Artisan

Passes the removal test with nothing left over. Ava is the product, not a layer on one - there is no underlying database subscription, sequencer or dialer that survives if the agent is removed, because the buyer is not licensing software, they are contracting an output. Founded 2023 and built agent-first from the start.

Third A on this axis in the index, and the pattern established with Aimdoc and Airspeed holds without exception: every AI-native vendor of 2022 vintage or later takes A, every established platform takes C.

The autonomy dial is the best-conceived autonomy control in the index and it is worth naming as a pattern others should copy: Artisan replaces the usual binary on/off switch with a continuous setting, so the buyer chooses how much of the outbound workflow Ava owns rather than accepting the vendor's default.

Equally creditable is that Artisan publishes where the dial actually sits in practice - it reports that over 90% of customers run Ava fully autonomously - which is honest about the fact that a configurable control is not the same as a used one.

Held below A because the dial is the whole published oversight story: nothing documents what each position actually gates, whether any human approval step exists at any setting, whether guardrails withhold a non-compliant message rather than sending it, what triggers escalation to a person, or whether an audit trail of Ava's actions and reasoning exists. Agentforce holds the A on this axis because it publishes that machinery; Artisan publishes the control surface without it.

No model provider, family or version named for any part of Ava - not the research layer, not the copy generation, not the autonomous reply handling. No model cards, no evaluation results, no accuracy or hallucination measurement, and no description of how the agent decides what to say. The "Level 2 autonomy" framing used in coverage of the company is a rhetorical device rather than a reference to any published taxonomy, and Artisan does not define it. For a product whose entire value is what the model writes and decides, the model layer is a black box by omission.

The finding on this record is an on-the-record disavowal by the vendor's own CEO. Artisan built its public identity on the "Stop Hiring Humans" billboard campaign, reported at over a billion impressions. Asked by TechCrunch whether he believes AI will replace people, founder and CEO Jaspar Carmichael-Jack said no, and described the billboards as mostly done for attention.

A buyer reading Ava's marketing is entitled to weigh that against every other claim on the site, and this index grades it as directly material to evidence quality rather than as colour. Beyond it: company-reported figures at Series A (around 250 enterprise clients, $5m ARR, 50% average month-on-month growth) are corporate metrics reported through funding coverage, not customer outcomes; named customers exist (SaaStr, Quora) without attributed results; and independent review consensus is consistently reserved, describing a limited track record, few independent case studies, and messages that read as robotic in practice despite hyper-personalisation claims.

No published operational treatment of CAN-SPAM, GDPR, CASL or state privacy law surfaced in this pass: no unsubscribe or List-Unsubscribe implementation documented, no opt-out handling described, no consent or lawful-basis framework, no suppression or do-not-contact mechanism, and no statement of where responsibility transfers to the buyer.

That silence is more consequential here than at most vendors, because Ava sends fully autonomous cold outreach at machine speed from a bundled database of hundreds of millions of records into both US and European markets, and the customer is the sender of record. C rather than lower because nothing in the marketing celebrates evading controls - this is absence of a published posture, not an adverse one. Worth re-verifying directly against the vendor's legal pages.

No privacy documentation surfaced in this pass: no Data Processing Addendum, no sub-processor list, no stated lawful basis for holding the contact database, no Data Protection Officer or EU representative named, no cross-border transfer mechanism, and no data subject request route for the hundreds of millions of individuals whose records the platform prospects from.

Artisan operates as both a data supplier and a sender, which is the combination that attracts the heaviest obligations, and the public surface does not address them. Recorded as observed rather than concluded: this is a failure to locate documentation across a targeted search, not a finding that none exists, and it should be re-verified against the vendor's legal pages before being quoted.

A bundled contact database is the platform's central economic argument - it is explicitly sold as replacing a separate Apollo or ZoomInfo subscription - and its provenance is entirely undisclosed. A stated 22+ data sources with waterfall enrichment, and not one supplier, aggregator or licensing basis named.

The record count also does not settle: Artisan's own blog cites over 270m B2B contacts while third-party coverage of Artisan's claims reports 250m+ and 300m+, a spread of fifty million records on the number that defines the product. Same shape as Alta and the same consequence for the buyer, who inherits controller responsibility for records they cannot trace.

Lower exposure than the multichannel positioning suggests, on the evidence available. Artisan runs its own sending infrastructure rather than borrowing the customer's - third-party comparisons make the point explicitly when contrasting Artisan's bundled stack against assembling Smartlead plus Apollo plus warmed inboxes - and no credential custody, account rotation or limit-evasion marketing appears anywhere.

Ava is email-first with multichannel capability, so the LinkedIn surface is materially smaller than at the social-automation vendors at the bottom of this axis. Held at B and not higher: no conformance position is stated for any platform, and where LinkedIn outreach does form part of the multichannel claim the delivery mechanism is never described.

The training boundary is unaddressed and the product design makes it pressing. Ava is described as learning from campaign performance and optimising for pipeline impact across a customer base, and nothing published states whether that optimisation is confined to the contributing tenant or pooled - which for a shared agent operating in the same target markets for competing customers is the question a buyer should ask first.

Also absent: any retention schedule for prospect research, generated copy or reply conversations; any encryption specification at rest or in transit; any sub-processor disclosure; and any statement on what the agent does with data pulled from a connected CRM.

The cleanest statement of the article 50 problem in the index: loud about being AI to the market, silent about it to the person receiving the email. Artisan spent a reported billion impressions telling the world its product is an AI employee, and publishes nothing on whether Ava identifies herself as artificial to the prospects she writes to.

The exposure is sharper than at vendors that only send: Ava handles replies autonomously, so a recipient can hold a two-way exchange with an agent carrying a human first name, which is squarely the reasonably-foreseeable-contact scenario the Article 50 guidance addresses and where the marking duty falls on the provider rather than the deploying customer.

No Article 50 position, no statement that generated messages carry any disclosure, no described mechanism by which the recipient could know. Bottom of the C band, alongside Alta, for silence compounded by human presentation.

Thin by strategy rather than by neglect, and the grade reflects the outcome either way. Artisan's positioning is consolidation - the pitch is that the buyer no longer needs separate data, sequencing and deliverability subscriptions because Ava bundles them - so the product is designed to replace the stack rather than connect to it.

CRM connection is implied by the workflow and is not documented: no named connector catalogue, no public API reference, no object-level sync description, no SSO or SCIM detail, and no MCP or agent-facing endpoint located. A platform that intends to be the stack still has to answer how it sits beside the systems of record a buyer will not replace, and this one does not.

Nothing published: no hosting region, no cloud provider, no residency option, no data centre location, and no statement on where prospect research, generated messages or reply conversations are processed and stored. Multi-tenant SaaS is implied by the delivery model and never stated. European buyers running autonomous outbound into their own markets have no basis on which to answer a residency question in their own audit.

A security page, trust centre, SOC 2, ISO 27001 or penetration test attestation did not surface in a targeted search across the vendor's public surface and third-party vetting coverage. Recorded carefully rather than concluded: Artisan reports roughly 250 enterprise customers, and enterprise procurement usually forces a SOC 2 audit, so a certification may well exist and simply not be discoverable.

That is itself the finding for a buyer - a security posture that cannot be found is a security posture that cannot be used in a review, and every other well-funded vendor in this index publishes at least a landing page. Weighed against the sensitivity of what the platform holds: connected mailboxes, CRM access and reply conversations across hundreds of enterprise accounts. Re-verify directly before quoting.

The most contradictory pricing evidence in the index, and the contradiction is the finding. Third-party sources published within roughly six months of each other report, for the same product: flat tiers of $0, $250 and $600 per month billed annually; a $280 per month entry point; around $500 per month; $1,000 to $2,500 per month; $1,500 to $2,000 per month with an annual contract; and $12,000 to $120,000+ per year.

One source calls Artisan unusually transparent for publishing flat tiers; another states the public pricing page presents customised plans with no tier names and no lead-capacity matrix. Artisan's own blog describes the model as usage-based, charging per action performed rather than per seat, and separately the structure is reported as splitting BDR seats from AE seats. Annual commitment is consistently reported. A buyer cannot budget from this, which is the established test.

Graded C rather than lower because at least one source reports published numbers on the vendor's page and the live page was not directly retrieved in this pass - the grade records irreducible uncertainty, not confirmed absence.

Annual commitment is consistently reported across independent reviews, with at least one explicitly flagging the risk of a twelve-month lock on still-maturing technology as the platform's main commercial hazard. Against that, no offboarding path is documented: no export format or mechanism, no post-termination data rights, no deletion timeline or confirmation, and no statement on what happens to the enriched contact records, prospect research, generated copy or reply conversations the agent accumulates.

The bundled-database question is unanswered and matters most here: a customer who leaves has been prospecting from Artisan's database rather than building their own, so what they can take with them is precisely the thing nobody has written down.

A genuine strength and a sharp contrast inside the same category. Ava ships email warm-up procedures and inbox placement testing as part of the product, running on Artisan's own sending infrastructure rather than the customer's connected mailbox, which puts Artisan in the small group - with Amplemarket - that treats deliverability as a product surface rather than an assumption. Set directly against Apollo's D on this axis in the same session: same buyer, same category, opposite answer.

Held below A because the documented surface stops at warm-up and placement: no domain health monitoring, no published bounce or complaint thresholds, no mailbox or domain rotation policy, no volume governance, and - the one that matters most for a fully autonomous agent - no stated response to reputation degradation when the sender is a machine running at 90%+ autonomy with no human watching the send rate.

The segment claim does not resolve. Independent reviews place Artisan as best for startups and SMBs without a full SDR team; the company reports roughly 250 ENTERPRISE clients; and a third review concludes the price band puts it out of reach for most smaller teams. Those three cannot all be right, and the vendor publishes no segment definitions, no employee-count or revenue bands, and no ideal-customer framing that would settle it.

Named customers are two (SaaStr, Quora), both US technology companies. No geographic coverage stated, no supported markets or languages enumerated, and no per-region data density for a database sold on hundreds of millions of records.