FP Alpha

Machine reading carries the whole throughput and an encoded expertise layer carries the judgement, which is the same split that placed the nearest competitor in this band and is applied here for the same reason. Removing the extraction does not leave a leaner product, it leaves an empty one, because there is no workflow spine, no data asset and nothing to analyse until a document has been read.

What holds this off the top grade is the analytical layer: the company describes its insights as coming from expertise contributed by dozens of specialist professionals, which is encoded human knowledge applied to extracted fields rather than reasoning by a model.

Worth recording that the extraction task here is materially harder than the competitor's, because wills, trusts, powers of attorney and healthcare directives are free prose with no fixed structure, where a filed tax return is a standardised government form with known fields. That places this record at the top of the band rather than moving it above.

One framing gestures at oversight and no design is published behind it. The company describes what it delivers as adviser verified outputs, which locates the review step correctly and asserts nothing about how the product supports it. Nothing executes, so the exposure is a wrong document rather than a wrong action.

Searched for a confidence measure on an extracted data point, any flagging of a passage the system found ambiguous, any account of behaviour on an unusual instrument or a poorly scanned page, and any reconciliation view placing an extracted value beside the source text it came from, and located none.

That last absence is the sharpest, because interpreting a trust is not field matching: provisions are conditional, drafted in varied language and frequently ambiguous on purpose, and an adviser reviewing a clean flowchart has no way to see which reading the system chose.

Searched for an extraction accuracy figure, an error rate by document type, a benchmark, a validation method, a quality assurance procedure or a revalidation cadence, and located none. Accuracy is asserted through adjectives and through the credentials of the people whose expertise was encoded rather than through any measurement.

The unmeasured quantity is unusually consequential because of the document types involved: a misread bracket produces a wrong projection an adviser may catch, while a misread distribution provision produces a flowchart that confidently shows a family the wrong inheritance, and there is no obvious signal to the reviewer that anything went wrong. Nothing published states how the system performs on ambiguous or unusually drafted instruments, which is where the value and the risk both concentrate.

Quantified customer claims exist on the vendor's own pages, including a stated saving of more than eight hours per client, and no attribution was visible in the material retrieved, so the named reference test cannot be met from what was actually read. This row is therefore recorded as unverified rather than evidenced, and it may well move. Independent editorial coverage of the product exists at a major research house, which is a credibility signal rather than a customer outcome. The founder leads an established advice firm, which explains the product's origin and is not a customer reference either. Queued check: the testimonials with attribution and any case study library.

Searched for any statement on whether uploaded tax returns, wills, trusts, directives or insurance policies are retained after analysis, used to improve extraction, or used to train, tune or evaluate any model, and located none in either direction.

The vendor's own marketing raises the question and leaves it open, because the capability is repeatedly described as adviser trained and expert trained artificial intelligence, which is a training claim with no account of what the training material was. If that means professionals contributed rules and review, it is unremarkable.

If it means the system learned from real client documents, that is a disclosure about a very large number of families' wills and tax returns which nothing addresses. This is the second instance this session of an accuracy or training claim that implies a corpus and never accounts for it.

Recorded as an unverified absence. Searched for a retention schedule, a deletion commitment, a statement of whether uploaded documents persist after analysis, or published processing terms available to a professional firm, and located none in the material opened.

The credit model raises a specific question nothing answers: because a credit is consumed per household and the resulting analysis is stored for later reference by any seat in the firm, the documents and their extracted contents plainly persist somewhere for some period, and neither the period nor the deletion route is described. Queued check: a privacy policy and processing terms.

Recorded as an unverified absence. Searched the pages opened 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. The consideration is arguably the highest of any vendor added this session, because of what a single household upload concentrates: a filed tax return with government identification numbers and full income composition, a will and trust naming beneficiaries and their entitlements, powers of attorney, and a healthcare directive setting out medical wishes. Few systems anywhere hold that combination for one family in one place. Queued check: a trust or security page.

A software supplier holds no licence and requires none here, and no penalty attaches. The boundary question is more pointed for this vendor than for most in the pocket because of what the output resembles. Analysis of a will and trust that identifies problems and recommends next steps looks like legal work, analysis of insurance coverage gaps looks like insurance advice, and the company sells to estate attorneys and accounting firms as well as to advisers, so the same output crosses three professional perimeters with different rules about who may say what. Searched for any statement drawing the line between analysis and advice, or any account of what a professional must add before an output leaves, and located none.

Searched for any evaluation of extraction performance across document types, drafting styles, jurisdictions or scan quality, and any account of how observations are prioritised, and located none. Two effects sit unexamined. Extraction reliability plausibly varies with how a document was produced, so a will drafted by a large firm using standard templates is likely read more accurately than one drafted decades ago or by a small practice, which means the households with the least professionally prepared paperwork get the least reliable analysis.

And the observation engine surfaces more where there is more structure to find, so complex wealthy estates generate substantial reports and simple ones generate thin ones, which shapes where an adviser invests planning effort.

Searched for a warranty, an accuracy commitment, a correction obligation or any allocation of responsibility between vendor and professional firm, and located none. The vendor's own description of its deliverables as adviser verified outputs is the allocation, made in a phrase rather than a term: the professional who reviewed it owns it.

The exposure travels further than usual because the outputs are client facing by design, since an estate flowchart and an insurance gap summary exist specifically to be shown to a family and used to prompt action. A household that acts on a misread trust may not discover the error for years, and the person whose documents were read has no described route to see what was extracted about them or to correct it.

Searched the pricing page, the solutions material, the product announcements and third party coverage for a model provider, a model family, a version, an inference host or any subprocessor, and located none. The capability is consistently attributed to the company's own engine and to the specialists whose expertise informs it, which describes ownership of the logic and says nothing about who performs the reading.

For a product whose entire function is passing a family's most sensitive legal and financial documents through a language model, whether that model is operated by this vendor or by a third party is a question a professional firm needs answered before its first upload, and nothing published answers it.

Recorded as an unverified absence rather than an evidenced one, and the distinction matters here because this is a layered tool whose value depends on its findings reaching the plan. Third party review material states that the product integrates with existing financial software without naming a single system.

Searched the material opened for a named client relationship system, a named planning platform, a programmatic interface or any account of whether findings flow outward or stay in the product, and located none. What is established is internal rather than external: permissions, usage tracking and customisation by team or office, plus a credit bank shared across firm seats. Queued check: an integrations directory, which for a vendor at this maturity very probably exists.

Recorded as an unverified absence. The product is reached through a browser with documents uploaded to it, and nothing about hosting, storage location, tenancy or residency was located in the material opened. Searched for a region, a residency commitment or any statement about where legal and tax documents are processed and stored, and located none.

This is close to the most complete commercial disclosure in the index and it is worth setting out because so few vendors attempt it. A public pricing page carries actual annual figures for the combined licence and for a single module bought alone, distinguishes a two year billing rate from a single year one, and states how pricing scales between an individual adviser and a firm.

Beyond the headline rate the unit economics are published too: the licence carries a bank of credits, one credit is consumed per tax document reviewed and one per estate household regardless of how many documents that household has, additional capacity can be bought as needed, and credits reload annually and are drawn by any seat in the firm.

A buyer can therefore model annual cost against expected case volume before contacting anyone, which is exactly what this axis exists to reward and what the direct competitor in this category, sold contact only, does not permit.

Three distinct professional buyers and three planning domains against a single jurisdiction. Licences are sold to sole practitioners, to advisory teams and to enterprise wealth firms running multiple offices, and beyond the adviser market the company states it also works with accounting firms and estate attorneys, which are genuinely different professions buying the same document reading for different reasons.

Product breadth matches, spanning tax, estate and personal insurance rather than one of them. Held below the top grade because the entire design is bound to one country's tax code, estate instruments and insurance conventions, and because no bank, insurer or asset manager is addressed as a buyer.