Holistiplan

The extraction reading applies and the vendor's own candour is what caps this grade. Machine document reading carries the throughput unambiguously: a full return with schedules is read in roughly 45 seconds, and removing that capability does not leave a working product with less automation, it leaves a calculation engine with nothing to operate on, since the entire commercial proposition is that nobody keys the return. That is the extraction led shape rather than the legacy platform shape.

What prevents the top grade is that the analytical layer is not modelled and the company says so plainly: its chief technology officer describes the calculation engine as built by humans and validated against a large corpus of real returns. So the opportunity flags, the bracket and surcharge analysis and the scenario projections are deterministic tax computation and encoded rules. Machine reading carries the volume; human built logic carries the judgement.

The product acts on nothing, which limits the exposure, and no oversight design is published for the part that matters. The output is a report an adviser reviews and presents, so no transaction executes and no record updates without a person, and the documented process includes a review step where the professional decides which flagged opportunities apply.

What is absent is any handling of the failure mode the product actually has: searched for a confidence measure on an extracted field, any flagging of a low confidence read, any description of behaviour on an unusual or poorly scanned form, and any reconciliation between extracted figures and the source, and located none. The tell is behavioural.

Independent reviewers of this product repeatedly instruct users to verify machine read figures against the original return before trusting a projection, which is an oversight step the user community has had to invent because the interface does not surface it.

A validation basis is stated and no result is, which puts this above the silent norm and short of the top grade. The company's chief technology officer, named and quoted, states that the calculation engine is built by humans and validated against hundreds of thousands of real tax returns.

That is a disclosed validation method with a stated corpus scale, attributed to an accountable officer, and it is accompanied by unusually candid architectural transparency, since the vendor volunteers that the analytical layer is deterministic rather than modelled at all.

Held below the top grade because the two quantities that decide whether a generated report is safe to hand a client are given as adjectives: extraction is described as highly accurate and the combined result as near complete data accuracy, with no field level error rate, no benchmark, no error rate on opportunity flags, and no revalidation cadence against annual changes in tax law.

The evidence is an independently measured market position rather than a curated reference, which is the stronger form. The 2026 edition of a long running independent adviser software survey, run by two named industry researchers and fielded across the profession, records this vendor as leading the tax planning category at 39 percent market share, with the highest satisfaction rating in that category. Neither figure is selected by the vendor and neither can be.

That sits alongside a competition win at a fee only adviser network in the company's founding year and reported record adoption growth. This index treats an independent party's quantified finding as an alternative route to the top grade beside the named customer with a measured outcome, and the same reasoning was applied in this session to another vendor in the same pocket. Absent, and worth naming: no individual advice firm is named reporting its own result.

Searched for a statement on whether uploaded returns are retained after processing, used to improve extraction, or used to train, tune or evaluate any model, and located none in either direction. The gap is pointed here rather than routine, because the vendor's own accuracy claim raises the question and leaves it hanging: an engine described as validated against hundreds of thousands of real tax returns implies a corpus of real filed documents belonging to real households, and nothing published states whose returns those were, whether they were customer uploads, on what basis they were used, or whether that use continues. A validation claim is a reason to trust the product and, stated this way, also a disclosure that something happened to a very large number of tax returns which is nowhere accounted for.

Recorded as an unverified absence. Searched for a retention schedule for uploaded returns, a deletion commitment, a statement of whether a document persists after a report is generated, or any published processing terms available to an advice or accounting firm, and located none in the material opened.

The question has a specific edge for this buyer: tax practitioners in the United States operate under an explicit obligation to maintain a written information security plan covering the taxpayer data they hold, and a practice adopting this product needs the vendor's position to complete its own plan. Queued check: a privacy policy and any processing terms.

Recorded as an unverified absence rather than an evidenced one. Searched the vendor's news material, product descriptions and third party profiles for a named attestation, an information security standard, an audit report or a dedicated trust page, and located none, and no security page was found to open.

The consideration is as sharp as it gets for a vendor of this size, because the object the product ingests is a complete filed individual tax return, which carries the taxpayer's government identification number, those of a spouse and dependants, full income composition, employer, charitable giving and account details. Few documents in a household's possession concentrate that much. Queued check: a trust or security page, which for a vendor with this market share and this private equity backing very probably exists.

A software supplier holds no licence here and requires none, and no penalty attaches. The position deserves precision because the output sits closer to a regulated act than most: a client facing document setting out a household's tax position and recommending conversion, harvesting or charitable strategies looks like tax advice to the person receiving it, and it carries the advice firm's own branding because the report is white labelled.

The professional obligation stays entirely with the adviser or accountant who signs it out. Searched for any statement about the boundary between analysis and advice, or about what the firm is expected to review before the report leaves, and located none.

Searched for any evaluation of extraction performance across form types, preparer software, filing complexity or document quality, and any account of how planning opportunities are prioritised, and located none. Two effects sit unaddressed. Extraction quality plausibly varies with how a return was produced and scanned, so households whose returns come from a small preparer or arrive as a photographed copy may be read less reliably than those filed through major software.

And the opportunity engine surfaces more where there is more to find, so a complex high income return generates a substantial report while a simple one generates a thin one, which quietly signals to an adviser which clients repay tax planning attention and which do not.

Searched for a warranty, an accuracy commitment, a correction obligation or any allocation of responsibility between vendor and professional firm, and located none. The exposure is unusually concrete because of how the output travels: the report is white labelled and client facing, so a single misread figure propagates into a document delivered under the advice firm's own brand and used to justify a conversion or a harvesting decision with real tax consequences.

Independent reviewers advise verifying every extracted figure against the source before relying on it, which means responsibility currently rests with the professional by community practice rather than by any published term. The taxpayer whose return was read has no described route to see what was extracted about them or to correct it.

The layer that is not modelled is disclosed and the layer that is remains unattributed, which is an unusual inversion. The vendor volunteers that its calculation engine is human built, which is genuine information about where no external dependency exists.

Searched the news material, product pages and third party profiles for the provider behind the document reading capability, whether extraction runs on a commercial service or an internally trained model, any model family or version, or any inference host, and located none. Given that extraction is the capability the entire product rests on, the identity of whatever performs it is the single most consequential unnamed dependency on this record.

The integrations are real, named and chosen well, and they are a short list rather than an estate. Connections are named into widely used adviser client relationship systems and financial planning platforms, which are the two categories that matter for this workflow because the tax findings need to reach the plan and the client record.

The positioning is explicit that this is a layer over an existing stack rather than a platform replacement, which is a coherent strategy and makes integration the product's dependency rather than a convenience. Held below the top grade because searched for a programmatic interface, a connection into tax preparation software where the source return originates, a custodian link or any account of bidirectional write depth, and located none of it.

Recorded as an unverified absence. The product is reached through a browser with documents uploaded to a portal, and nothing about hosting, storage location or residency was located in the material opened. Searched for a region, a residency commitment or any account of where uploaded returns are processed and stored, and located none. The buyer base is domestic and the documents are domestic, so a single region is the expected answer, but expected is not published.

This vendor is close to a demonstration of why this index refuses to grade commercial terms from aggregators. Three independent review sites carry three materially different accounts of the same product's price, one describing entry pricing near 79 dollars a month, another a band of roughly 1,500 to 3,000 dollars a year by tier, and a third simply recording contact sales. They cannot all be current and a buyer reading any one of them in isolation would form a wrong expectation.

What is established from the vendor is structural: pricing is by tier against the number of returns analysed and the number of seats, a seven day trial exists, and a setup fee is commonly reported. Searched the vendor's own material for a published rate and located none.

Two distinct professional buyers and three product domains against one jurisdiction and a stated ceiling on firm size. The product is sold both to financial advice firms and to accounting and tax practices moving from compliance work into advisory work, which are different buyers with different workflows, and it now spans tax analysis, property and casualty insurance analysis and estate document handling rather than tax alone.

Against that, the entire design is bound to one country's tax system, since the object it reads is a United States individual return and the analysis concerns that country's brackets, surcharge thresholds and retirement account rules, and published positioning describes the buyer range as running from solo practitioners to firms of around fifty employees rather than into enterprise. Held below the top grade on the single jurisdiction and the stated size ceiling.