Pipl.ai

The model layer carries the personalisation half and the infrastructure half survives its removal intact, which is the more interesting half of this product. Strip the models and the buyer loses the campaign and sequence writer, one click text randomisation, personalised opening lines built from scraped context, reply sentiment classification and the reply agent.

What remains is the thing this vendor actually engineered: a private warmup network of stated size, sender and address rotation, provider matching, address verification, a unified inbox, scheduling and volume control. The credit table confirms the split by pricing personalisations as a separate consumable alongside verifications, scrapes and address lookups, so the intelligence is metered on top of a deliverability platform rather than being the platform.

A reply agent answers real prospects on the two upper tiers and no oversight machinery is described for it. Nothing states whether a human approves a generated reply before it sends, what the agent may or may not commit to on the customer's behalf, when it escalates, or whether its actions are logged.

Real administrative controls do exist and are published in the comparison table: user roles and administration from the middle tier upward, dedicated client workspaces, unlimited team members, and a global block list that suppresses addresses across the whole account.

Held at the middle band because those are access and suppression controls rather than constraints on the autonomous component, and the component that talks to strangers without supervision is the one this axis is asking about.

One genuine disclosure sits in the feature table and no model estate is described anywhere else. The upper two tiers offer the option to connect the customer's own provider key, which names a major commercial model provider by implication and lets a customer route generation through their own account. That is a real and unusually useful option, and it is the only place the model layer becomes visible.

For the default path, no base model, provider, family, version or hosting arrangement is stated, and the warmup network is described as running its own model to match tone and schedules with nothing said about what that model is either.

Genuine named attribution undermined by defects in the same block. Around fifteen testimonials carry a full name, a role and in many cases an employer logo, several with per customer metrics for reply rate, positive reply share and meetings booked, plus video testimonials and one named chief executive of an identified company. That is real, checkable evidence and better than most vendors this size publish.

Against it, the headline testimonial contradicts itself on the same page: the same named person, quoting the same client count of seventy three, appears once describing over a year of use and twenty million prospects contacted and once describing three months and one million, a fourfold difference in duration and a twentyfold difference in volume. The vendor's own name is misspelled twice inside that quote. And one testimonial praises content planning, audience engagement tracking and post scheduling, which is social media management copy at a cold email vendor.

The vendor answers this question directly and ships the enforcement mechanic that matters most. Both the European data protection regulation and the United States commercial mail statute are named, and the position taken under them is specific rather than a claim of compliance: every message is stated to include an automatic unsubscribe and sender identification details, which is platform level insertion rather than a template a customer might forget.

A global block list suppresses addresses account wide on every paid tier. The vendor states plainly that it does not store or sell personal lead data and that enrichment is triggered by the user, and it transfers responsibility explicitly by saying it is the customer's job to follow the laws of their region.

Off the top band because no consent requirement is stated anywhere, no prohibition on purchased or scraped lists appears, no complaint rate threshold or account review trigger is published, and the platform simultaneously ships text randomisation and sender rotation whose purpose is to make bulk mail read as individual mail.

No privacy policy, processing addendum or cookie policy was reached, so this row is held at the middle band and flagged rather than graded as an absence. One substantive statement is verified and worth recording in the vendor's favour: it states that it does not store or sell personal lead data and that all enrichment is user triggered, which if accurate means the vendor is a lookup service rather than a data holder and materially narrows the exposure.

Against that, the platform runs a browser extension scraping prospects from more than a hundred sources, enriches from dozens more, holds credentials for potentially hundreds of connected mailboxes per customer, and operates a warmup network stated at over three hundred and fifty thousand accounts, all of which needs covering somewhere.

The source count is published three times with three different numbers and no source is ever named. Enrichment is described as drawing on thirty five sources in the product navigation, eighty sources in three separate places, and the browser extension as scraping from more than a hundred, and none of those figures is reconciled or itemised. No database, partner, licence, coverage figure, accuracy rate or lawful basis appears for any of them.

The credit table denominates consumption in scrapes rather than records, which tells a buyer the collection method while telling them nothing about the right to collect. Credited in the vendor's favour: the statement that personal lead data is neither stored nor sold and that enrichment runs only when a user triggers it, which is a meaningful limitation if it holds.

The evasion stack here is the most extensive in this index and every element is published as a paid feature. The unit of account is itself the tell: credits are denominated in websites scraped and professional network profiles scraped, so the vendor prices extraction by the act. A browser extension scrapes prospects from more than a hundred sources.

Sending runs behind address rotation at the network layer, sender address rotation, rotating domains, provider matching between sender and recipient, sender randomisation by provider and domain and tag, and infrastructure randomisation. The sharpest item is the warmup network: more than three hundred and fifty thousand accounts that exchange mail with a customer's inboxes and, in the vendor's own words, mark messages as important and as not spam while simulating opens, replies and stars.

That is deliberate manipulation of the recipient providers' own classification systems using a coordinated account network, which those providers prohibit, and it is sold as the product's central capability rather than hinted at.

One real mechanism and no stated position. The upper two tiers let a customer connect their own model provider key, which is a genuine stewardship option because it moves generation onto the customer's own account and takes the vendor out of the retention and training question entirely for those who use it. Very few vendors in this index offer it.

Against that, nothing states what happens on the default path: no non use commitment, no retention arrangement, no statement about whether customer content, scraped prospect context or the contents of connected mailboxes contribute to any model, and no sub processor list. The exposure is wide, covering connected mailbox contents across potentially hundreds of inboxes, uploaded lead lists and the scraped context assembled about each prospect.

No position on article 50 of the European artificial intelligence regulation appears anywhere, and a reply agent on the upper tiers composes and sends responses to real people with nothing stating a machine wrote them. Text randomisation varies the wording of bulk messages so each arrives looking individually written, and sender address rotation varies who a recipient appears to hear from.

Held at the middle band rather than lower for a reason that genuinely counts: the vendor states that every message carries automatic unsubscribe and sender identification details, so the recipient is always told who is contacting them and always given a way out, which is more than most vendors in this band provide.

Note also that the elaborate simulation of human behaviour, marking messages important and not spam, is aimed at the recipient's mail provider rather than at the recipient, and is graded on the platform terms row where it belongs.

Better than the category norm at the programmatic layer and gated above the entry tier. The upper two plans carry interface access, webhooks, two workflow brokers, a messaging platform alert integration, one named system of record, one named agency platform, the option to supply the customer's own model provider key, and a protocol server the vendor describes as connecting directly to external assistants including named commercial ones.

A protocol server at this price point is unusual and is the strongest item here. Off the top band on three counts: everything above alerts is reserved for the middle tier and higher, so the entry plan has effectively no programmatic surface; two of the most commonly required systems of record are absent and reported by independent research as still on the roadmap; and no developer documentation, marketplace or integration directory was located.

Fragments of architecture are published and no residency position exists. The done for you provisioning tiers name a specific public cloud for the agency option, state that dedicated addresses are deployed, and commit to sending addresses located in the United States only, which is a genuine location statement even though it concerns outbound infrastructure rather than data storage.

The top software tier offers an isolated sending server and dedicated connector addresses, and the highest managed tier describes a multi domain multi region architecture. None of that answers where customer data, connected mailbox contents, scraped prospect records or the warmup network itself are stored, and no hosting provider, country, region or residency option is published for the platform.

No certification, audit, attestation, penetration test, control set, trust centre, status page or vulnerability route was located, and the row is held at the middle band and flagged rather than taken to the floor because the page footer was truncated before it rendered and badge images were beginning to appear, so an assurance surface elsewhere on the site cannot be ruled out.

What is confirmed is that the fully enumerated navigation and resources menus, which list every free tool, the glossary, the community and the training academy, contain no security, trust or compliance entry at all. The only security language located is a marketing description of warmup infrastructure as isolated and secure.

Set against credentials for potentially hundreds of connected mailboxes per customer plus a warmup network of over three hundred and fifty thousand accounts, confirming whether any assurance surface exists is the priority check on this vendor.

Four separate commercial ladders are published in full and a buyer can compute any of them. The software runs three tiers with real monthly figures, each carrying a slider that reprices the plan across four or five enrichment credit bands, so the buyer sees the exact rate at every volume rather than a single headline. Monthly and annual are both displayed side by side with the annual benefit expressed concretely as two months free.

Every allowance is published per tier covering monthly send ceilings, active leads in campaigns, leads stored, workspaces and credits, and extra workspaces carry their own rate. The decisive disclosure is the credit conversion table, which states exactly what a credit buys across five different actions at every tier, so a buyer can convert their intended workload into a plan.

Beyond the software, done for you mailbox provisioning is priced per inbox for two providers and per domain for a third, a standalone placement testing product runs three priced tiers, and a fully managed service carries a published monthly figure. A fourteen day trial takes no card. The only gap is that agency tier enrichment credits are quoted as custom.

Storage ceilings are published and no exit position is. The comparison table states how many leads may be stored in lists at each tier, rising from three hundred thousand to one million to unlimited, which tells a customer how much of their material the platform will hold. Programmatic routes out exist on the upper tiers through interface access, webhooks and a protocol server, which makes bulk retrieval technically possible for those customers.

Against that, no export function, file format, post termination retrieval window, deletion timeline or retention period was located anywhere, and the entry tier has no programmatic access at all, so the cheapest customers have no described route to their own records. Terms of service and any privacy policy were not reached and are flagged.

This vendor has more deliverability machinery than anything else in the index and publishes a claim no sender can honour. Real discipline first, because it is substantial and some of it runs against the vendor's own interest: sender authentication records are auto configured across all three standards, open tracking is off by default with the reason stated as protecting domain reputation, volume is throttled, a monitoring layer alerts before placement degrades, address verification is included, and a separate paid product tests inbox placement on a recurring schedule.

Turning off open tracking by default costs the vendor analytics and is the single most creditable choice here. Against that, the comparison table lists a ninety nine point seven percent deliverability guarantee as a feature on every tier including the free one, and no sender can guarantee delivery because the receiving infrastructure decides.

And the scaling model is rotation at every layer, covering sender addresses, network addresses, domains and provider matching, alongside text and volume randomisation, which is the position this index has consistently held to be the opposite of discipline rather than an expression of it.

The buyer range is stated plainly and the ladders back it up. The vendor says it serves solo founders, small teams, agencies and large sales organisations, and quantifies the span rather than gesturing at it, describing customers sending fifty messages a week at one end and managing five hundred inboxes at the other. Each software tier carries a written statement of who it suits.

The agency case is supported by real product decisions rather than a claim, with dedicated client workspaces, white labelling per client, an isolated sending server, unlimited team members and custom credits. Four distinct commercial ladders address four different buying postures, from self serve through done for you provisioning to a fully managed service. Off the top band because no region, country, language or industry coverage appears anywhere, no customer count is published, and there is no statement of who this product is wrong for.