. The vendor runs an annual customer success leadership study now in its seventh year, produced with named partners and examining how the discipline is changing, alongside an annual practitioner conference in its fourth year. Sustaining a research programme for seven years is a real commitment and gives the company standing that marketing alone does not buy.

It also measures the profession rather than the software: nothing in it establishes what happens to churn, retention or expansion at companies using this platform. On the product itself the published evidence is thin. No outcome study with a stated customer population, period and method was located, and no retention or churn reduction figure attributable to the platform appears.

Independent review aggregation is substantial, with one analyst site reporting a satisfaction rating in the mid nineties across more than two thousand reviews drawn from four separate review platforms, and reviewers converge on support quality, health scoring flexibility and automation depth as strengths. That is a reliable signal of how users find the product and not a measure of what it achieves. Ask for churn or retention improvement measured across a stated customer cohort and period, and how it was isolated from other factors.

The people receiving these messages are already paying customers, which changes the posture substantially, and the mechanics are still unpublished. Lifecycle email campaigns, in application messages and survey invitations all go to users of the customer's own product under an existing contractual relationship, so consent to be contacted is a different question here than for a cold prospect and the volume patterns that make this axis bite do not arise.

That is why this sits mid band rather than lower. What is absent is everything below that. Nothing published describes unsubscribe handling for lifecycle email, whether a recipient can decline in application messaging without losing product function, how a suppression preference propagates between the two channels, or what obligations the vendor considers to transfer to the customer whose brand the messages carry.

The survey programme raises the same question in a sharper form, since research contact is regulated differently from service communication in several jurisdictions. Ask what unsubscribe mechanism applies to lifecycle email and in application messaging, and whether an end user can opt out of one without affecting the other.

A serious compliance record carrying one claim that should have been retired years ago. The substantive material is real: an independent health information attestation, an information security management certification, an audited control report covering security, availability and confidentiality, and a published guide in which the vendor tells buyers to insist on a current processing agreement and to check whether a vendor publishes subprocessor lists at a standing address.

The last of those is a company setting out the standard it wishes to be measured against, which is a good sign and also invites the measurement. Against that, a vendor press release listing its accreditations includes compliance with the transatlantic transfer framework that a European court invalidated in 2020, published years after that decision.

This index treats a live claim to that framework as a warning sign, and the charitable reading is that the vendor means the successor arrangement that replaced it, in which case the fix is to say so. Either way a buyer relying on it for a European transfer is relying on the wrong thing. Beyond that, no retention period, subprocessor list or residency election was established, and the platform is stated to be wholly hosted in one public cloud with no region named. Ask which transfer mechanism currently applies, for the subprocessor list and retention periods, and whether any regional hosting option exists.

Almost nothing on this axis applies and the grade records the scope rather than a shortfall. The vendor supplies no contact database, no firmographic data, no intent feed and no enrichment. Everything the platform holds arrives from the customer's own systems: product usage telemetry from their application, support histories from their help desk, contract and account data from their record platform, and survey responses their own customers gave them.

There is no purchased corpus whose origin a buyer needs to trace, which removes the exposure this axis exists to capture and is a genuinely favourable position. One question is open rather than absent, and it follows from the newer capabilities.

The assistant and the agent operate over that material, and nothing published states whether they draw on any reference data, benchmark set or cross customer aggregate in order to score or recommend, which would introduce material from outside the customer's own tenancy. The vendor's annual research programme also collects data across the profession, and nothing states whether customer platform data contributes to it. Ask whether the assistant or agent uses any data originating outside the customer's own tenancy, and whether platform data feeds the published research.

Official connections throughout, and one dependency that runs in the opposite direction from most records here. The integration set covers more than twenty named systems across customer record platforms, help desk and support tools, product analytics and team messaging, each through that product's own documented route, and the connection to the largest record platform is singled out by reviewers as the strongest part of the product.

Nothing scrapes a professional network, drives a browser extension against one or automates a social account, so the exposures that dominate this axis elsewhere do not arise. The distinctive dependency is inward rather than outward: in application messaging means this vendor's code executes inside the customer's own product, in front of the customer's own users, so a defect or an outage in the messaging layer surfaces as a problem in the customer's software rather than in a back office tool.

Nothing published describes what that integration requires, how it is versioned, what happens if it fails, or what the customer's obligations are for the code they embed. Held below the top band because no conformance position is stated for any connected platform and the embedded layer is undocumented in these terms. Ask what the in application messaging component requires and how failures in it are contained.

The vendor has published the exact list of questions a buyer should ask about artificial intelligence features and has not answered them about its own. Its guide for customer success leaders enumerates the sensitive material these platforms hold, naming health scores, contract details, private feedback, support histories and usage patterns, and advises careful review of the tools and artificial intelligence features a team brings in. That is an accurate and useful list.

It is also a precise description of what this platform holds and what its own assistant and agent process. Nothing published states whether customer data trains or tunes any model, which provider processes material when the assistant or the agent runs, what retention applies to prompts and generated output, or whether anything crosses a tenant boundary, and no governance document, evaluation record, red teaming artifact or audited management standard for artificial intelligence was located.

The gap is more consequential since June 2026, because an agent that executes rather than recommends acts on that material rather than summarising it. The compliance centre is live and rendering, so a buyer can ask through a route that exists. Ask whether customer data trains any model, which providers process agent workloads, what retention applies, and whether any artificial intelligence specific governance is documented.

The people on the receiving end are the customer's own users, which is the lightest position this axis records, and the newest capability reopens the question. Messages arrive inside a product the recipient already uses or by email from a company they already pay, so they know who is contacting them and why, and nothing here presents a persona, a synthetic voice or an agent posing as a person. Survey invitations and lifecycle messages are ordinary vendor communication.

What is unaddressed is authorship. The assistant drafts content and the agent product executes rather than only suggesting, so a message reaching a customer may be composed and sent without a person writing or reading it, and nothing published states whether generated content is reviewed before it goes, whether it is marked in any way, or what the customer's own users would be told if they asked.

The setting sharpens it slightly: a message that arrives inside the product itself carries more implied authority than an email, because it appears to be the software speaking rather than a person. Ask whether generated messages are reviewed before sending, and whether an end user can tell that a message was machine composed.

A named set spanning the four system categories this product actually needs, plus an embedded layer most competitors do not offer. Published connections cover two customer record platforms, two help desk and support systems, a product analytics platform, an issue tracker and two team messaging tools, with the total stated at more than twenty and the record platform integration repeatedly singled out by reviewers as the strongest element.

The categories matter more than the count for this product, because a customer success platform is only as good as the usage, support and contract data reaching it, and all three sources are covered natively rather than through a generic automation bridge. The in application messaging component adds a delivery surface inside the customer's own product, which is depth of a different kind and is not something a connector list conveys. Two things hold it below the top band.

No public developer documentation or programmatic interface was located on this pass, so a customer wanting to move data the vendor has not connected has no visible route, and no partner marketplace or directory was found. And the enumerated list, while covering the right categories, is modest against the larger platforms in this category. Ask whether a documented interface exists and for the full integration list.

The access architecture is described in unusual detail and the geography is not described at all. What the vendor publishes about how the platform is reached is specific and checkable: hosting wholly within one named public cloud, a least privilege access model with access reviewed every ninety days, internet access to the platform disabled at the edge with the only route in through an encrypted tunnel, an explicit statement that no bastion or jump host is used, customer configurable session expiry, and continuous vulnerability and anomaly monitoring through three named third party tools.

Naming the monitoring vendors and stating what is deliberately absent are both better practice than the usual list of adjectives. What is missing is everything a buyer with a location obligation needs. No region is stated, no residency election is offered or ruled out, no tenancy model is described, and no recovery time or recovery point objective appears anywhere.

For a platform holding a customer's entire customer base including contract values, private feedback and support history, and marketed to organisations serving regulated sectors on the strength of a health information attestation, the absence of a stated region is the first question a review will raise. Ask which regions host the platform, whether any regional option exists, and what the recovery objectives are.

A live compliance centre with a dated artifact a buyer can actually download, which is rare enough to carry the grade. The portal renders publicly and carries an audited control report for 2025 covering security, availability and confidentiality, an information security management certification, and an independent health information attestation.

The element that separates this from the usual pattern is the penetration test: the vendor states an annual third party test was conducted in March 2026 and publishes a customer letter summarising the scope, methodology and findings for download, so a reviewer can read something rather than request something. Alongside it sits a separately published security paper describing the access architecture in specific terms, and security leadership is named publicly with stated credentials.

The vendor also publishes a buyer's guide setting out what to demand from platforms in this category, which is a company inviting the standard it is then measured against. Two things hold it below the top band. The audited report itself still requires a request rather than being downloadable, and no auditor is named for it, so the strongest single artifact remains gated while the secondary one is open. And no audit period beyond the year is stated. Ask for the report with its audit period and auditor, and whether the certification scope covers the newer agent capabilities.

No figure on any vendor surface, a metering unit that makes budgeting harder than a seat count would, and a discount range that suggests the list price means little. Nothing is published: no tier, no rate, no calculator, no self serve path, with evaluation running through guided demonstrations and a proof of concept.

The primary pricing dimension is the number of end customer accounts managed rather than users, which is defensible for this product and does mean a buyer's bill scales with their own growth in a way seats do not, and no published guidance helps them model it. Layered on top are seat counts for the success team and module selection, so three variables interact and none is published.

Third party figures do not converge into anything usable: an entry point near eight hundred and fifty dollars a month appears in one account and a starting figure around twelve thousand dollars a year in another, real deals are reported between fifteen and eighty thousand a year with mid market clustering between twenty five and fifty, and procurement data puts buyers managing two thousand accounts or more above one hundred thousand.

Implementation is reported separately at five to twenty thousand. One reported detail is worth more than the ranges: buyers are said to secure discounts between forty five and seventy one percent off list, which if accurate means the list price is not a price at all. Sources also disagree on whether a trial exists. Ask for the rate per account band, what modules cost, the implementation fee, and whether a trial is available.

Nothing published on leaving, and the accumulated asset is larger than it looks. No export scope or format, no post termination access window, no retention or deletion timeline and no notice term was located on any surface reached. What a departing customer would be leaving behind is worth stating plainly, because it is not simply a record set.

A mature deployment holds years of health score history, the configured scoring model itself, play and journey definitions, success plan templates, survey response history from the customer's own customers, and the engagement record that gives all of it meaning. The scoring configuration is the part that took longest to get right and is the least likely to be portable, since it encodes how a particular business defines a healthy customer.

A second consideration is unusual to this product: the in application messaging component runs inside the customer's own software, so leaving involves removing code from a shipped product rather than switching off a login, and nothing published describes that process or its timing.

Ask what exports and in what format, specifically whether scoring configuration and play definitions are included, what happens to survey and health history at termination, and what removing the embedded component requires.

The platform sends lifecycle email at volume on the customer's behalf and publishes nothing about the mechanics. Automated plays and journeys drive email campaigns to the customer's own users across onboarding, adoption, renewal and win back sequences, which at a customer base of several hundred to several thousand accounts is real recurring volume rather than occasional correspondence.

Nothing published describes which infrastructure sends, whether mail leaves from vendor domains or the customer's own, who configures the sender authentication records, what bounce or complaint handling applies, or whether any threshold pauses a journey. The exposure is genuinely lower than for outbound products because every recipient is an existing paying customer, so complaint rates should be structurally low and the relationship provides its own consent basis.

It is not zero: renewal and win back sequences reach customers who are already disengaged or unhappy, which is precisely the population most likely to mark mail as unwanted, and the sending domain is usually the customer's own. In application messaging bypasses mail entirely for part of the volume. Ask which infrastructure sends, who owns the authentication records, and what bounce or complaint rate pauses a journey.

The fit boundary is unusually well defined, and mostly by other people. Independent pricing analysis and reviewers converge closely on the same answer: the product suits organisations managing roughly five hundred to two thousand end customer accounts with a multi person success team, and is described as materially overpriced below that range with lighter alternatives named for smaller teams.

That is a clear and consistent boundary, and the metering unit reinforces it, since charging by customer accounts means a buyer can locate themselves on the ladder by counting something they already know. Sector coverage extends beyond software through an independent health information attestation, which opens regulated healthcare buyers, and the information security certification supports enterprise procurement.

Category standing is real rather than asserted, resting on a research programme in its seventh year and a practitioner conference in its fourth. Two things hold it below the top band. The vendor states none of the fit boundary itself, so a buyer learns where they sit from third parties rather than from the company.

And no geographic coverage, language or residency statement was located, which for a platform embedded in customers' own products serving their end users internationally is a gap. Ask what account volume the entry configuration assumes, and which regions and languages are supported.