Warmly

The differentiating half of the product is model dependent and the reliable half is not. Company level identification runs on reverse network address lookup enriched against third party databases, which is deterministic matching and would work with every model removed.

Everything the vendor sells above that is not: person level resolution across cookie, pixel and identity graph signals is probabilistic inference, the chat agents generate conversation, the autopilot tier performs goal setting, qualification and decision making, and the entry product is named for the AI that powers it. Strip the models and what remains is a reverse lookup tool, which is a materially smaller product sold at a fraction of this price.

Held below the top band precisely because that remaining piece is the part that works: the company level match is reported at roughly sixty five percent against roughly fifteen percent for the person level layer, so the deterministic component carries most of the delivered value while the modelled component carries the positioning. Ask what the current company and person level match rates are and how each is measured.

A named autonomy control with a human in the loop setting, which is more than most vendors running autonomous agents publish. Independent review reporting describes the outbound and inbound agents as running autonomously within configured guardrails and states they can be set to require human involvement instead, so the operator chooses the autonomy level rather than inheriting it.

That is a real control and it is the specific thing absent from several comparable records in this index, where agents act and no setting is described. The architecture also limits blast radius in one respect, since the agents engage visitors who are already on the customer's site rather than initiating contact into a cold population.

Held below the top band because the guardrails themselves are not described anywhere: no published statement covers what an agent may commit to in a chat conversation, what volume ceiling applies to automated outreach, what escalates to a person, or what halts activity already in flight. The top product tier is sold on removing human involvement, which raises the value of documenting exactly where the boundary sits. Ask what the configured guardrails actually constrain, and what stops an agent mid conversation or mid sequence.

Agents are named by function and nothing beneath them is disclosed. The product line identifies distinct agents for chat, qualification, outbound and total addressable market work, and describes capabilities including goal setting, decision making, slide generation and auto learning that improves performance over time.

That last claim is the one most in need of disclosure and gets none: a system described as learning from its own conversations invites questions about what it learns from, whether that crosses customers, and how a buyer would know if it drifted. No model provider, family or version is named, no model card exists, and no evaluation or error rate is published for the chat agents, the qualification logic or the identity resolution.

The credit unit that meters model consumption is itself undefined. Ask which models power the agents, what the auto learning mechanism trains on, and what evaluation covers agent output before it reaches a visitor.

The headline promise is three times more qualified pipeline, and the results offered behind it are customer quotes rather than measurements. A demand marketing manager at Orum says the team cut its acquisition cost by about half, and a chief executive reports nine opportunities in two weeks; neither figure comes with a baseline, a population or a period.

The data quality page gives 95% deliverability on verified emails and 137 million sessions processed a year, and describes the scoring factors that raise or lower confidence in a match. What a visitor identification product is bought on, the share of visitors resolved to a person and to a company, carries no figure on Warmly's own pages. A 4.8 rating from more than 200 reviews and a row of review site badges sit where measured accuracy would go.

Automated outreach is triggered by identification rather than by consent, and no suppression apparatus is described. The platform resolves a visitor to a named individual and can queue email and social outreach and retargeting against them within the same session, so the person moves from anonymous browsing to being contacted without any intervening action of their own.

Nothing published addresses the obligations that attach: no statement on consent basis for contacting an identified visitor, no suppression or opt out mechanism, no description of whether an objection recorded anywhere prevents re engagement, and no guidance on what transfers to the customer.

The jurisdictional question is live rather than theoretical, since the controller entity and processing are United States based while the platform identifies visitors from anywhere, and independent analysis notes that whether person level deanonymization of European visitors is lawful for a given use case is a question for counsel rather than a feature list. Ask what suppression mechanism exists, what consent basis the vendor asserts for contacting identified visitors, and what its stated position is on European traffic.

A jurisdictional notice exists and the instruments behind it were not reachable. The privacy policy names a United States entity as controller for European data protection purposes, states that data may be processed and stored in the United States, and carries a dedicated notice for the European Economic Area, the United Kingdom and Switzerland, so the vendor has recognised that European visitors will be identified and has written something addressing them.

What could not be established on the routes taken is any of the substance: no processing agreement, subprocessor list, transfer mechanism, retention period or data protection officer was located, and no privacy certification is claimed.

The stakes are set by what the product does rather than by its size, since person level identification of individuals who never submitted their details is among the more sensitive operations in this index, and the resulting profiles are enriched from third party providers and retained. Ask for the processing agreement, the subprocessor list covering the identity graph providers, the transfer mechanism for European visitor data, and the retention period for resolved profiles.

The enrichment layer names its suppliers and the identification layer does not. Independent reporting identifies four named data providers supplying firmographic and intent enrichment, two of which are separately indexed here, so a buyer can trace part of the corpus to known vendors and evaluate them on their own terms. That is real and worth crediting. The gap sits at the layer that actually distinguishes this product.

Person level resolution is described as combining cookie matching, email pixel tracking and third party identity graphs, and no identity graph provider is named, no collection basis is stated for the pixel and cookie signals, and no account is given of how a graph came to associate a browsing session with a named individual. That is the most privacy sensitive input in the system and the least documented. No indemnification position was located. Ask which identity graph providers supply person level resolution, what the email pixel mechanism tracks and under what basis, and whether provenance is indemnified.

Outreach and identification both touch third party platforms and neither method is described. Retargeting and automated outreach are stated to run across email, a professional social network and advertising channels, and person level identification depends on cookie matching, pixel tracking and identity graphs assembled from sources the vendor does not name.

Nothing published states what conformance position governs the social channel activity, whether it runs through that platform's sanctioned advertising and messaging interfaces or through other means, and no named platform commitment was located.

The identification layer carries a second and less common exposure, since it depends on third party cookies and pixel based tracking whose availability is set by browser vendors rather than by this vendor, and nothing published describes what happens to person level match rates as those restrictions tighten further. The grading convention treats undescribed collection against social platforms cautiously by default, and that is where this sits. Ask how social channel outreach is executed and under whose terms, and what happens to person level identification as third party cookies are deprecated.

An auto learning claim sits on the surface with no stewardship position underneath it. The top product tier advertises chat performance that improves over time through automatic learning, which describes a system training on its own operation, and nothing published states what that mechanism consumes, whether one customer's conversations inform models serving another, whether visitor data enters training at all, or what retention applies to chat transcripts and resolved profiles.

No model provider is named so the supply chain is invisible, no artificial intelligence governance document, evaluation record or red teaming artifact was located, and no independently audited management standard is claimed. The corpus is sensitive in a specific way: it holds named individuals who never identified themselves, their browsing behaviour, enrichment appended from third parties, and transcripts of conversations they held with an agent. Ask what the auto learning mechanism trains on, whether visitor data or transcripts cross tenant boundaries, and what retention applies.

The product premise is reaching people who deliberately declined to identify themselves, and no disclosure position accompanies it. The vendor's own material frames the value plainly: most visitors leave without completing a form, the company never knew they were there, and deanonymization changes that. The method for the person level layer is cookie matching, email pixel tracking and third party identity graphs, none of which a visitor would recognise as happening.

Once resolved, that named individual may be engaged by an AI chat agent on the site, receive machine written follow up, be sent automatically generated demonstration slides, and enter outbound sequences across email and social channels, all triggered by browsing they had no reason to think was attributable to them. Nothing published states whether the chat agent identifies itself as automated, whether generated outreach is marked, or what a person is told about how they were identified.

The upper product tier is explicitly named for running without human involvement. This lands at the bottom of the published scale rather than mid band because the identification method, the automated engagement and the absence of any disclosure position compound rather than sit separately, and because the vendor's own framing treats the visitor's choice not to identify as an obstacle the product exists to overcome. Ask whether the chat agent discloses that it is not human, and what a person contacted this way is told about how their identity was determined.

Integration is the delivery mechanism for a product whose value is speed of reaction. Named routes cover customer record synchronisation, messaging alerts for real time notification, sales engagement platform connections, webhooks for arbitrary downstream action, and activation into advertising retargeting across multiple channels, which together let an identified visitor trigger action in whichever system the team already works in rather than requiring them to watch a dashboard.

Enrichment draws on several named external providers rather than a single source. Reported deployment time of two to four weeks covering onboarding, customer record integration and signal configuration is consistent with a genuine integration surface rather than a script and a login.

Held below the top band on verification and on gating: no developer documentation, interface reference or rate limits were reached on this pass, no integration inventory was counted, and customer record synchronisation sits inside the entry product while chat and autonomous execution require the higher tiers. Ask for the interface documentation and which integrations are included at the tier quoted.

United States processing stated, with no residency option and no architecture published. The privacy policy states that data may be processed and stored in the United States and names a United States entity as controller including for European data protection purposes, which is at least a clear answer about where data goes.

No hosting provider or region is named, no European or United Kingdom residency election is described, no tenancy model is stated, and no recovery time or recovery point objective is published. The absence carries more weight for this product than for most, because the data being moved to the United States includes named individuals resolved from European web traffic who never submitted their details, and a European buyer deploying the tracking script on a site serving European visitors is the party that must answer for that transfer. Ask whether any European processing or residency option exists, which provider and regions host the platform, and what the recovery objectives are.

No certification, attestation or trust surface was located on the routes taken this pass. No trust centre, no service organisation control report, no international information security certification, no penetration testing statement, no vulnerability disclosure route and no security page enumerating controls was reached.

This records what a buyer could establish before contacting sales rather than asserting that nothing is held, and for a venture funded company of this size some security programme very likely exists behind a sales conversation. The gap matters in proportion to what the platform holds, which is resolved identities of named individuals who never volunteered them, enrichment appended from several third party providers, browsing behaviour and conversation transcripts, all processed in a single jurisdiction.

A buyer whose own customers appear in that corpus inherits the vendor's posture without being able to inspect it. Ask which certifications or attestations are held, with audit periods and auditors, and whether a completed security questionnaire is available.

Real figures published in two billing cycles, undercut by a undefined billing unit and by the page contradicting its own arithmetic. The pricing page publishes three products with annual and quarterly figures: identification at ten thousand dollars a year or four thousand eight hundred and seventy five a quarter, chat at twenty thousand or six thousand five hundred, and the autopilot tier at thirty thousand or nine thousand seven hundred and fifty.

Two add ons carry published figures of ten thousand dollars a year each. Publishing both cycles is better than most vendors at this grade manage. Three things hold it here. The annual figures are prefixed as starting points rather than prices, every button routes to a sales conversation, and no self serve path exists, so the numbers are opening positions.

Every tier states a floor of ten thousand credits a month and the page never defines what consumes a credit, which is the unit that meters the entire product. And the billing switch carries a saving claim of thirty percent that does not match its own numbers: against four quarters, the chat and autopilot tiers save twenty three percent while the identification tier saves roughly forty nine percent.

The starting at prefix on the annual side may mean the two ladders describe different configurations, which would explain the gap without excusing the label. Ask what a credit is consumed by, the overage rate, and which saving figure is correct.

Nothing published addresses leaving, and independent analysis identifies the switching cost as material. No statement of export scope or format was located, nothing describes whether resolved visitor history, enriched profiles, conversation transcripts and orchestration rules leave with a departing customer, and no retention period, deletion timeline or notice term appears.

Independent review reporting describes the platform as a strategic purchase where migrating away requires meaningful re-architecture of revenue workflows after deep customer record and signal integration, which is a switching cost assessment from outside the vendor and consistent with a product whose value comes from triggering action across several connected systems. Two specific questions follow and neither is answered.

Resolved identities were assembled partly from third party identity graphs and enrichment providers, so what a customer may lawfully retain and use after termination is a licensing question rather than an export one. And conversation transcripts are a record of exchanges with the customer's own prospects. Ask what exports and in what format, what licence attaches to resolved profiles after termination, and what the deletion timeline is.

The platform sends automated outreach and the sending discipline is reported as thin. Automated email follow up and outbound sequences are core to the middle and upper tiers, so this axis applies in full rather than by scope. Independent review reporting states directly that email deliverability management is basic, that the platform lacks the managed mailboxes and dedicated warmup that specialist sending products provide, and that teams running heavy outbound will need a complementary tool.

That is a specific capability finding from outside the vendor rather than an absence of information. Nothing published from the vendor addresses which infrastructure sends, whether mail leaves from the customer's own connected mailboxes or vendor systems, who configures authentication, what bounce or complaint thresholds trigger intervention, or whether warmup exists for a new sending identity.

The risk profile is raised by the input: recipients are resolved from browsing behaviour and enriched addresses rather than from a verified list. Ask which infrastructure sends, whether addresses are validated before automated sending, and what bounce handling applies.

Warmly sells to B2B revenue teams and keeps separate solution pages for startups, mid market and enterprise companies, and for sales leaders, marketing, revenue operations and SDR teams. Price sets the practical floor: identification starts at $10,000 a year, or $4,875 a quarter for a team that wants to try it before an annual term, with no free plan and no checkout.

Reach runs beyond the customer's own website, since the signal layers add social activity, Bombora research intent, job changes, hiring and funding events for accounts that have never visited. The customers quoted on the site, Orum among them, are B2B software companies, and the positioning still spans startup to enterprise without naming the team size, traffic level or region where the product works best.