Quantified AI

The cleanest removal test in this session, because the product is a simulated human counterpart and there is no version of that without the model. Take it away and there is no persona to converse with, no scoring across more than fourteen hundred behavioural dimensions, no path that adapts to a representative's tenure and gaps, no real time scoring of language against a cleared content library, no generation of scenarios and rubrics from approved material, and no readiness signal.

What remains is a content library and a recorder, which is the previous generation of this category and precisely what the vendor positions against. Six capabilities are named and all are stated to run on one engine.

Unusually well described for this category. Every practice and coaching session is scored in real time against the customer's own cleared content library, off label language triggers an instant fail rather than a warning, every session produces an audit ready record automatically, one audit trail is stated to span the whole platform, and audit trails on every model interaction are named separately.

Certification is gated on demonstrated behaviour rather than course completion, and the customer configures the playbook, the rubrics, the scoring model and the visibility. Regulatory review of authored content is described as a structured workflow step rather than a bottleneck at the end, which is a genuine approval gate.

Holding it off the top band, the edge case is unaddressed: field coaching includes live call analysis, and nothing states what the system may do during a real conversation, who reviews the guidance it gives, or what a representative can override. No equivalent of the content review gate is described for the coaching the model delivers in the field.

The architecture is described in detail and the model estate is not. Six capabilities are named, one personalisation engine is named beneath them, the model is described as fine tuned, private and purpose built for life sciences, private deployment is offered for sensitive use cases, and scoring runs across more than fourteen hundred behavioural dimensions in forty or more languages.

What is never stated on any page read is the base model, provider, family or version behind any of it, what the fine tuning corpus consisted of, or who hosts a private deployment. Describing a model as fine tuned and private tells a buyer about the deployment boundary rather than about the model itself, and in a regulated procurement the provenance of the base model is a question that gets asked.

Seventeen customer logos are named and several are among the largest pharmaceutical companies in the world, and seven testimonials carry a full name, a specific job title and a named employer, with headshots, across four different customers. A case studies library and a customers page sit in the navigation. That is stronger named attribution than almost anything graded in this session.

Holding it off the top band, every aggregate figure is unsourced: six times more practice, forty percent faster time to readiness, nineteen percent more good selling outcomes and a ninety seven percent mastery rate all appear without sample, period, denominator, control or method. One of them has also migrated.

A vendor product listing elsewhere states forty two percent faster ramp and a nineteen percent improvement in win rates, while the site now states forty percent faster time to readiness and nineteen percent more good selling outcomes. The same number has been reattached from a measurable commercial result to a softer and unfalsifiable one.

This axis sits oddly on a product that does not send, and the vendor nonetheless carries more compliance apparatus than most vendors that do. Published: real time scoring of every practice and coaching session against the customer's regulatory cleared library, an instant fail on off label language, guardrails described as aware of the relevant promotional oversight regime, compliance drift tracked by therapeutic area, market and indication, and an audit ready record on every session so that inspection preparation is continuous rather than a project.

A dedicated compliance readiness page exists. Holding it off the top band, this is compliance with promotional regulation of the customer's own messaging rather than with the contact and consent regimes this axis was written for, and no position on those is taken anywhere. Graded on what is knowable with the note carrying the context, per the convention for vendors in this category that do not send.

Residency is offered across three named regions covering Europe, the United States and Asia Pacific, which is a concrete answer where most vendors are silent, and identity runs through the customer's own provider via federated sign on and directory synchronisation rather than a separate credential store. Audit trails on every model interaction are stated. A privacy policy exists at its own route and was not opened, so this is flagged.

What holds it off the top band is the thing the pages read never address, and it is specific to this product: the platform records named employees speaking, repeatedly and at length, and scores them across more than fourteen hundred behavioural dimensions, which makes it a workforce measurement system as well as a training tool. No retention period, employee facing notice, processing addendum or works council position appears anywhere, at a vendor selling into large European employers.

Cleaner than most in this index because the working corpus is the customer's own. Simulations are authored from material the customer's regulatory process has already approved, personas are tuned to the customer's ideal customer profile, objections are described as drawn from the customer's real field conversations, and scoring runs against the customer's cleared library.

The vendor sells no contact data and compiles no third party records, so the provenance question that dominates this category largely does not arise. Holding it off the top band, one layer sits outside the customer's own material and is unsourced: the behavioural scoring model rests on a research lineage the company describes as computational social science applied since 2012, and neither the corpus behind the fourteen hundred dimensions nor any validation of those dimensions against commercial outcomes is published.

Little exposure exists by construction. Nothing sends, nothing scrapes, no account rotates, no browser extension ships and no professional network is touched. Integration runs through named enterprise connectors into systems of record, content management platforms, learning systems and human resources platforms, plus federated sign on and directory synchronisation, all of which are sanctioned routes rather than worked around.

The one area worth naming is content: the platform ingests approved promotional material out of regulated content management systems and generates derivative scenarios, persona variants and rubrics from it, and whether a generated derivative inherits the approved status of its source is answered procedurally by routing it back through review, which is the right answer. Off the top band because no conformance position is published for any connected platform.

The vendor takes a position and it is more precise than most: customer content does not train foundation models, private model deployment is available for the most sensitive use cases, residency spans three named regions, and audit trails are generated on every model interaction. Four concrete commitments where much of this index has none. The limit of the position is visible on the same site, and it is the finding worth carrying.

The platform is described as one adaptive engine learning from every representative practice and coaching session, while the training disclaimer is scoped specifically to foundation models. Both statements can be true simultaneously and together they leave the actual question open: whether one customer's practice sessions improve the engine that serves another.

The material at stake is not marketing copy but recordings of named employees rehearsing, scored across fourteen hundred dimensions. Off the top band on that gap, and because none of the four commitments is shown to be contractual on any page read.

Two distinct questions arise and the site answers one of them by construction. The person conversing with the machine is the customer's own representative, who knows the persona is a simulation, so the ordinary form of this axis does not apply.

The second question does apply and goes unaddressed: field coaching covers pre call preparation, live call analysis and post call debriefs, which means real conversations with healthcare professionals are being processed by a model, and nothing states whether the other party is told, how consent is obtained, or what is retained.

The vendor has published an argument that recording live conversations in this setting erodes trust and creates compliance exposure, which shows the question is well understood inside the company, and its own live call analysis is never distinguished from the practice that argument criticises. No position on the European marking obligation appears anywhere.

Named breadth across four distinct categories, which is more than almost anything graded in this session: two major systems of record plus a life sciences specific one, four learning platforms named individually, three approved content management systems named individually, a human resources platform, two named identity providers, a named data warehouse, and federated sign on with directory synchronisation stated explicitly for enterprise rollout.

Coaching activity, readiness scores, certification status and compliance posture are stated to synchronise back into those systems. Holding it off the top band, the programmatic surface is claimed rather than shown: an open interface is named with no documentation, reference, developer hub or marketplace anywhere, one link offering full integration details resolves to a dead anchor, and the integration logo grid renders with empty image sources.

Residency is stated across three named regions covering Europe, the United States and Asia Pacific, and private model deployment is offered for the most sensitive use cases, which answers both where data rests and where inference runs. Together that is a fuller deployment answer than any other vendor graded in this session.

Holding it off the top band, whether the customer selects the region or it is assigned is never stated, no hosting provider, data centre or infrastructure partner is named, and the dedicated security page where this detail would ordinarily sit was not opened and is flagged. No corporate address is published anywhere on the pages read, with a Texas telephone area code the only geographic signal for the company itself.

A named certification with its type stated, which the conventions treat as materially stronger than an unqualified assertion: compliance to the second type of the relevant service organisation standard, stated as covering the whole platform.

Around it sit private model deployment, residency controls, automatically generated audit trails on every model interaction, a dedicated security and trust page in the primary navigation, and a statement that security documentation is available through a trust centre.

Holding it off the top band, the report itself is not served, no audit period, scope or auditor is named, no penetration test, status page or vulnerability disclosure route was located, the trust centre is described rather than linked, and the security page was not opened and is flagged. One small inconsistency: the pricing page badge states the standard without its type while the platform page states the type.

A pricing page exists in the primary navigation, carries two named packages and a genuine capability by capability comparison showing exactly what each includes and excludes, and contains no number of any kind. Both packages read quoted for your team and program, and every route forward is a request. No unit, seat price, minimum, term, range or starting figure appears anywhere.

The vendor explains the position rather than hiding it, noting that a two hundred representative specialty launch and a three thousand representative primary care field force are not the same deployment, and confirms that capabilities can be added later without rebuilding scenarios or reconfiguring integrations. What pushes this to the bottom band is a defect on the pricing page itself.

Two of the frequently asked questions are answered with text belonging to different questions: the question asking whether there is a minimum number of users is answered with a definition of coaching, and the question asking what implementation includes is answered with an explanation of why the product is not a learning management system. Those two questions are precisely the two whose real answers would have let a buyer estimate a bill.

The structural position genuinely reduces the cost of leaving and none of it is portability. Identity runs through the customer's own provider, the content library originates in the customer's own approved materials, and readiness scores, certification status, coaching activity and compliance posture synchronise continuously into systems the customer already governs, so a meaningful part of the output already lives outside the platform.

Against that, no export function, format, post termination right, deletion timeline or retention period appears on any page read. The asset with no stated route out is the one the platform is built to accumulate: the behavioural record of every named employee across every session, scored on fourteen hundred dimensions and held as an audit ready archive. Terms and conditions unread and flagged.

The product sends nothing. There is no mailbox, no sequencer, no dialer and no outbound channel of any kind, so warmup, ramp, authentication, complaint handling and reputation management have no application here. This is the structural non applicability recorded when the enablement category was created, and the convention is to grade on what is knowable with the note carrying the context rather than to leave the row blank or inflate it. Nothing in the vendor's published material bears on sending discipline in either direction.

The fullest segment disclosure graded in this session. Five industries each carry a dedicated page, eight use cases each carry another, and the focus is stated rather than implied: built for life sciences and regulated industries, with pharmaceuticals first among them. Seventeen customers are named including several of the largest pharmaceutical companies in the world, plus an insurer and a lender, so the claimed focus is evidenced rather than asserted.

Scale is published at more than ten thousand users, language coverage at forty or more with a stated fluency level, and geography at three regions through the residency answer. The boundary is drawn explicitly and more than once: the product is not a learning management system, it is described as the layer between learning and selling, it is stated to be additive rather than a replacement, and most teams are expected to run both. Naming what a product is not, to a buyer who already owns the adjacent thing, is the disclosure this band asks for.