Salesforge

The vendor's own pricing architecture settles the removal test before any analysis is needed. The pricing page presents a human path and an agent path as separate purchases and tells the buyer to choose either or combine both, and the vendor states in its own answers that the platform runs human led campaigns without the agent and that the agent can run without the platform's campaigns.

A company that sells the product with the model layer removed, at a published price, has demonstrated that a complete product remains after removal. What remains is substantial: sequencing across email and social channels, a unified reply inbox, mailbox and address rotation, warmup, validation, workspaces and integrations, sold from forty eight dollars a month.

The model layer above it is metered separately as personalisation credits, at one credit per generated opening, two in the all source mode and three for a generated reply draft, and a capability billed by consumption apart from the subscription is detachable from it by construction. The agent adds autonomous prospecting and reply handling on top and is a genuine model dependency, but it is an addition to a platform rather than the platform. Founded 2022 as a cold outreach product.

This grade will read oddly against a homepage that leads with the agent, and it is the grade the convention gives. Ask what the platform does for a customer who never buys the agent, and which capabilities consume personalisation credits.

A named approval mode, published rather than reported, on a product that otherwise sends without a person. The agent is sold with two operating modes the vendor names as autonomous and as an approval mode in which a person reviews output before it goes out, so the operator chooses the autonomy level rather than inheriting it.

Two supporting mechanisms are described: a knowledge base the agent draws on when generating replies, which constrains what it can assert to material the customer supplied, and sentiment analysis that classifies each reply thread and tags it, which gives an operator a triage surface over conversations the agent is holding. Grounding replies in a customer controlled corpus is a real constraint rather than a policy statement. Held below the top band on three absences.

Nothing published states what the agent may commit to in a conversation, what volume ceiling applies to it, or what halts a sequence already running beyond the ordinary stop on reply. The approval mode's scope is not described, so whether it applies per campaign or across an account is unknown. And the agent tier is reachable only through a booked demonstration, so its configuration surface cannot be verified from the published site at all. Ask what the agent is permitted to assert beyond the knowledge base, whether approval mode is set per campaign, and what stops it mid sequence.

Unusually specific about what generation costs and silent about what performs it. The vendor publishes the mechanics of its own metering in a way almost nothing else in this index does: one credit generates an opening from a single data source about a contact, two credits generate one from every available source in a mode the vendor names, three credits generate a reply draft, and regeneration is free.

It publishes twenty one named languages for generation, which is a concrete capability boundary rather than a claim of global coverage. That is disclosure of the economics and the scope. What it does not disclose is the system. No model provider, family or version is named for message generation, reply drafting, sentiment classification or the agent, no model card exists, and no evaluation, accuracy or quality measurement is published for any of them.

The platform offers a protocol connector and command line interface for connecting external assistants, which discloses what the product connects to rather than what runs inside it. The gap matters because message quality is the entire value proposition and the customer's own domain reputation carries the output. Ask which models generate messages and replies, what evaluation covers output before it sends, and whether the customer can inspect the instructions used.

The most rigorous evidence about this product was produced by somebody else. An independent reviewer reports running the agent across three accounts over ninety days and publishes the full arithmetic: twelve thousand four hundred emails sent, two hundred and eighty seven replies, forty one meetings booked, with a stated conclusion that the agent covered roughly sixty percent of the role and that reply handling, meeting preparation and account research still required a person.

That is a denominator, a period and a population, which is what this axis asks for, and it did not come from the vendor. What the vendor publishes alongside it is testimonial. Two named customers appear on the pricing page and both praise deliverability and ease of use without a number attached, a case studies section exists, and a forecasting tool invites a buyer to model their own expected outcomes rather than showing what existing customers achieved.

No study measures reply rate, meeting rate or deliverability across a stated customer population and period. The vendor also publishes a large public directory of competing tools and agencies, which is a content asset rather than evidence about its own performance and should not be read as one. Ask for measured reply and meeting rates across a stated cohort and period, and for the deliverability outcome its testimonials describe in numbers.

High volume cold sending into multiple jurisdictions with no compliance position published anywhere reached. The platform sends cold email and social messages at volumes the plans express directly, at fifty thousand emails a month on the second tier before purchased overage, to contacts drawn from a supplied database of people who have no relationship with the customer.

The category editorial names the decisive question as how much of the commercial email mechanics the platform enforces against what it leaves to the customer, and nothing published answers it: no statement on opt out capture and honouring across email and social channels, no sender identity or physical address requirement, no consent basis for the supplied contacts, no jurisdictional screening, and no suppression apparatus described beyond email validation.

Validation confirms an address exists rather than that it may lawfully be contacted, and the two are frequently conflated in this category. The exposure is heightened rather than reduced by the vendor's European establishment, since a European processor sending cold outreach into European jurisdictions is operating where consent requirements are strictest. Ask what unsubscribe mechanism is enforced across both channels, what consent basis supports the supplied contacts, and which party the vendor treats as controller for prospect data.

The processing addendum page is the finding here, and what it contains is a description of a document rather than the document. It states that an addendum is offered to European controllers, that it supplements the terms and privacy policy, and that a customer must request it by email and sign it electronically. No terms appear on the page: no subprocessor list, no transfer mechanism, no retention period, no security schedule, no data protection contact.

The page carries a last updated date of April 2023, so it has not been revised across the launch of an autonomous agent and five sibling infrastructure products. The vendor also states that it will not sign a customer's own addendum and will not make individual changes to its own, and gives its reason directly, that it is a small team without a legal team on staff and that counsel would be cost prohibitive.

That candour is genuine and is worth recording as candour; it is also a firm statement that the contractual terms are not negotiable. Set against that, legal pages are localised into four languages and a privacy policy and terms are published. The corpus at stake is prospect personal data at volume, sourced by the vendor, processed for generation, and stored with reply content. Ask for the addendum itself before purchase, with the subprocessor list, the transfer mechanism, the retention period and the security schedule, and confirm whether it has been revised since 2023.

The vendor is a data supplier as well as a data user, and says nothing about where any of it originates. Contact data reaches the customer by three published routes: a search product the vendor describes as offering more than five hundred million verified business contacts, a feature that finds email addresses from professional network profiles, and an interface sold for retrieving verified addresses at scale, which means the vendor sells the data onward as a product in its own right.

Naming those routes is not the same as naming a source. No provider, aggregator or licensing arrangement is identified for the database, nothing states whether records are collected, purchased or licensed, no lawful basis is stated for holding personal data on individuals unknown to either party, no indemnification position was located, and the mechanism by which an address is derived from a profile page is not described.

Verified is used throughout to describe deliverability rather than provenance, and a buyer should not read it as the latter. Selling the data through an interface raises the stakes rather than lowering them, because a customer using it becomes a downstream recipient of a corpus whose origin neither party can establish. Ask where the database originates, on what basis the records were collected, how addresses are derived from profile pages, and whether provenance is indemnified.

Unlimited automated identities on a professional network, contact extraction from its profile pages, and no statement of method anywhere. The second tier connects unlimited profiles to the platform for ninety six dollars a month, and social actions covering profile views, connection requests and messages are sold as bulk credits with a stated expectation of four hundred to five hundred actions per profile per month.

Alongside the messaging sits extraction: a browser extension and a named feature derive email addresses from profile pages, and the results are sold onward through an interface. The vendor offers a whitelabel programme and agency positioning for running client outreach at scale, so operating many identities across many accounts is a marketed use rather than an edge case.

Nothing published describes how any of it executes, whether through sanctioned interfaces or otherwise, and no conformance position appears. What does appear in the footer is a disclaimer that the vendor is neither affiliated with nor endorsed by the network, which acknowledges the relationship without addressing the terms.

The category editorial states that a vendor which will not say how it operates has answered this axis by construction, and unlimited identities plus extraction is the fullest version of that. Two records hold this band for a different reason, marketing evasion itself through behaviour simulation and published guidance on stepping around stated limits, and nothing of that kind appears here. Ask how social actions and address extraction are executed and under whose terms, and who bears the loss when an account is restricted.

No stewardship position was reached, on a platform where the model layer touches the most sensitive material it holds. Nothing published states whether prospect data, generated messages, reply content or the agent's accumulated learning train or tune any model, whether one customer's material informs models serving another, what retention applies to prompts and generated output, or which providers process it.

No governance document, evaluation record or red teaming artifact was located, and the trust portal on the vendor's own subdomain was not retrieved on this pass, so material may sit behind it that this pass did not see. Two published details make the questions concrete rather than generic. The agent is sold with long term learning as a named capability, which describes accumulation across time without stating across what boundary.

And the warmup service is described as leveraging machine learning and generative technology to keep mailboxes warm, which means the traffic circulating to build sending reputation is itself machine generated, a design choice with its own stewardship questions about what that generated mail contains and where it goes. Ask whether prospect data, messages and replies train any model, what the agent's long term learning retains and for how long, and whether anything crosses a customer boundary.

Machine written outreach sent under the customer's identity, with tone selected as a setting, and nothing anywhere addressing what the recipient is told. The agent researches a contact, composes the message and sends it from the customer's own mailboxes and connected social profiles, then handles the reply, so a prospect can hold an entire exchange without a person having written a word of it.

The vendor's own public demonstration makes the design intent explicit: a visitor supplies an address and receives a generated email from the agent with the tone chosen from a list of seven options including playful, hilarious and appreciative. Emotional register presented as a configuration setting is the clearest available statement that the output is engineered to read as personal correspondence.

The agent is named after the founder and referred to as a person throughout the site, and the platform is sold on running outreach without the headcount that would otherwise write it. Nothing published states that a recipient is told a message was machine generated, that the agent identifies itself as automated when asked, or that generated replies carry any marking, and the disclosure obligation under Article 50 of the European artificial intelligence regulation, in force since 2 August 2026, is unmentioned by a vendor established in Europe. Ask what a recipient is told, what the agent replies when asked directly whether it is a person, and whether any generated message is marked.

A wide surface for a company this size, assembled outward rather than deep into one platform. Published routes include an application interface, a protocol connector and command line interface for driving the platform from external assistants, a browser extension, applications on both mobile stores, an integrations directory, and template libraries for three separate automation platforms, alongside a public roadmap and feature request portal, a help desk, an academy and a certified partner programme.

A whitelabel programme lets a reseller run the platform under their own brand, which is an unusual depth of ecosystem commitment and creates a partner channel rather than only a connector list. Two things hold it below the top band.

The interface, the protocol connector and the integrations are all conditional on the second tier, so the entry plan is a closed system, and no integration inventory was enumerated on the pages retrieved, so breadth is asserted through a directory rather than counted.

And the architecture pushes integration work onto the buyer in a specific way: because each product in the family carries its own subscription and its own surface, a full deployment means assembling several products rather than configuring one. Ask which integrations exist by name and which require the upper tier, and how the products in the family share data between them.

Nothing on hosting, region, tenancy or recovery was established on the routes taken, and the question is squarely applicable. No provider, region, residency election, tenancy model or recovery objective appears on the pages retrieved, and the trust portal that would ordinarily carry this material was not retrieved on this pass, so this records what a machine readable pass could establish rather than a finding of absence.

Circumstantial signals point to European operations, including a European scheduling instance, legal pages in four European languages and a processing addendum framed around European controllers, but none of that is a residency statement and it is recorded here as inference rather than fact.

The footprint is wider than the subscription suggests, because the vendor provisions and operates sending domains and mailboxes, holds a large contact database, stores generated messages and the replies prospects write, and spreads all of it across several separately subscribed products whose hosting may not be common. Ask which provider and regions host each product in the family, whether the data stays in one jurisdiction, and what the tenancy model and recovery objectives are.

Two compliance badges in the footer, both linking to a trust portal that was not retrieved on this pass. One badge refers to a service organization control audit and the other to an accessibility standard, and both point at a trust subdomain the vendor operates. A portal of that kind ordinarily carries certification status and a document request route, so material very likely sits behind it, and nothing here should be read as asserting the vendor holds no certification.

What can be said is what a buyer establishes from the published site alone, which is a badge and a link: no audit period, no audit date, no auditor, no scope, no penetration testing statement, and no enumerated control description on any page retrieved. A published responsible disclosure route was not located either.

This index has consistently held that pattern to the middle band, a certification asserted with nothing independently checkable before a sales conversation, and the same grade was applied to a far larger enablement vendor on the same reasoning earlier in this project. Ask for the report with its audit period, scope and auditor, whether the audit covers the sibling infrastructure products or only the core platform, and for a penetration testing summary.

A buyer can compute their own bill on this page, including the parts that usually decide it. Both self serve tiers publish monthly and annual rates with full feature lists, and the annual claim of two months free is arithmetically exact against twelve monthly payments on each.

Underneath sits a live calculator across four separately metered dimensions, active contacts and emails, social action credits, personalisation credits and validation credits, so a buyer models overage before purchase rather than discovering it after. The metering definitions are published rather than implied, including what consumes a credit in each dimension and the expected social action volume per connected profile per month.

The agent tier publishes its rate with its billing cycle, its included contact volume and, unusually, its overage rate at twenty five cents per contact beyond two thousand, and a capacity add on carries its own published price. A consulting engagement is priced. The trial is specific down to the credits it includes.

Most notably the page carries an estimator that assembles the whole family and shows each sibling product's cost, which is the vendor volunteering that its own headline price is not the real bill, a disclosure most vendors make only under questioning. Three qualifications belong on the record without moving the grade. The agent tier cannot be bought without a booked demonstration.

The agent's rate appears as five hundred and ninety nine dollars quarterly on its own page and four hundred and ninety nine annually here, which reconciles across billing cycles but reads as two prices to anyone skimming. And in the state the estimator returned, its itemised product costs did not sum to the total it displayed.

The commercial exit is clean and published, and the data exit is not addressed at all. Terms are stated plainly: no contract commitment on the self serve tiers, monthly or annual by choice, changes and upgrades charged on a pro rata basis, cancellation at the customer's convenience, and invoices issued and editable in the application. That is better than most and it removes the lock in question at the contract level. Nothing published covers what leaves with a departing customer.

No export scope or format is stated for contact records, sequences, generated message history, reply threads in the unified inbox, personas or agent configuration, and no post termination retention or deletion timeline appears. Two structural points sharpen it. Purchased credits are consumed against a subscription and the relationship between an unused balance and cancellation is not described.

And because the sending infrastructure is bought through sibling products under separate subscriptions, the domains and warmed mailboxes an outbound programme depends on are held under a different contract from the platform, so a customer leaving one may or may not retain the other, and warmed sending reputation takes weeks to rebuild elsewhere. Ask what exports and in what format, who holds registration of the domains and mailboxes, and what happens to unused credits at cancellation.

One of the fuller control sets in this index, sitting beside two practices that keep it off the top band, which is the same shape as the record that leads this axis. Published and included: unlimited warmup through a sibling deliverability product, automatic mailbox rotation, automatic address rotation, matching of sender provider to recipient provider, validation credits metered per contact, bounce handling, and a reputation score per mailbox.

The agent's infrastructure publishes a per mailbox sending rate of roughly fifteen a day across twenty mailboxes, which is conservative and stated rather than left to the buyer. Around the product sits a free public diagnostic toolkit including inbox placement testing, a deliverability check, a blacklist checker, a provider identifier and a spam copy scanner, available without an account, which is deliverability infrastructure in its own right and the same thing that earned credit elsewhere on this axis.

Three points hold it below the top. The vendor publishes a tool for generating spintax, and copy variation whose purpose is to keep bulk sends from being recognised as bulk is filter evasion rather than sending discipline, which is exactly the distinction the leading record on this axis was graded against.

Warmup is described as leveraging machine learning and generative technology, meaning the reputation is built on generated traffic, and the vendor does not say whether subscriber mailboxes send to each other. And unlimited mailboxes with automatic rotation and no published complaint threshold or stopping rule is volume architecture pointing the other way. Ask whether warmup traffic circulates between customer mailboxes, and what bounce or complaint rate halts a campaign.

Coverage is expressed through structure rather than assertion, and one stated market does not survive the vendor's own contract page. The tier ladder runs from a single seat plan for solo operators to an unlimited user plan, dedicated solution pages address agencies, sales teams and growth teams separately, and a whitelabel programme lets an agency resell the platform under its own brand, which is a distinct market served with a distinct product rather than a positioning line.

International coverage is concrete rather than claimed: twenty one named languages for message generation, and the site and legal pages localised into three further European languages. A published academy, glossary and course material lower the entry barrier for the least experienced buyer the entry tier invites. The gap is at the top.

The pricing page states that the plans cater all the way up to the largest enterprises, while the vendor's own processing addendum page states that it will not sign a customer's agreement and will not modify its own because it has no legal team, and an enterprise procurement process cannot generally proceed on those terms. The two statements sit on the same site and are not reconciled. Ask what an enterprise deployment looks like given the stated contract position, and what the smallest viable deployment is in sending volume.