Catchlight

The score is the product and everything else exists to feed it or act on it. A predictive conversion model trained on a stated corpus of more than 170,000 client conversions carries the entire proposition, which the company states plainly as helping an adviser know which prospect to call next. Remove it and what remains is an enriched list with no ordering, which is a data product nobody is buying this for.

The enrichment layer is a genuine asset in its own right and it is upstream of the model rather than an alternative to it. Worth distinguishing from the deterministic scoring elsewhere in this pocket: this is a model learned from outcome data rather than financial mathematics applied to a household's numbers, so the objection that quantitative scoring is the wrong kind of model does not apply here.

A per record confidence signal is surfaced in the interface, which is the control this pocket almost universally lacks and the reason this sits above the norm. The product marks a lead where it holds high confidence and shows where attributes are missing and uncertainty is therefore significant, so a user can see which scores rest on a thin profile before acting on them.

Set against three other vendors added this session whose independent reviewers had to invent a manual verification habit because no such signal existed, publishing uncertainty at all is a meaningful choice. Held below the top grade because the signal reflects completeness of the underlying profile rather than the model's confidence in its own prediction, it is presented as a mark rather than a graded measure, and nothing describes an approval step, an audit record or what a firm should do with a low confidence score beyond noticing it.

The size of the training corpus is disclosed and the performance of the model is not, and only the second is what this axis measures. More than 170,000 conversions and over 100,000 adviser and client interactions is a specific and useful statement about what the model learned from, and continuous testing is referred to in passing.

Searched for any measure of how well the score actually predicts, whether an accuracy rate, a calibration statement, a discrimination measure, a holdout result, a comparison against no scoring at all, or a revalidation cadence, and located none. The data itself is described as institutional grade, which is an adjective standing where a figure belongs and is the same construction this index has flagged in three other vendors.

One named customer executive is quoted by full name and title at a named advisory firm, which is exactly the bar this grade describes. Around it sits scale evidence of mixed quality: trade reporting placed adoption at roughly 300 firms in late 2024, which is a third party figure but a journalist's rather than a measured survey, and the training corpus scale of more than 170,000 conversions is an input claim rather than an outcome.

Held below the top grade because the one quantified performance claim, an improvement in conversion of two to three times, is self reported, attached to no named firm and unaccompanied by any statement of what it was measured against. For a product whose entire value is a lift in conversion rate, that figure is the one that most needed a customer's name on it.

Searched for any statement on whether a customer firm's uploaded leads, its recorded outcomes or its adviser activity are used to train, tune or evaluate the shared model, and located none, and the product material points toward yes without saying it.

Scoring is described as getting smarter over time with adviser use, which in a single shared model means one firm's conversion results improve the predictions another firm receives, and that is either a substantial benefit or a substantial leak depending entirely on terms nobody publishes.

This is the third instance this session of a training claim that implies a corpus and never accounts for it, and the most concrete of the three, because the corpus is explicitly described as real adviser and client interactions.

This is the most consequential unanswered row on the record and it deserves stating without softening. The people this product profiles are prospects rather than customers. They have no relationship with the advice firm running the search, they did not supply the two thousand attributes assembled about them, and nothing indicates they are told a profile exists or that a wealth estimate has been produced in their name.

The inputs are described as public records and first party marketing data from the corporate parent, which means information a person gave to one company in one context is being used to estimate their value to another. Searched for a consumer facing privacy notice, an account of lawful basis, a route for a profiled individual to see or delete what is held, a retention schedule, or any position on state consumer privacy rights, and located none in the material opened. Queued check: the privacy policy, which for a product of this design is the single most important page to read.

Recorded as an unverified absence. Searched the product pages, the support material and third party listings for a named attestation, an information security standard, an audit report or a dedicated trust page, and located none, and no security page was reached.

A subsidiary of a major asset management group, distributed through a large custodian's marketplace, will have satisfied both parents' security requirements, so published credentials very probably exist and were not found in what was opened. Queued check: a trust or security page.

No licence is held and none is required, and the seventh application of the principle that a parent's supervision does not read across applies: this is a wholly owned subsidiary within a major regulated group's innovation arm, and that group's authorisations attach to its regulated businesses rather than to a marketing technology subsidiary.

What raises this above a routine entry is the specificity of the vendor's own disclaimer, which states not only that nothing constitutes investment, tax or legal advice but that the company does not endorse any investment adviser or make recommendations to any prospective client about selecting or retaining one. That second clause is an unusually precise disavowal, and it exists because a system that ranks consumers by their value to an adviser is close enough to steering that the line needed drawing.

No vendor assessed in this pocket sits closer to the reason this axis exists, and nothing at all is published. A model trained on which prospects historically became wealth management clients has learned the pattern of who advisers have historically taken on, and in United States wealth management that pattern is shaped by inherited wealth, geography, race and gender rather than by who would benefit from advice.

A system built on it will rank people resembling existing clients higher and rank others lower, so advisers stop calling the second group, those people never convert, and the next model trains on that. The projected client revenue estimate compounds it by attaching a monetary value to an individual inferred from public records. This runs across roughly three hundred firms simultaneously.

Searched for any fairness evaluation, any testing of score distribution across demographic or geographic groups, any account of which attributes the model uses, or any published impact assessment, and located none.

An explicit and total disclaimer exists, which is different from silence and lands on the same grade, and that is the fourth instance of the limitation already flagged on this axis. The vendor states in terms that it is not responsible for the accuracy of any information provided nor for the outcomes of any interactions entered into in reliance on it, which is a complete allocation of risk to the buying firm rather than an absence of one.

Searched for any warranty, accuracy commitment or correction obligation sitting alongside it and located none. The party with no route at all is the profiled individual, who cannot know a score exists, cannot see the attributes behind it, cannot correct an error in it, and experiences the consequence only as an adviser who never called.

The models are described as proprietary, which is a genuine disclosure that no external model provider carries the scoring, and it is the only part of the chain that is disclosed. The data supply chain is the one that matters for this product and it is described only by category, as first party marketing insights from the corporate parent and as public records, with no source, broker, aggregator or record type named.

For a system whose output is a wealth and conversion estimate about a named individual, the identity of the data brokers feeding it is the equivalent of a model provider elsewhere in this index, and searched for it across the product, support and third party material without locating any.

Three distinct routes into a firm's existing estate are published and they cover the systems that matter for this workflow. An application on a major customer relationship platform is named, an interface exists for programmatic use, and bulk loading handles batches of up to 2,500 leads for firms whose data does not sit in a connected system at all.

Distribution through a large custodian's provider marketplace adds a fourth route that is commercial rather than technical but does place the product inside an estate advisers already work in. Held below the top grade because searched for named connections to the client relationship systems most independent advice firms actually run, as distinct from the general enterprise platform named, and located none. Queued check: an integrations directory.

Recorded as an unverified absence. The product is delivered as a hosted application with an interface and a connected platform application, and nothing about hosting, region, tenancy or residency was located in the material opened. Searched for a storage location or any residency commitment and located none. The market is stated as the United States only, which makes a single domestic region the expected answer without it being a published one.

Searched for a rate, a tier, a per seat or per lead figure and located none. The vendor's own support material refers to pricing plans in the plural, so a tier structure exists and is described nowhere that was opened. The unit question matters more here than for a subscription tool, because the product consumes leads in batches and enriches per record, so whether a firm pays per seat, per lead enriched or per volume band changes the economics entirely for a firm running high volume acquisition. Queued check: the pricing page.

Genuinely narrow on every dimension this axis measures, which is what this grade means rather than a comment on quality. One buyer type, being an independent advice firm or registered investment adviser, with no bank, insurer, broker dealer or asset manager addressed. One country, stated explicitly on the vendor's own pages as being offered to persons in the United States only.

One function within that buyer, since this addresses acquisition and touches nothing in planning, servicing or compliance. What keeps it from being narrower is channel rather than coverage: the product reaches firms directly, through an application on a major customer relationship platform, through an interface, and through a large custodian's marketplace, and a separate programme exists for small firms, so within its one buyer type it reaches across the size range.