Empiraa Signal

Signal is sold as a stack replacement rather than as an artificial intelligence product, and the pitch survives removal of the model. The value proposition on both the homepage and the features page is consolidation and price: one system in place of a CRM, a scheduler, a proposal tool, a lead source and a deck builder, with a calculator that totals what those five cost separately.

Remove the assistant and a complete saleable product remains, being the pipeline, deals, records, decks, scheduling and reporting that the tier meters are denominated in. The assistant is real and it is present on every tier, metered at 500, 2,500 and 10,000 credits a month, which is why this sits at the top of the band rather than lower. The vendor states the relationship in its own words, that a traditional CRM stores data and Signal acts on it, which places the model on top of a system of record rather than in place of one.

Published a can and cannot table on the AI Transparency page, which is a rarer artefact than a policy sentence. The stated limits are that the assistant cannot take autonomous action outside the platform, cannot send emails or messages without instruction, cannot access data outside the workspace and cannot modify account settings or billing without approval, closing with the line that all generated output is presented for review and decisions remain with the customer.

The homepage puts the same commitment in a question, asking whether the assistant will take actions without approval and answering no. The approval posture is unconditional rather than configurable, which is the stronger form and matches the best position in this band.

Held off the top because no audit trail of agent actions is described, no guardrail withholds a violating output, no escalation by risk exists, and the approval promise is written about the assistant while a separately metered automation engine of 250 to 10,000 runs a month executes configured workflows the promise does not reach.

A dedicated AI Transparency page, dated April 2026 and carrying the legal entity and company number, names the infrastructure the assistant runs on: large language models on Google Cloud Vertex AI. It publishes the data path plainly, from browser to Empiraa servers to the inference platform over transport encryption and back, and links the provider's own data governance documentation for the buyer to check. Naming the provider at all puts this vendor in a small minority of the index.

Held off the top because Vertex is a platform hosting many model families rather than a model, so nothing here identifies which model, version or maker is answering, and because the product makes consequential judgements that carry no published accuracy: lead scoring, deal risk detection, stalled deal flagging and prospect enrichment. A system that tells a rep which deals to stop working is making an accuracy claim whether or not a number is attached to it.

Across the homepage, the platforms page, the Signal features page and the pricing page there are no named customers, no customer logos, no case studies and no outcome attributable to any identified buyer. The results slot carries three unattributed multipliers instead, being three times more deals closed, eighty percent less admin time and two thousand dollars a month saved, with no basis, sample, period or method.

Two calculators generate further figures for the reader, one returning a saving of 2,351 dollars a month and the homepage version returning ninety four percent cheaper, and the comparator in the homepage version folds two thousand dollars or more of admin and specialist time into what is presented as a software cost. The only independent review record located is a single review, and it is filed against the sibling strategy product rather than against Signal. This is a maintained site with a blog, a glossary and a video library, so the gap is not a production problem.

Signal sends sales sequences from the buyer's own connected mailbox and one real control is documented, being reply detection that stops a sequence when a prospect answers, which appears in the privacy policy rather than only in marketing. Beyond that, no sending regulation is named anywhere for the customer's outbound: no unsubscribe insertion, no consent requirement, no suppression list, no regional filtering and no duty placed on the buyer.

The consent language that does exist covers the vendor's own marketing to its own users, being email and mobile message consent with an opt out route, which is the vendor as sender and not the customer as sender. The sharpest version of the gap is jurisdictional.

This is an Australian company whose home statute, the Spam Act 2003, requires consent, accurate sender identification and a functional unsubscribe on commercial electronic messages, and the vendor sells outbound sequencing without naming it once.

A current policy, version 1.1 effective July 2026, naming the Australian legal entity, the hosting country, the collection categories and a breach response, with access and correction rights for everyone. The Google integration section is unusually specific for a vendor this size and is written about the product rather than about a website: scopes enumerated, sent copies written back to the buyer's own Sent folder, message metadata read to detect replies, tokens stored encrypted and deleted or invalidated on disconnect, and adherence to the Limited Use requirements stated.

What holds it here is the population the policy never mentions. The tiers sell reveal of up to 10,500 companies and 210,000 contacts a month, and those people appear nowhere in the notice: no source, no notification of the kind Australian Privacy Principle 5 contemplates, no removal route and no lookup.

Portability and deletion are also framed as extras for users in the European Economic Area in certain circumstances rather than as product functions, and there is no data protection officer, no sub processor list and no stated breach notification window.

The reveal meters are among the largest published on any tier card in this index, at up to 10,500 companies and 210,000 contacts a month on the top plan, and the whole of the published provenance for that data is the marketing verb.

Signal analyses an ideal customer profile and surfaces companies, and it maps stakeholders and buying signals, with no source, supplier, licence, collection method or refresh position stated on any page read, and no route by which a person in the database can see or remove their record.

Held at the bottom of the band rather than lower because no harvesting or evasion mechanic is marketed, nothing is described as taken from a platform that forbids it, and the sending path runs through the buyer's own consented mailbox rather than through vendor infrastructure. Where a data vendor in this index has earned a higher grade it has done so by naming source classes or by publishing a removal route, and this vendor does neither.

The exposure that dominates this axis elsewhere is structurally absent. There is no social automation, no browser extension, no account rotation, no proxying, no scraping described and no undetectability language anywhere on the site.

Sending runs through the buyer's own mailbox under the buyer's own provider relationship: the connection is made by consent, the vendor holds an encrypted token rather than a password, the buyer can disconnect in the product or revoke access in their own account permissions, and the vendor states that its use of information received from Google interfaces adheres to the Limited Use requirements. Held off the top on two openings.

No position is taken on bulk sender requirements at the mailbox providers, which is the conformance question for a product whose meters run to 25,000 activities a month. And because the acquisition method for the reveal database is unstated, whether any of that data came from a source with terms of its own is unknowable from the published record.

The cross tenant training question is answered directly and in detail on a dedicated page, which few vendors this size attempt. Customer data is stated not to be used to train, fine tune or improve the provider's models or any other models. Data sent for inference is stated not to be retained, shared or made accessible to the provider or any third party beyond that processing.

Processing is scoped to the customer's own workspace with no combination across customers, and no secondary use is claimed. The Google integration clause in the privacy policy repeats it for the most sensitive input, stating that data from those interfaces is not used for advertising, not sold and not used to train generalised models. What keeps it off the top of the band is that the same promise appears at three different widths on the same site.

The homepage answers the safety question by ruling out training of public models, the narrower claim. The transparency page rules out any model, the broad one. And the vendor's own rendering of the provider's commitment qualifies it, saying customer data submitted to the platform is not used to train that provider's foundation models by default.

Alongside that, the privacy policy is the later document at July 2026 and it reserves the right to retain anonymised user content indefinitely for internal product development, and no contractual instrument carrying the broad promise was located.

The correspondent is real in every respect this axis tests. Mail leaves the operator's own named mailbox, a copy lands in that operator's own Sent folder, and there is no agent persona, no invented human, no generated voice, no local presence dialling and no automated action on any social platform. A human approves before anything sends, which the vendor states without conditions, so a real person signs every message that reaches a prospect. Held here rather than higher for three reasons.

The prospect is contacted on the strength of a record they never supplied and cannot see. The drafting is done by a model and presented as the sender's own writing, with no disclosure position anywhere on the site. And Article 50 of the European artificial intelligence regulation, in force since August 2026, goes unmentioned by a vendor selling generated outbound correspondence into the markets it reaches.

Roughly nine integrations are reported, being Slack, Microsoft Teams, Zoom, Twilio, Mailchimp, Brevo, Meta Lead Ads, QuickBooks Online and Stripe, alongside Google and Microsoft sign in through a workflow engine, and the metered automation runs on every tier confirm an internal workflow layer exists.

That list comes from a third party directory, and the reason is worth recording: on the vendor's own features page the integration set is published as a single image file, so the connector list cannot be read by a machine, by a screen reader or by a model answering a buyer's question. For a product whose central claim is that it replaces five named tools, the catalogue of what it connects to is the load bearing evidence and it is a picture. Public interface documentation, webhooks, a developer surface and an agent endpoint were all absent from the pages read.

The privacy policy names the hosting country plainly, stating that information is hosted on cloud based servers located in Australia, with processing also occurring in other countries where partners and service providers sit. Naming a country at all puts this ahead of most of the band.

What holds it at the bottom is that the half of the product the vendor markets hardest has no published location: the AI Transparency page states processing runs in regions consistent with applicable data protection requirements, then asks the buyer to make contact to confirm specific regional settings. A residency answer that resolves in a sales conversation is not a published one, and it covers the inference path, which is exactly where a buyer with a residency constraint asks. No region choice is offered, no sub processor list is published, and the two stated locations are never reconciled with each other.

An enumerated control set is published, and it sits on the AI Transparency page rather than on a security page: encryption in transit, 256 bit encryption at rest, role based access control, logical isolation between customers, monitoring for unusual access patterns and internal audit logs, with encrypted token storage described separately in the privacy policy.

Against that, this vendor claims no certification of any kind, and no auditor, report, audit period, trust centre, status page, penetration test or vulnerability disclosure route was located across five pages read. The one attestation named on the site belongs to the infrastructure provider rather than to Empiraa, the page describing the inference platform as enterprise grade and compliant and listing the provider's own frameworks, which is the recurring catch in this index: a host's certificate is not the vendor's. The privacy policy's own security paragraph is generic and closes by disclaiming any guarantee.

Three tiers at 49, 149 and 299 dollars a month with the currency stated as Australian on the page, self serve signup on all three, a fourteen day trial with no card, plan changes at any time, and no quote only ceiling anywhere. The piece that earns the grade is what the pricing removes: users are unlimited on every tier including the entry one, so the variable that makes a sales system bill unpredictable is off the table and the published number is the number paid.

Around it, ten meters are quantified on every tier rather than ticked, being companies revealed, contacts revealed, record ceiling, active deals, pipelines, decks, activities, assistant credits, automation runs and storage. Two gaps are recorded and neither is enough to move it. The assistant credit is the meter governing the part of the product the marketing leads with, and the question defining it is a collapsed element that did not render on retrieval, so the unit itself was not read. And nothing states what happens when a meter is exhausted, whether an overage charge, a hard stop or a prompt to upgrade.

Governing terms are published and a retention position is stated, which keeps this above the bottom of the scale, but the position is written as a standard rather than as a duration: registration information is retained for a commercially reasonable time and for as long as there is a valid purpose, and anonymised user content may be retained indefinitely for internal product development.

Deletion and a structured electronic copy are framed as additional rights held by users in the European Economic Area in certain circumstances, so the Australian and United States buyers this vendor targets are left with access and correction. The concrete absence matters most here because this is a system of record: no export function is named on any page read, on a product whose top tier holds up to 100,000 records and 2,000 active deals. The best position in this category on this axis puts unmetered export on the free tier and says so on the pricing page.

The architecture is the good part and it is the same one that separates the better cold email vendors in this index from the worse: sequences send from the buyer's own connected mailbox under their own provider relationship, so the buyer builds and inherits their own sending reputation rather than borrowing shared infrastructure, and a sent copy is written back to their own Sent folder so the thread is real mail in a real mailbox.

One genuine control is documented, being reply detection that halts a sequence when a prospect answers. Everything else that constitutes sending discipline is missing: no warmup, no ramp schedule, no authentication guidance, no per mailbox rate, no bounce or complaint threshold, no blocklist monitoring, no placement testing, and no verification described for the contacts the platform itself reveals. Activity ceilings of 1,000, 5,000 and 25,000 a month are published as entitlements rather than as pacing.

The buyer is described consistently and in more than one place. Each tier card names who it is for, being small teams starting out, growing teams needing more capacity, and teams scaling activity across more users and contacts, and a dedicated solutions page addresses small and medium businesses directly.

The positioning is stated against named comparators rather than in the abstract, the stack calculator listing the specific products it expects to displace, and the vendor publishes addresses on two continents. Held off the top because there is no headcount band, no vertical list, no qualifying condition and no stated ceiling: nothing says at what size a buyer outgrows this, and the claim that it replaces the entire sales stack is made without limit. The strongest segment statements in this index name the point at which the buyer should leave.