AiSDR

Nothing survives the removal test as a product this company could sell. Every capability offered is inference: building the ideal customer profile from a conversation, searching for leads by plain description, researching the three most relevant data points on each contact, composing email and social messages, interpreting replies, handling objections and booking meetings.

Strip the model layer and what remains is not a diminished version of this product but two purchased inputs, a licensed contact database and a set of mailboxes bought from a mail provider, with nothing proprietary joining them. The vendor makes the same argument from its own side, positioning explicitly against sequencing platforms on the basis that those help a team send while this one decides what to send and to whom, which is a claim only a company whose entire deliverable is model output can make.

Founded 2023 as an agent company rather than as a platform that added an agent later, which is the vintage split this convention describes, and it is the third record in this index to take the top band here after two other agent native companies founded after 2022.

Four named controls, published by the vendor rather than reported by a reviewer, which is more than any agent record graded in this index so far. The operator can run the agent autonomously with replies inside five to ten minutes or switch to a copilot mode in which a person approves messages before they leave, so the autonomy level is chosen rather than inherited.

A volume ceiling is enforced on contacts researched each month, with an in application notification as the limit approaches and an explicit choice at the limit between pausing campaigns and paying for overage, which is a hard stop that exists by construction rather than by policy. Outbound can be paused at any time, and the vendor names the situations it expects that to be used for. Suppression lists are supported at setup and existing customer lists can be used for suppression.

Held below the top band because the boundary itself is undescribed: nothing states what the agent may commit to when it handles an objection or answers a question about the product, whether the copilot setting applies per campaign or across the account, what escalates to a person automatically, or what happens to a conversation the agent has already mishandled. Ask what the agent is permitted to assert about the product, and whether approval mode can be set per campaign.

A training claim on the marketing surface and nothing underneath it. The strategist agent is advertised as trained on fourteen thousand real campaigns, which is a specific and load bearing statement about how the system came to work, and no accompanying material states what those campaigns were, whose they were, what was extracted from them or how a customer would know whether their own campaigns are in that corpus.

No model provider, family or version is named anywhere reached on this pass, no model card exists, and no evaluation, accuracy or quality figure is published for lead research, message composition, reply interpretation or objection handling.

The vendor states that its own engineers write the prompts in the background so the customer never has to, which is a usability argument that also means the instruction layer determining what is said in the customer's name is neither visible nor editable by the person whose reputation carries it. A dedicated page addressed to artificial intelligence assistants exists on the vendor's site and was not retrieved on this pass. Ask which models power research and composition, what the fourteen thousand campaigns consist of, and whether the customer can inspect or edit the prompts sent in their name.

The most useful number the vendor publishes is its most modest one, and the impressive ones carry no population at all. In an answer about expected results the vendor states a conversion rate of one to three percent from cold prospect to booked meeting, or one to three meetings per hundred contacts, and labels it directional, states it is drawn from historical performance across teams using the product, and warns that it varies by market fit and offer strength.

That is a caveated planning benchmark rather than a boast, it is unusual in this category, and it is enough for a buyer to model cost per meeting against the published tiers. Beside it on the same page sits a carousel of figures with no denominators and no attribution to any named account: a reply to demonstration rate of twenty eight percent, an email reply rate of twenty six point six percent, a positive response rate of about eleven point eight percent and a response rate of about three point five percent, several of which cannot describe the same population and are not distinguished.

One testimonial states that a customer spent a figure rendered as the placeholder letter X and closed one point eight times that in revenue, so the denominator is literally absent from the published claim. Customer count is given as more than two hundred and fifty. Ask which population and period each headline rate describes, and for the meeting acceptance and qualification rate rather than meetings booked.

Real suppression apparatus, and no stated position on the law that governs any of it. What exists is concrete and published: suppression lists are collected during a setup the vendor times in minutes, existing customer lists in the connected record system can be used as suppression, unsubscribed leads are cleaned out of campaigns, bounce checks run three times before a send, and a monthly ceiling caps how many new people can be contacted.

Those are mechanisms rather than promises and they are better than the cohort. What is absent is the legal frame around them. Nothing published states the consent basis for contacting people sourced from the vendor's own database who have no relationship with either party, whether prospects in jurisdictions requiring prior consent are screened or excluded, whether any preference or do not contact register is checked, or what obligations the vendor considers to transfer to the customer whose domain and name the message carries.

The vendor operates a European cookie consent surface and so plainly sells into Europe, which makes the omission a live question rather than a theoretical one. Ask what consent basis supports the contacts supplied, whether prospects are screened by jurisdiction, and which party the vendor considers the controller for the prospect data it sources.

A privacy statement, terms and a granular consent manager are published, and the instruments that would answer a buyer's questions could not be reached. The vendor publishes a privacy statement, terms of service, a responsible disclosure policy and a separate European cookie policy, and its consent manager offers functional, preference, statistics and marketing categories separately rather than as a single accept control, which indicates a deliberate rather than a default configuration.

Beyond that nothing could be established on the routes taken. The trust centre is a hosted compliance portal that renders client side and returned no content to this retrieval, so what it holds regarding a processing agreement, subprocessor list, transfer mechanism or retention period is unknown rather than absent, and this grade should be read as recording what a machine readable pass could establish.

The stakes are set by the corpus rather than the company size, because the platform holds researched profiles of individuals at prospect organisations who never contacted anyone, their social activity, the messages sent to them and the replies they wrote. Ask for the processing agreement, the subprocessor list, the retention period for researched prospect profiles and reply content, and how a contacted individual exercises rights over data held about them.

A database of three hundred million records is offered and no account is given of where any of it came from. The vendor states that prospecting draws on a native global database of more than three hundred million leads, on a professional network database for real time social data, and on any publicly available lead data from the web used to tailor messaging, and lead search and enrichment are included in the subscription rather than sold as a separate data licence.

Naming those three sources by category is more than nothing. What is missing is everything that makes a source assessable: no supplier, aggregator or licensing arrangement is named for the native database, nothing states whether records are collected, purchased or licensed, no lawful basis is stated for holding personal data on individuals who have no relationship with either party, no indemnification position was located, and the phrase describing the use of publicly available web data does not say what collects it or under what terms.

Enrichment consumes roughly three credits per contact according to the vendor's own knowledge base, which tells a buyer what enrichment costs and not what it draws on. Ask which providers supply the native database, on what basis the records were collected, what mechanism gathers the publicly available web data, and whether provenance is indemnified.

Substantial automated activity against a professional network with the method left entirely undescribed, which is where this convention places the middle band rather than the bottom. The agent sends connection requests, direct messages and mail credits, and at the upper tiers adds voice notes and video messages, while prospecting tracks profile visitors, keywords and post engagement and draws on the network's own sales database for real time data.

Connected accounts scale with the tier from one to five to twenty, so a large deployment runs automation across twenty separate identities on that network. Nothing published states how any of it is executed, whether through sanctioned interfaces or otherwise, and no conformance position, terms reference or allocation of account risk appears anywhere. The account that gets restricted belongs to the buyer.

This sits above the bottom band on a distinction worth stating precisely, because two vendors in this index hold that band for a different thing: they market evasion itself, selling human behaviour simulation, published guidance on stepping around the network's stated limits, or instructions for creating a fictitious account. Nothing of that kind appears here. Undescribed method plus multiple accounts is the middle band; evasion as a marketed feature is what takes a record below it. Ask how social actions are executed and under whose terms, whether the network's sales database is accessed by sanctioned interface, and who bears the loss if an account is restricted.

One advertised sentence raises the central stewardship question and no published material answers it. The strategist agent is promoted as trained on fourteen thousand real campaigns. Real campaigns are customers' campaigns, containing their positioning, their messaging, the sequences that worked for them and the replies their prospects wrote, so the claim describes a system that learned from the operational material of the customer base.

Nothing published states whether that training corpus is consented, whether it is aggregated or identifiable, whether a current customer's campaigns and replies feed models serving other customers including their competitors, or whether an opt out exists. For a product where message quality is the entire value, learning across customers is commercially obvious and competitively sensitive at the same time, which is exactly why it needs stating.

Around it, no model provider is named so the processing chain is invisible, no governance document, evaluation record or red teaming artifact was located, and the hosted trust portal renders client side and returned no content to this retrieval, so material may exist there that a machine readable pass could not see. Ask whether customer campaigns and prospect replies train models serving other customers, whether an opt out exists, and what the fourteen thousand campaigns consist of.

The stated design goal is that machine written outreach is indistinguishable from the customer, sent in the customer's name, and at the upper tier extended to a synthetic voice and face. Onboarding asks the customer to supply one real email so the agent can learn their voice, so voice matching against a specific identifiable person is a documented setup step rather than an emergent effect.

Messages then leave from mailboxes carrying the customer's domain and from that person's own social account, and the upper tier adds generated video in messages and generated voice notes in social messages, so a recipient may receive what presents as a personal recording from someone who never made one. The agent also answers replies and handles objections within minutes, so a prospect can hold an entire exchange believing they are corresponding with the named sender.

Nothing published states that a recipient is told a message was machine generated, that the agent identifies itself when asked, or that generated audio and video are marked as generated, and the disclosure obligation under Article 50 of the European artificial intelligence regulation, in force since 2 August 2026, is not acknowledged anywhere by a vendor operating a European consent surface.

This band is where the index records a product whose design intent is that the automation should not be apparent to the person receiving it. Ask what a recipient is told, what the agent says when asked directly whether it is a person, and whether generated voice and video carry any marking.

Two deep integrations and a short list around them, with no route to build anything else. The vendor states its integrations are a mail provider pair, two customer record platforms, one dialer and one scheduling tool. The record platform integrations are genuinely deep rather than nominal: two way synchronisation, active and static list sync, personalisation from record properties, activity logging, account scoring, call task creation with scripts, contact creation and enrichment write back, with the second record platform reserved for the top tier.

Beyond those six named systems nothing exists. No public interface documentation, no automation connector service and no additional record platform support were located, and an independent review of the workflow reaches the same conclusion, which means a buyer running a different record system or wanting the agent's activity in a warehouse or a business intelligence tool has no supported path.

Website visitor tracking with enrichment and outreach appears at the top tier, which extends the product's reach without extending its connectivity. Ask whether a public interface exists or is planned, and what a deployment looks like for a team on a record platform outside the two supported.

The question is squarely applicable here and the answer could not be established on the routes taken. No hosting provider, region, tenancy model, residency election or recovery objective was reached on any surface retrieved, and the hosted trust portal that would ordinarily carry this material renders client side and returned no content to this retrieval, so this records a retrieval limit rather than a finding that nothing exists.

What can be said is what the architecture requires regardless of where it runs: the platform provisions and operates mail domains and mailboxes on the customer's behalf, holds researched profiles of prospects across whatever territories the customer sells into, stores the messages sent and the replies received, and connects into the customer's record platform, so the data footprint is wider than the subscription price suggests.

A buyer contacting European prospects from a United States incorporated vendor of this size should treat residency and transfer as first order questions rather than procurement formalities. Ask which provider and region host the platform, whether any regional processing option exists, what the tenancy model is, and where mailbox content is stored.

A hosted trust portal exists and would not render to a machine, which is a different finding from an absent security surface and is recorded as such. The vendor links a trust centre on its own subdomain, built on a recognised compliance automation platform, alongside a published responsible disclosure policy. A portal of that kind ordinarily carries certification status, policy documents and a request route, so material almost certainly sits behind it.

It renders client side and this retrieval runs no scripting, so the page returned descriptive metadata and no body, and no certification, audit period, auditor or report could be confirmed on this pass. That distinction matters for a buyer as much as for this record: a security posture that only a browser can read is unavailable to procurement tooling, to automated vendor risk assessment and to any assistant a buyer asks, which for a vendor of this size is a practical disadvantage rather than merely a presentational one.

Nothing here should be read as asserting the vendor holds no certification. Ask for the current attestation with its audit period, scope and auditor, and for a machine readable summary or a document set that does not require a rendered portal.

The most complete pricing disclosure recorded in this index, and it comes from one of the smallest companies on this list. All three tiers publish a monthly rate and an annual rate, and the annual figures are internally consistent at exactly twenty percent below twelve monthly payments on every tier, which was checked rather than assumed.

Each tier states the metering unit as contacts researched per month at two hundred, eight hundred and two thousand five hundred, the number of users, and the number of domains, mailboxes and social accounts included, so a buyer can see both the price and the capacity that price buys.

The usage economics that decide the real bill are published rather than implied: unused messages roll over while the subscription is active and expire at its end, mid quarter upgrades are prorated, volume caps are enforced with notification before the limit, and the choice at the limit between pausing and paying overage is stated.

Both managed service options carry published prices, at one hundred and forty nine dollars per campaign at the middle tier and two thousand five hundred dollars a month at the top, where the convention explicitly withholds this grade from vendors whose load bearing additions are unpriced. The entry tier is purchasable through a checkout link with no sales conversation, and the absence of a free trial is stated plainly with the vendor's reasoning rather than buried.

Two gaps are worth naming without moving the grade: the overage rate itself is not published, and the price of additional domains appears only in the knowledge base. The finding this record contributes is the one this index keeps recording, that scale predicts disclosure everywhere except price, where the smallest vendors publish and the largest do not.

Cancellation terms are clear and the question of what the customer owns afterwards is not answered anywhere. Termination itself is straightforward and published: the subscription can be cancelled at any time, and unused message credits expire when it ends. Activity is written continuously into the connected record platform, so contact activity, call tasks and enriched fields already sit in a system the customer owns, which is portability by construction for the operational record.

The unresolved question is specific to this product's architecture and consequential. The vendor buys and warms the sending domains and mailboxes on the customer's behalf as part of onboarding, and warmed domains take the vendor's own stated thirty days or more to reach full sending capacity, so they are a real asset the customer has paid to build. Nothing published states who holds registration of those domains, whether they transfer at termination, or what happens to the mailboxes.

A departing customer may be leaving behind the sending reputation their outbound depends on. Beyond that, no export statement covers researched prospect profiles, message and reply history, personas, sequences or campaign configuration. Ask who owns the domains and mailboxes at termination and whether they transfer, and what exports and in what format.

A real control set, published per tier, and one method left undescribed at the point where the comparable record in this index made its name. Mailbox and domain provisioning is part of the product with counts stated per tier at three, six and eighteen mailboxes across one, two and six domains, which caps volume architecture transparently rather than offering unlimited accounts.

Around it the vendor publishes ongoing domain warmup, automated inbox rotation, domain health tracking, three bounce checks before a send, bounce prevention and automatic removal of unsubscribed leads from campaigns, and independent review reporting adds sender authentication record configuration, sending caps, sender balancing and randomised sending windows.

The vendor also builds warmup time into its commercial terms, stating that maxing deliverability takes thirty days or more and using that to justify the quarterly commitment, which is an honest statement of a constraint most vendors leave the buyer to discover. Held below the top band on two points.

The warmup method is never described, and the specific distinction this index has recorded before is whether a vendor uses reciprocal pools in which subscriber mailboxes send each other synthetic mail to manufacture reputation; the record that earned the top band here named that practice and refused it, and this vendor is silent. And no complaint threshold, bounce ceiling or stopping rule is published for a system sending on the customer's own domains. Ask whether warmup uses a reciprocal network or genuine traffic, and what bounce or complaint rate halts sending.

The tier ladder is itself a fit statement, and the vendor supplies the tools to test fit before buying. Each tier is labelled with the buyer it is for, from a single founder selling alone, through a team proving outbound as a channel, to a team making it a reportable pipeline source, and the specifications underneath make the boundaries concrete in contacts, users and sending infrastructure rather than in adjectives.

A published readiness guide asks whether the product suits the reader, industry pages address healthcare technology, education technology and marketing technology specifically, and a return calculator and a strategist tool let a prospective buyer generate a sample plan before any conversation. That is more pre purchase self qualification than any other record in this cohort offers. Held below the top band on two absences.

Nothing states a floor, meaning the deal size, sales cycle or contract value below which outbound at this volume cannot pay for itself, which is the condition that actually decides fit for the smallest buyers the entry tier invites. And no localisation or non English capability is stated anywhere, while the lead database is described as global, so a buyer selling into non English speaking markets cannot establish whether the agent writes in those languages. Ask what deal size makes the entry tier pay back, and which languages the agent composes in.