Trigify

The product is named for artificial intelligence and the substance underneath it is collection. What the platform actually does is monitor public activity on a set of platforms, resolve an observed action to a person, enrich that person, and route the result, and none of those four steps requires a model.

Monitoring runs on profile watches and Boolean searches on stated schedules, resolution matches an account to an identity, enrichment appends company and contact fields, and routing pushes to connected systems on rules. Strip the model layer and a complete social monitoring and enrichment engine remains, which is what a buyer is paying for.

The model layer sits above it and is real but additive: an assistant that configures searches and workflows from plain language, agents that reason over a signal, draft copy and classify sentiment, and a score that ranks intent. The positioning is worth reading carefully rather than at face value, because the vendor describes itself as the signal layer for artificial intelligence agents, which is a claim about what consumes its output rather than about what produces it. Being built for agents is not the same as being one. Ask which capabilities stop working with the assistant and the agent nodes disabled, and what the intent score is computed from.

Two oversight mechanisms here are rare enough in this index to carry the grade, and both are product features rather than policies. The first is replay and backtesting: the vendor states that any workflow can be run against historical signals so that scoring is tuned before it ships, which lets an operator see what an automation would have done before it does it to anyone.

Almost nothing else graded in this index offers a way to evaluate an automation against real past data prior to deployment. The second is a published outputs table described as recording every action the platform has taken, queryable and exportable, which is a complete audit trail rather than a dashboard.

Alongside those, monitoring frequency is an explicit setting at twelve hourly, daily, weekly or monthly rather than a fixed cadence, and the plan tiers cap searches and workflows numerically. Held below the top band because the boundary is undescribed where it matters most.

Nothing states what the copy drafting agent may generate without review, whether signals and drafted output pass any approval gate before they push into a connected sending tool, what volume ceiling applies to a workflow, or what halts one already running. Ask whether generated copy is reviewed before it leaves the platform, and what stops a workflow mid run.

The interfaces to the model layer are documented in detail and the model layer itself is not described at all. What is published is unusually concrete for a company this size: a protocol server endpoint with example calls, an interface for searches, results, enrichment and usage, and a command line tool, so a technical buyer can see exactly what the agents can invoke. That documents the connection rather than the reasoning.

No model provider, family or version is named for the configuration assistant, the reasoning agent, the copy agent or the sentiment classifier, no model card exists, and no accuracy, precision or evaluation figure is published for any of them. The scoring is the consequential gap.

The platform assigns an intent score that determines which people are surfaced and routed, and nothing published describes what the score is computed from, how it is weighted, or how a user would know it had drifted, which matters more here than for a generative feature because the score decides who gets contacted.

A bring your own key option appears as a published plan feature, which suggests the customer may be able to supply their own model credentials, but the comparison matrix did not render in a way that reliably maps that row to a tier, so its availability and price are unestablished. Ask which models run the agents and the assistant, what the intent score is computed from, and on which plan a customer may use their own model key.

Three statistics carry the argument on the front page and none of them is attributed there. The vendor states an average cold email reply rate of zero point seven percent for 2026, that seventy three percent of business buying journeys begin on social platforms, and that a contact made in context is eleven times more responsive, and no source appears beside any of the three.

The vendor does attribute figures elsewhere, naming a professional network's sales report and a review platform's buyer study in its published guides, so the omission on the page that sells the product is a choice rather than a habit. Those attributed figures also differ from the front page ones, at seventy eight and eighty seven percent against seventy three, which a reader moving between the two surfaces could easily conflate.

Customer evidence is a logo strip repeating six names twice under the label of powering signals for teams, and it does not distinguish customers from integration partners, two of which appear elsewhere on the site as connected products. One entry in that strip renders as a placeholder image on both pages where it appears, which is recorded as observed rather than interpreted.

No measured outcome across a stated customer population and period was located, and independent reviewers report signal volume and low intent noise as the most common complaint, which is precisely the thing a measured precision figure would settle. Ask what proportion of delivered signals a customer acts on, measured across a stated cohort, and for the sources behind the three front page figures.

The platform does not send, and it supplies both halves of what makes a send possible. Contact details are appended by the product itself, with email and phone enrichment as named workflow steps, and the qualified person is then pushed into a connected sending or sequencing tool. So a person who liked a post can be resolved, given a verified email address and a phone number they never provided, and placed into an outbound sequence, with the platform performing every step up to the send.

Nothing published addresses the obligations that attach to that. No consent basis is stated for enriching an individual with contact details on the strength of a public post, no suppression or objection mechanism is described, nothing states whether a person who has asked one customer not to contact them is excluded when a second customer's workflow surfaces them, and no guidance addresses which obligations transfer to the customer who does the sending.

The jurisdictional exposure is real rather than notional, since a vendor operating from the United Kingdom monitoring public platforms worldwide is enriching people across regimes with different rules on contacting them. Ask what consent basis supports contact enrichment from public activity, what suppression exists, and what the vendor considers the customer responsible for.

One combined policies page, and a retention model that is priced rather than stated. The privacy policy and terms are published together as anchored sections of a single page which was not retrieved on this pass, so what they contain is unestablished rather than absent, and no processing agreement, subprocessor list, transfer mechanism or data protection contact was reachable from the pages that were read.

What the pricing page does establish is unusual and worth recording: retention is a plan feature. Search history runs seven days on the entry plan, twelve months on the middle plan and all time on the enterprise plan, and stored table data runs thirty days, ninety days and a custom period on the same ladder.

So how long the platform holds records about monitored individuals is determined by what the customer pays rather than by a stated policy, which is the second record in this project to gate a retention control commercially. The corpus makes this substantial: the platform holds identified individuals who have no relationship with the customer, their social activity history, an inferred intent score about them, and appended contact details.

Ask for the processing agreement and subprocessor list, what retention applies to a monitored person's records independent of the customer's plan, and how a person exercises rights over data held about them.

Provenance at the signal level is architectural here, and it is the strongest form this axis recognises. The vendor states that every signal is a named person with the original source post attached, and the product surfaces the triggering post alongside the record, so a user does not accept a score on trust but inspects the public artifact that produced it.

That is verifiable by the buyer rather than asserted by the vendor, it survives any disagreement about scoring method, and no comparable record in this index ships the underlying evidence with each row. The signal library is published by type, so what is collected is enumerated rather than described in the abstract. Two things hold it below the top band and both concern layers other than the signal.

The collection basis is unstated: nothing describes the legal footing on which public activity is captured, monitored on a schedule and retained, and the vendor does not name the principal platform it draws from anywhere in its copy. And the enrichment layer has no provenance at all, since email and phone appended to a person come from suppliers that are never identified, under arrangements never described, which is the part of the record a buyer will actually contact someone with. No indemnification position was located. Ask which providers supply contact enrichment and under what licence, and what legal basis supports the monitoring and retention of public activity.

Systematic collection from a professional network the vendor will not name, with the method undescribed. The published capabilities are specific about what is captured: profile monitors, Boolean searches, engagement capture, retrieval of the likes and comments on a post, and monitoring of top accounts on stated schedules, with history reaching back as far as all time on the upper plan.

Extracting who engaged with a post and resolving each of them to an identity is data extraction, which the category editorial names as one of the decisive questions, and nothing published states how it is performed or under whose terms. The vendor's own copy refers to business networks and professional networks throughout and does not name the platform at any point reached on this pass, which is a choice a reader should notice rather than a conformance position.

This sits above the bottom band on a distinction that matters. The vendor does not operate the buyer's own account, automate messaging or connection requests on their behalf, or market evasion, so the buyer's account is not the thing at risk, and an independent reviewer makes the same observation that external monitoring carries less account exposure than browser driven automation.

What the buyer inherits instead is continuity risk in a service whose principal input depends on access the platform controls. Ask how engagement data is collected and under whose terms, and what happens to the service if that access is withdrawn.

Model processing runs across the most sensitive material the platform holds and no stewardship position was located. The agents reason over signals, draft copy and classify sentiment, and the configuration assistant is described as building searches and workflows from plain language, so model processing touches monitored individuals' posts, the profiles resolved from them and the intent inferences drawn about them.

Nothing published states whether that material trains or tunes any model, whether one customer's signals or configurations inform models serving another, what retention applies to prompts and generated output as distinct from the stored signal records, or which providers process it. No governance document, evaluation record or red teaming artifact was located, and no security or trust route appears in the site footer, which offers only privacy and terms.

One published feature bears directly on this and could not be pinned down: a bring your own key row appears in the plan comparison, which would let a customer route model processing through their own provider account under their own data terms, and that is a genuine architectural control rather than a policy statement, but the matrix did not render in a way that maps the row to a tier. Ask whether monitored activity or generated output trains any model, whether a customer may supply their own model key and on which plan, and what retention applies to prompts.

The monitored person is never told, and the input is at least genuinely public, which is what separates this from the bottom of the scale. Every signal originates in an act the person chose to perform publicly, a post, a comment, a like, a role change announced on a professional network, which is materially different from resolving anonymous browsing that the person had no reason to think was attributable.

That distinction is architectural and it is why this sits mid band rather than lower. What follows the observation is where the gap opens. The person is resolved to a named identity, appended with an email address and a phone number they did not supply, scored for intent, and pushed toward a private channel, and nothing published states what they are told about how they came to be contacted or whether the message reaching them was drafted by a model.

The vendor's framing of the value, that a public interaction becomes a private conversation, describes exactly the transition that this axis exists to record. A copy drafting agent sits inside the workflow, and no disclosure position accompanies it. Ask what a contacted person is told about how they were identified, and whether copy generated by the platform is marked as generated when it reaches them.

An open interface surface that is unusually complete for a company of this size, and a connector list that is not. Three programmatic routes are published with working detail rather than a promise: an interface covering searches, results, enrichment and usage, a protocol server with its endpoint and example calls shown on the marketing page, and a command line tool for terminal workflows.

The vendor's stated position is that everything the product finds is available to external software, and the published calls bear that out, so a customer with engineering capacity can treat the platform as a data source rather than a destination. That is a genuine architectural openness and it is the opposite of the pattern this index usually records, where the interface is the gated upper tier feature.

Named connectors cover a sequencing tool, a customer record platform, a messaging tool, a documents tool and a spreadsheet, several of which are separately indexed here, plus webhooks. Held below the top band on breadth rather than depth: the named list is short, the remainder is described only as more, no integration directory was located, and independent reviewers report the absence of native connectors to the major sequencing platforms, which for a product whose output is meant to reach an outbound motion is the gap that matters. Ask for the full connector list and which sequencing platforms are supported natively rather than through webhooks.

Nothing on hosting, region, tenancy or recovery was reached on the routes taken, and the combined policies page that might address it was not retrieved on this pass, so this records what was establishable rather than a finding of absence. The question is squarely applicable and the reason is the corpus rather than the company's size.

The platform continuously collects and stores public activity by identified individuals across territories, appends contact details to them, retains that material for periods running to all time on the upper plan, and offers no statement about where any of it sits. A buyer in a regulated sector or with a residency obligation of their own cannot answer their regulator from anything published.

Circumstantial signals point to United Kingdom establishment through the brief and the founder's published presence, and none of that is a residency statement. Ask which provider and regions host the platform and the stored signal records, whether any regional processing option exists, and what the tenancy model and recovery objectives are.

No security route is offered anywhere a buyer would look on the pages retrieved. The site footer carries exactly two links of this kind, a privacy policy and terms of service, both anchors on one combined page, with no trust centre, no security page, no certification badge, no vulnerability disclosure route and no request path for a report. That is an observation about the published surface rather than an inference, because the footer was read directly on two separate pages.

It is a weaker position than the comparable young vendors graded earlier in this session, both of which at least operated a hosted trust portal even where it would not render to a machine. The combined policies page was not retrieved on this pass and may carry security terms, so nothing here asserts that no programme exists.

The stakes follow the corpus: continuously collected activity records about identified individuals who are not the customer's contacts, with appended email addresses and phone numbers, held for up to all time. Ask whether any attestation or certification is held or in progress, whether a penetration test has been performed, and for a security contact and a completed questionnaire before data about third parties is entrusted to the platform.

Two self serve tiers published in full with a real comparison matrix, an overage rate on the page, and one number missing that decides the bill. Published are the entry plan at forty dollars a month and the middle plan at one hundred and ninety nine, with a custom enterprise tier, alongside a comparison across five feature groups covering searches, workflows, history, storage, tracking, assistant usage, interface access, integrations, support and seats.

Monthly credit allowances are stated at four thousand, forty thousand and two hundred thousand or more, searches and workflows at twenty five, one hundred and one thousand, and seats at two, five and unlimited. The overage rate is published outright at one point two cents per credit, which most vendors in this index will not state at any price. A fourteen day trial requires no payment card, which removes the evaluation barrier entirely. What holds it below the top band is the burn rate.

The vendor states that one credit pool powers four distinct activities, signals, enrichments, engagement retrieval and assistant runs, and nowhere publishes what any one of those consumes. A buyer therefore knows exactly what a credit costs and cannot forecast how many they will use, which is the calculation that decides whether four thousand credits is a month or a week. The two vendors graded top band on this axis earlier in this project both published their per action consumption. Ask how many credits a signal, an enrichment, an engagement pull and an assistant run each consume.

Export is a product feature rather than a promise, which is the distinction this axis usually has to record in the other direction. The vendor publishes an outputs table described as holding every action the platform has taken, queryable and exportable, and the same records are reachable through a published interface and a command line tool covering searches, results and enrichment.

A customer can therefore extract their signal history and the actions taken on it continuously, using documented routes, without depending on the vendor's cooperation at the moment of termination, and that is architecture rather than a support commitment. Held below the top band on the terms around it.

No statement describes what happens at termination, what retention or deletion applies to a departing customer's stored signals, monitored account lists, workflows and enriched profiles, or over what notice period. The plan structure complicates it further, because stored table data is retained for thirty days on the entry plan and ninety on the middle one, so a customer who downgrades or lapses may lose history before they think to export it, and nothing published warns of that. Ask what is retained and for how long after cancellation, and whether workflows and monitored lists export in a form that can be rebuilt elsewhere.

The platform operates no sending infrastructure, so most of this axis does not bite, and the part that does is the quality of what it hands to the tools that do send. One relevant control is published: email enrichment is described as returning a verified address, which reduces the bounce exposure a customer inherits when a signal becomes a sequence. That is a real contribution to a downstream sender's reputation and it is worth crediting. Nothing else is addressed.

No statement describes how verification is performed or by whom, what happens when an address cannot be verified and whether the person is passed on regardless, whether role based or generic addresses are filtered before routing, or what volume of contacts a workflow may push into a connected sending tool in a given period.

The risk profile is specific to this product's shape: the recipients are people surfaced by an intent score rather than drawn from a list the customer built, so a customer's sending reputation becomes a function of the precision of somebody else's scoring, and independent reviewers report low intent volume as the most common complaint about that scoring. Ask how addresses are verified and by whom, what happens to unverifiable contacts, and whether a workflow can be rate limited before it pushes into a sending tool.

The plan ladder implies the market and the vendor never states it. Pricing from forty dollars with two seats, through a middle plan at five seats, to an unlimited enterprise tier maps a range from an individual operator to a larger team, and the interface, protocol server and command line surface point clearly at a technical buyer comfortable assembling their own stack. That is an inference a reader can draw, not a fit statement the vendor provides.

Nothing published names the company size, sector or motion the product suits, states a minimum viable deployment, or says who should not buy it. Coverage of the underlying surfaces is stated as eleven or more platforms and the signal library as growing monthly, with the specific platforms named in published writing rather than on the product pages, so a buyer whose market lives on one particular surface cannot confirm support from the pricing or product pages alone.

No localisation, non English monitoring capability or regional coverage statement was located, which matters for a monitoring product whose entire function is reading what people write. Ask which platforms are covered and in which languages the monitoring and scoring operate, and what team size and motion the middle plan assumes.