Close

The established platform pattern, and the vendor's own launch language settles it: for more than a decade Close helped sales teams engage leads with built in calling, texting, email and automation, and now it is taking the next step with an agent.

Remove every model and what remains is a complete and mature product, namely a full pipeline with leads, contacts, opportunities, custom objects and smart views, a native telephone system with power and predictive dialing, recording, voicemail drop and live coaching, two way email with templates and scheduling, text messaging, web forms, automated workflows and a reporting suite through to funnel analysis.

That is what 11,500 businesses bought and it is what the two billion logged interactions were run on. The agent layer is real, generally available and included on every plan, and it is an addition to a decade old product rather than the reason the product exists.

Top of the band on an oversight artefact most vendors here do not have, held off the highest grade by where that oversight sits in the sequence. The agent acts on its own: it telephones a new lead as it arrives, holds a qualifying conversation, books a meeting and writes the result back to the record, with no approval step anywhere in that loop.

What the vendor publishes against that is a record rather than a gate, and it publishes it plainly: every action the agent takes is logged, reviewable and editable by the team, every call is transcribed and summarised automatically, and a user can review, edit and leave feedback from the lead record at any time.

Three further controls are real and enumerated: the operator supplies the script, qualification criteria and knowledge sources, so the agent's scope is configured rather than open; blackout dates prevent automated email and text sends on holidays or when a team is unavailable; and the top tier can restrict which users see the agent's configuration, prompts and settings at all. The gap is that all of it is retrospective. Nothing withholds a violating output, nothing escalates by risk, and nothing pauses the agent before it dials.

Top of the band on one strong sentence that answers a different question than the axis asks. The vendor states that the agent operates entirely within Close and that customer data never leaves the platform to a third party AI vendor, which is a clear and checkable claim about where data goes and is more than most competitors offer. What it is not is model disclosure.

No provider, family or version is named anywhere, no inference location is stated, and no retention position is published for prompts, transcripts or generated content. The data boundary claim also invites an explanation it does not receive, because a platform whose customer data never reaches a third party AI vendor is either running its own models or holding contractual carve outs with the providers it uses, and neither is described. Recorded as observed: the security page was not read this pass and is the most likely place for anything further.

Strong on scale and attribution, thin on measured outcomes. The scale figures are specific and checkable: more than 11,500 businesses, two billion calls, emails and meetings run through the platform, and a review platform rating of 4.7 from more than two thousand reviews with the reviewer base concentrated at small business, which corroborates the positioning rather than contradicting it.

The agent launch disclosure is unusually precise and deserves credit, giving a beta of 306 businesses and more than 818,000 calls before general availability, where most vendors announce a launch and no numbers at all. Testimonials carry a full name, job title and named employer rather than an initial.

Off the top of the band because none of it is a measured result: there is no analyst evaluation, no commissioned economic study and no customer outcome with a stated basis, and the quoted praise is impressionistic, including a customer scoring the agent nine out of ten and another saying it outpaces a former team of callers, neither with a number behind it.

Four shipped controls rather than a policy page, which is what earns the band. Unsubscribe management is built into the platform rather than left to the sender. Sending limits actively prevent a connected mail account from exceeding the thresholds its provider sets, described in the vendor's own words as safeguarding sender reputation. Blackout dates stop automated email and text sends on holidays or when a team is unavailable, which is a timing control almost nothing else here has.

And the messaging feature carries a plain statement that carrier registration is required before texting United States numbers, which is the obligation most platforms leave the buyer to discover. Two things hold it off the top. The predictive dialer, which dials multiple numbers simultaneously and connects whoever answers, is shipped on the top tier with nothing published about the abandoned call rate, which is the single regulated parameter of that technology and which a competitor in this index names explicitly. And three of the four controls, unsubscribe, sending limits and blackout dates, arrive only on the third tier, so compliance machinery is sold as an upgrade.

Bottom of the band, graded on a published surface rather than on contents. Five separate documents are linked from the footer of every page: a privacy policy, a dedicated General Data Protection Regulation page, a dedicated California regime page, a security page and terms of service. Splitting the jurisdictional material into its own pages rather than burying it in a single catch all is a real choice and it is uncommon at this price point.

The posture is also structurally lighter than most vendors graded here, because the platform is a processor of records its customers supply rather than a broker of third party personal data, so the difficult question about people who never contracted with anyone barely arises.

Off the top of the band on four counts: none of the documents was read this pass, no processing addendum appears in the footer alongside them, no data protection contact or European representative is named on any read surface, and no sub processor list was located despite the platform carrying telephony, transcription and model inference.

The lightest version of this question, with one gap in it. The platform does not sell a contact database, does not claim one, and holds what its customers put into it, so the provenance of the bulk of the data is the customer's own activity: calls placed, emails synced, forms submitted, records imported.

Against that the product does ship enrichment on every tier, described as automatically enriching contact and company records with external intelligence and elsewhere as pulling live company and contact data from public sources. Public sources is the whole of the published provenance. No supplier is named, no method is described, no accuracy or coverage figure is given, and there is no notification route or self service lookup for the people whose records are being enriched. That is the same one phrase disclosure another vendor in this index gives for the same feature, and it is graded the same way here.

A clean posture built on official routes throughout. Nothing scrapes, nothing automates actions against another platform's user surface, no browser extension pulls data off professional profiles, and no credentials for third party accounts are held.

Mail and calendar connect through the major providers with two way sync, telephony runs on the vendor's own lines or a line the customer brings, conferencing integrates with the three main platforms for meeting capture, and the wider catalogue runs through documented interfaces with both key and delegated authorisation available.

The detail worth naming is the carrier registration disclosure on the messaging feature, which shows the vendor tracking obligations at the network layer rather than only at the application layer. Off the top of the band because no conformance position is published in writing with mail providers or carriers beyond the sending limits feature itself, and because a business messaging channel appears in the product line with its platform policy obligations nowhere addressed.

Earned on a single plainly worded boundary statement that most vendors here decline to make: the agent operates entirely within the platform, and customer data never leaves it to a third party AI vendor. That is the clearest published answer in this index to the question of where the conversation goes, and it is paired with a second commitment that every action the agent takes is logged, reviewable and editable by the customer's own team, which gives a buyer a way to check the first.

Off the top of the band on two counts, and the second matters more as the product grows. The statement sits on a product page rather than in a contract, and the only holder of the top grade on this axis earned it with contractual terms rather than marketing copy.

And the cross tenant training question is untouched, which is conspicuous here because the platform is accumulating an unusually rich conversation corpus, with a notetaker recording and transcribing every synced meeting, a call assistant transcribing calls, and an agent holding live qualifying conversations at volume.

The most direct instance in this index of the shape that Article 50 of the EU AI Act was written for, and it is now generally available at scale. An agent carrying a human first name telephones a lead within moments of the lead arriving, holds a live two way qualifying conversation in synthetic voice, and books a meeting.

That is not a generated email or a cloned voice reading a script, it is a machine conducting a conversation with a person in real time, and the beta alone ran more than 818,000 calls. Nothing published states whether the agent identifies itself as artificial at the start of a call, and the regulation is not mentioned anywhere. Two things keep this at the middle of the band rather than lower.

Every call is transcribed and logged, so the record of what was said exists and the customer can inspect it, which is more than the vendors selling agent conversations without any artefact. And there is no manufactured location anywhere in the calling product, no area code matched to the prospect, which is a restraint several dialer vendors here do not show. Voicemail drop ships on the upper tiers with no consent position published.

The most complete agent era and developer surface graded so far, and the decisive detail is again placement rather than breadth. A native Model Context Protocol server is listed as an integration on every plan including the nine dollar entry tier, documented in the help centre, and marketed by naming both major assistants directly in the pricing matrix, so a solo operator paying the lowest price can put their pipeline in front of an external model.

That is the sixth such server in this index. Around it sits a genuine developer surface: a dedicated developer portal, programmatic access with both key based and delegated authorisation on every tier, and an event log interface for change capture, which almost nothing else here publishes.

Named integrations cover conferencing, customer data infrastructure, scheduling and an automation platform reaching thousands of applications, with more than a hundred applications in total, plus built in importers from three named competing platforms. Two operational artefacts complete it and are rare at any size: a public system status page and a public product changelog.

The residency question goes unanswered on every surface read this pass. No data centre region is named, no residency option is offered to the customer, no cross border transfer position is stated and no sub processor list appears.

That gap has more weight here than it would for a lighter product, because this platform holds not only the pipeline of record but the recordings and transcripts of every call and meeting a team runs through it, which is among the most sensitive material any vendor in this index processes, and because telephony and transcription both imply infrastructure whose location is a real question.

Recorded as observed rather than concluded: this is a failure to locate documentation rather than a finding that none exists. A dedicated security page and separate jurisdictional pages for the European and Californian regimes are all published and were not read this pass, and any of the three is a plausible home for a residency statement. Re verify there.

The certification is named with its type, which sounds trivial and is exactly what separates this from several vendors graded lower: SOC 2 Type 2, stated on every plan in the pricing matrix rather than buried. Around it the access controls are enumerated per tier rather than asserted, covering two factor authentication and single sign on across all plans, four predefined roles everywhere, and custom roles, lead visibility rules, user groups, required fields and field editing restrictions on the top tier.

Two operational artefacts add real assurance: a public system status page and a public changelog, both of which let a buyer verify behaviour rather than take a claim on trust. Off the top of the band because no auditor is named, no audit period is stated, no route to the report is published, no penetration testing is attributed and no vulnerability disclosure policy was located, and because the dedicated security page was not read this pass. One substantive gap worth naming: the only single sign on offered at any tier is through one consumer identity provider, with no enterprise federation available even on the top plan.

Complete across all three layers of the bill, which is rare for a product that mixes seats, telephony usage and model credits. The seat layer publishes four tiers at nine, thirty five, ninety nine and a hundred and thirty nine dollars per user per month annually, each with its monthly equivalent, alongside a full feature comparison matrix showing tick and cross state for every capability across nine categories, so the gates are visible before purchase rather than after.

The usage layer is the unusual part: calling and messaging are stated to be passed through at cost, with a linked full breakdown, an indicative rate near two cents a minute and lines from about a dollar a month. The model layer publishes an included credit allowance per tier, a shared organisation ceiling, and the fact that credits refresh monthly and do not roll over, which is a disclosure against the vendor's own interest.

Add ons carry published prices rather than a conversation, covering premium numbers per line, the call assistant at a monthly fee plus a per minute rate, and additional organisations. The terms match: a fourteen day trial with no card, a thirty day money back guarantee described as a full refund with no questions asked, and change or cancellation at any time on monthly plans. The vendor states plainly that no sales conversation is required.

Export is a property of the cheapest plan rather than an upgrade. Data import and export both appear on all four tiers including the nine dollar one, with no metering, no volume ceiling and no credit cost, and programmatic access with an event log interface sits alongside on every tier as a second and more granular extraction route. Migration is documented inbound from three named competing platforms plus spreadsheets.

Commercially the customer is not held either, with monthly plans cancellable at any time and a thirty day money back guarantee. One post termination statement is published and it is the reason this sits below the top of the band rather than at it: after a trial lapses the account pauses and the data remains available for a limited period before deletion. That is a real commitment to eventual deletion, which most vendors here do not make at all, and the window is undefined.

No deletion confirmation artefact and no formal data return mechanism is described, and nothing addresses what happens to call recordings and meeting transcripts specifically. Re verify in the published terms.

Bottom of the band, earned by one control that directly addresses the failure mode this axis exists to catch. Sending limits prevent a connected mail account from exceeding the thresholds its own provider sets, described by the vendor as safeguarding sender reputation, which is a governor almost nothing else in this index ships.

Around it sit unsubscribe management inside the platform, email scheduling, blackout dates that suppress automated sends when a team is away, template performance reporting, and a cap on connected accounts per tier. The architecture helps too: sending rides the customer's own mailbox through two way sync, so the buyer carries their own reputation rather than a shared pool, and bulk email is unavailable on the two cheaper tiers, which limits how much damage an unprepared team can do.

What is absent is the rest of the standard set: no warmup, no guidance on sender authentication, no bounce or complaint threshold, no blocklist monitoring, no inbox placement testing and no domain health reporting. And the governor arrives on the same tier as bulk sending rather than before it.

Top of the band, and unusually concrete about who each price is for. Every one of the four tiers carries its own published buyer statement in plain language: founders and solo representatives who need a real system rather than a spreadsheet, small teams wanting one place for every call, email and deal, teams ready to let the agent work leads while representatives close, and high volume organisations needing advanced controls and the fastest dialing.

A separate selection guide restates the choice at the foot of the page and tells a customer when to move up. Beyond price the segmentation is published by role, with dedicated pages for founders and owners, sales leaders and representatives, and by industry, with pages for coaching, business software, financial services and insurance. Independent review data corroborates the positioning rather than contradicting it, with the reviewer base heavily concentrated in small business.

Off the top of the band because no headcount, revenue or geographic band is stated numerically anywhere, and because the vendor publishes no ceiling: a team that outgrows the top tier is routed to a custom conversation rather than told where the product stops.