Endgame

The product is an answer surface and nothing survives removal of the model. A rep asks a question in natural language and receives an analysis of an account, a deal, a stakeholder or a pipeline, assembled from connected systems; the skills, the scheduled tasks and the Slack app are all wrappers around that one act. There is no records layer, no sequencer, no dialer and no mechanical tier underneath.

The vendor built the second generation of the product around this explicitly, having previously sold a product usage signal tool, and its own comparison of itself is against consumer chat assistants rather than against sales software. When a vendor argues that its differentiator is the grounding rather than the generation, the model is still the thing being sold.

Scheduled unattended work is a shipped feature and it is published in detail: tasks that run on a schedule or on an event trigger, with examples of a weekly pipeline summary every Friday, a deal risk monitor every Monday and a daily account research reminder, each showing a last run and a next run. No approval gate, review step, confidence threshold, audit trail of agent actions or escalation path is described for any of it.

What holds this at the top of the band rather than lower is the blast radius. The output of an unattended run is a briefing delivered to the customer's own team, not a message to a prospect, so an error surfaces in front of the people best placed to catch it. The product does draft outbound follow ups, and the send is a human act performed elsewhere.

At the top of the band for a shipped mechanism rather than a claim: every response carries clickable source citations back to the underlying record, which the vendor presents as the answer to how a rep verifies what they are told, and a reviewer quote confirms citations are how they validate. A research mode with a stated maximum reasoning effort is exposed to the user as a visible control. What none of that discloses is the system.

No model, provider, family or version is named, no accuracy or error rate is published for the deal risk detection, the stakeholder inference or the pipeline analysis, and no fallback is described. This is the second vendor graded in one session whose transparency mechanism is about the answer rather than about what produced it, and the distinction is worth holding: tracing an output to its evidence is real and useful, and it is a different question from model transparency.

Three named customers, each attributed to a named individual with title and employer, each with its own linked case study: a revenue vice president at Hex, a co founder and chief operating officer at Monte Carlo, and a revenue enablement lead at Retool. The quotes are substantive rather than decorative, one describing an evaluation of more than a dozen competing tools and why this one differed, another describing what the product replaced.

One quantified figure appears, being adoption above eighty percent of a revenue team, attributed to a named executive at a named company. An implementation estimate is published with unusual candour, being roughly five minutes per integration, one to two weeks to configure methodology and messaging with an administrator, and the observation that the real work is organisational rather than technical.

Off the top of the band because the evidence is almost entirely qualitative, no methodology sits behind the single percentage, and no independent analyst position or review platform record was located.

This product does not send. It drafts follow ups tuned to an individual and a human sends them from their own systems, so the axis bites less here than on a sequencer and the grade is recorded on what is knowable, per the standing rule for categories that do not send.

What can be said is that no sending regulation, consent standard, suppression duty or unsubscribe mechanic appears anywhere, and an acceptable use policy is published but was not read this pass, which is where a restriction on how generated messages may be used would sit if one exists. The vendor's own marketing consent language covers its own list and includes an opt out route.

The compliance question that does bite on this product is a different one and it is graded on the privacy and licensing axes: what may lawfully be assembled about a person before a seller writes to them.

Rights routes are published and one of them is a first for this index: alongside deletion, correction and an opt out of sale or sharing, each addressed to a named privacy mailbox with a prescribed subject line, the vendor offers an interface for automating those requests to anyone who asks for it. Programmatic privacy request handling is rare at any size. Against that sits a set of absences a regulated buyer asks about first.

The policy was last updated in March 2025 and is around seventeen months old at grading. No legal basis, data protection officer, European representative, supervisory authority, retention period or sub processor list appears. International transfer is addressed by disclosing the risk rather than naming a mechanism, stating that information may be stored in jurisdictions that do not provide equivalent protection. The marketing page meanwhile advertises full compliance with the European and Californian regimes, and the policy contains none of the machinery the first of those requires.

At the top of the band, and the policy is unusually explicit about a practice most vendors leave implicit. The vendor names a category it calls lead enhancement information, being contact details, employment and education history, social media activity and coarse location, collected about its customers' current, former and potential customers from third party data providers, social media platforms and publicly available sources, and states plainly that it makes that information available to its customers.

Naming the categories collected and the source classes is more provenance than most of this index publishes, and a deletion route exists for the person in the data. Off the top of the scale because no supplier is named, no lawful basis is stated for assembling social media activity and location about people who have no relationship with the vendor, and the route is reactive: the person must already know they are in the system, which the benchmark on this axis solves by notifying on first entry and publishing a lookup.

Most of the surface is clean and conventional: official connectors into systems the customer already licenses, reading rather than acting, no browser extension, no credential rotation, no automated action taken from a customer's own accounts, and a documented five minute connection flow per system. Two named sources keep it at the top of the band rather than higher.

The professional network appears in the published integration list with no method stated, and the privacy policy names social media platforms as a collection source for third party personal data. Both are categories of source whose terms restrict automated collection, and neither is accompanied by a stated method or a conformance position.

A separate exposure sits alongside and is unaddressed: conversation intelligence and enrichment platforms flow into this layer, and several of those licences restrict derivative use of their data by an outside system.

The training question is answered here, and it is answered in the affirmative, which makes this vendor the exception in a corpus where the usual finding is silence or a refusal. Under its analytics purpose the privacy policy states that information is used to improve the service and to develop new products and services, including to train the vendor's own artificial intelligence models. Scope deserves precision rather than alarm.

That policy governs what the vendor collects as a controller, being website, account, marketing and the lead enhancement records it assembles about third parties, and it states separately that customer content is processed as a service provider under customer instruction, with those terms sitting in an agreement that was not read this pass.

So the disclosed training permission covers at least the third party enrichment corpus, and the position on the customer's own connected calls, mail and records is undetermined here. Recorded as a question the policy makes urgent rather than a finding about the customer corpus. The counterweight is real: a vendor that writes an adverse practice down is more transparent than one that stays silent.

The impersonation record is clean and the surveillance depth is among the greatest graded, which is the shape this axis first met at Cirrus Insight. Nothing pretends to be a person: no agent persona, no synthetic voice, no manufactured location, no automated social action, and the seller who sends the follow up is real. What the person on the other side does not know is the depth of the dossier behind it.

Call transcripts, internal messages, mail history, the professional network, regulatory filings and news are assembled into an account picture, and the vendor separately collects employment and education history, social media activity and coarse location about that person from third party providers. The product's stated aim is a follow up speaking to each person's exact pains, which is effort manufactured from material the recipient never disclosed. Article 50 of the European artificial intelligence regulation goes unmentioned.

Depth on the read side is the strongest part of the product and it is documented rather than asserted. Named sources are the system of record including mail, commercial history and notes, two conversation intelligence platforms for transcripts, internal messaging, a document store, the professional network, regulatory filings, earnings calls, news and company websites, with a stated setup time of roughly five minutes per system.

A public documentation site runs on its own subdomain, and the messaging app integration is a working surface rather than a notification feed, answering a question posed in a channel with a structured reply. Off the top of the band because the connections are consumption rather than exchange: no public interface documentation, webhooks, marketplace or agent endpoint was located, so nothing described lets another system query this one.

The transfer section states that the vendor is headquartered in the United States with operations and vendors there and in other countries, and that information may be transferred to, stored in or accessed from jurisdictions that do not provide protection equivalent to the reader's home jurisdiction. That is a disclosure of the exposure with no mechanism attached and no geography named.

No hosting provider, region, residency option, customer choice or sub processor list appears anywhere across the pages read. The gap is more consequential here than on a lighter product because of what is being centralised: an aggregate of a customer's calls, internal messages, mail and commercial history in one external system, which is the exact combination a European or regulated buyer takes to their own reviewers first.

At the bottom of the band, and what earns it is that the certification is named with its type rather than asserted loosely, which is the distinction this band turns on: an audited service organisation report at type two, with reports stated to be available on request. Alongside it, single sign on with a named federation standard, encryption claimed for data in transit and at rest, and a public documentation site. Held down by what is missing around it.

No auditor is named, no audit period or report date is given, no trust portal or status page was located, and no penetration test or vulnerability disclosure route appears. The self service standard set earlier in this index, where the current report and the penetration test attestation can be retrieved without asking anyone, remains the bar. Reports on request is one step short of it and a long step ahead of an assertion with no type.

No price, tier, unit, range, floor or pricing model is published, and no pricing page exists in the navigation or the footer across the pages read. Every path resolves to a demo request, including the link labelled as signup. A search for this vendor's pricing returns only third party directory entries recording contact for pricing, plus a stale profile describing a previous version of the product.

The unit question is the one a buyer most needs answered and it is entirely open: this is sold to revenue teams as a knowledge layer with reusable skills and unattended scheduled tasks, and nothing indicates whether the meter is a seat, an account, a query, a task run or a platform fee. The vendor publishes an unusually candid implementation estimate down to minutes per integration, which makes the absence of any commercial figure beside it more conspicuous.

Governing terms and an acceptable use policy are published, and a deletion route for personal information exists with a named mailbox and an offer of an interface to automate such requests, which is what holds this above the bottom of the scale. What a departing customer would actually ask about is absent. No retention period, post termination window, deletion timeline, deletion artefact or export function is named on any page read.

The terms of use were not read this pass and are where a return or destruction commitment would sit. The specific exposure worth naming is that the asset accumulated here is derived rather than supplied: the value builds in the configured methodology, the organisation's skills library and the assembled account picture, and none of the source systems that fed it holds a copy of that.

No sending infrastructure exists and none is claimed. Drafted follow ups leave through whatever the seller already uses, so sender reputation, authentication, pacing and bounce handling all sit with the customer's own mail provider and engagement platform rather than with this vendor. Graded on what is knowable with the structural context recorded, per the standing rule for products that do not send.

The one adjacent observation worth keeping is that a product generating individually tailored follow ups at the volume a full pipeline implies puts pressure on a sending system it neither controls nor describes, and nothing in the published material advises a buyer on that boundary.

The buyer is stated consistently and the reference customers corroborate it rather than contradict it. The product is positioned for enterprise sellers and revenue teams, sold through solution pages by function, and the three named customers are technology companies of a consistent profile with the quotes coming from a revenue vice president, a chief operating officer and an enablement lead, which is the buying committee the positioning describes.

The signup form collects sales team size, so the vendor segments on headcount internally. Off the top of the band because none of that is published as a boundary: no headcount band, deal size, vertical statement or stated ceiling appears, and nothing tells a reader at what size this becomes the wrong purchase. The strongest statements on this axis in the index name the point at which a buyer should leave.