A

Aderant

The models are the engine of a real capability layered on a business system that plainly functions without them, which is the B band. The AI is substantial and named: MADDI, introduced in 2023 and described as the AI foundation embedded across Aderant cloud solutions; askMADDI, generally available in four named surfaces, Stridyn Analytics for natural-language business queries, Onyx for guideline terms and rule lookups, iTimekeep for work descriptions and narrative suggestions, and the cloud general ledger; and Agent Center, seven purpose-built agents opened to early access in August 2026.

What decides the grade against A is what the product is. Aderant sells practice and financial management, time capture, bill delivery, docketing and calendaring, and a talent suite, and it sold all of that for decades before MADDI existed. Remove the AI and Expert, iTimekeep, BillBlast, Milana, CompuLaw and the vi suite remain fully saleable systems. The vendor frames it this way itself, describing MADDI as making AI a core capability rather than an add-on, which is a claim about integration rather than about the models being the product.

Recorded on the other side, because it is the strongest argument for A: the 2026 positioning is aggressive, the home page leads with an intelligent ecosystem and AI built for the way law firms work, and the general ledger, receivables and time capture releases are all described as AI-driven. That is a platform adding an AI engine to a core capability, which is what the B band says. Verified 12 September 2026.

Reliability is asserted without measurement and grounding is claimed without a described method, which is C, and R15 governs how heavily it should read. This product generates no legal authority. askMADDI answers questions about a firm's own ledger, receivables and time data, and the agents work on billing, collections, rates, forecasts and evaluations, so the limbs about primary authority, openable citations and citator status do not bite and the record is not penalised for them.

What does bite is that the output is quantitative and consequential, and the vendor asserts quality without evidence. The assertions located: that Aderant applies controlled orchestration, safeguards and validation to help ensure AI is relevant and reliable, and that it delivers explainable, secure and performance-driven technology. No accuracy figure, error rate, test set or evaluation of any kind is published for MADDI, askMADDI or any agent.

Two grounding claims exist and neither is a described method: the Agent Center states that agents work with a clear record of the information used and the work prepared, and the Compliance Agent is described as citing the source language of the guideline requirements it finds. Both are real and both are one clause long. Recorded because it is the sharpest version of the risk here: a general ledger assistant that answers a partner's financial question wrongly, or a Time Agent that recommends the wrong UTBMS code across a matter, produces an error that reaches a client's invoice, and nothing published says how often either is right. Verified 12 September 2026.

A written commitment that the models work alongside a human decision-maker, with real controls, short of the full structure. The commitment is unusually crisp and is published on the Agent Center page in one line: agents prepare, prioritize, and draft, and the firm's team reviews and decides. That is a categorical boundary rather than a hedge, and it is repeated in the agent descriptions, each of which stops at drafting, ranking, flagging or recommending.

Around it sit two real controls. Agents are stated to follow product permissions and firm controls, so the firm's existing access model constrains what an agent can reach. And each agent is stated to keep a clear record of the information used and the work prepared, which is a review surface rather than a slogan because it lets a reviewer see the basis before accepting the draft. What holds it off A is that nothing below that line is published.

No threshold, confidence boundary or class of work is described at which an agent proceeds without review, no error handling or escalation path appears, and nothing states what happens after an agent is wrong. The GL Forecasting Agent is described as producing confidence ranges, which is a property of a forecast rather than a statement of model uncertainty, and it is recorded here rather than credited. The A case is real and is noted rather than taken: a categorical constraint can stand in for a numeric threshold on a product of this kind, which is a live band question logged in the pull's parking file, and it is not resolved in this vendor's favour unilaterally. Verified 12 September 2026.

Named customers in quantity and no figure attached to any of them, which is the B band. The naming is the strongest part of this record and it is specific: published client stories cover Chartwell Law, a National Law Journal 500 firm with nearly 300 attorneys adopting Expert Sierra; Michael Best and Friedrich, an Am Law 200 firm running BillBlast, iTimekeep and compliance solutions; Buchanan Ingersoll and Rooney on the vi suite; Miles and Stockbridge on Expert Sierra and Stridyn Analytics; and Gerber Ciano Kelly Brady and Tressler LLP. Those are real firms named with the products they bought.

The aggregate scale claims are also published: more than 2,500 client firms, 98 per cent of the Am Law 200, 86 per cent of the Global 100, 26 countries. What is missing is measurement tied to a customer. No named firm carries a figure for what changed, nothing is dated, and no method is published for any of it. The vendor's own headline metrics are unsourced and self-assessed, a net promoter score of 92, more than 35 new products released, and a claim of zero failed enterprise implementations at 100 per cent, none of which states a period, a population or a basis, so none is credited as outcome evidence.

Nothing published attaches any result to MADDI or to an agent specifically, which matters because the AI is the thing being graded. The client stories library is published and was not opened; under R25 it corroborates rather than carries a grade resting on the named-customer strip, and it is the artifact that would move this row if the studies carry dated figures with a stated basis. Verified 12 September 2026.

Confidentiality is asserted in general terms and cannot be read in a published instrument, which is the C band, with one specific commitment that keeps it off the floor. The specific commitment is on training and it is worth stating precisely: the vendor states that its AI architecture does not use any confidential, intellectual property, personally identifiable or sensitive data for model training, and that all input undergoes masking, anonymisation and filtering.

That is more than most records at this grade publish, and it is graded on the training signal as well as recorded here. Everything else on this axis is asserted rather than committed. MADDI is described as designed with the security, privacy and governance standards law firms require, as working within each firm's existing controls, and as helping keep customer data protected, private and governed. Those are three general assertions and none is a commitment a buyer could hold the vendor to.

The reason they cannot be is structural rather than accidental, and it is the finding on this record: no customer agreement is published anywhere on the estate. The only legal instruments in the footer are a privacy notice and an API terms of use, and the privacy notice expressly excludes data processed on behalf of customers in the vendor's role as processor through its products, which is precisely the client data a law firm cares about.

So privilege and work product are addressed nowhere, no retention or deletion commitment exists, no position on third-party model providers is published, and segregation is asserted through product permissions without being described. Verified 12 September 2026.

Nothing published on professional responsibility was located, which is the D band, and R15 requires naming which limbs bite before the grade is read as heavier than it is. The advice-line limb barely applies. This product does not produce legal work product, advise on a client matter, or generate anything filed with a court; the AI drafts e-billing appeals, collections outreach, time narratives, associate evaluations, rate explanations and financial forecasts.

The audience limb is answered and answered narrowly: the buyer is a law firm's finance, operations, talent and compliance functions, and the vendor is unambiguous about selling the business of law rather than the practice of it. What is genuinely absent, and what the grade records, is any statement connecting the output to the professional obligations of the people relying on it. Two places where it would bite are worth naming rather than passing over.

The Time Agent improves time narratives and recommends UTBMS task, activity and expense codes, and a time narrative is a representation to a client about work performed, made by a lawyer who signs the bill. The Compliance Agent turns outside counsel guidelines into machine-readable rules that then govern what the firm may bill, which is a contractual and ethical obligation being encoded by a model. Nothing published addresses the supervision either requires, no disclaimer of any kind was located on any surface read, and there is no published agreement in which such a position might otherwise sit. Verified 12 September 2026.

Principles are published without a mechanism, a testing regime or anything a buyer could audit, which is the C band. The principles are real and are set out as four named pillars: firm-centric by design, responsible by principle, grounded in legal expertise, and engineered for trust, the second glossed as designed to be governed, transparent and human-centered. Around them sit assertions of controlled orchestration, safeguards and validation, and a claim to deliver explainable technology.

None of it is a mechanism. No governance framework is named, no certification such as ISO 42001 is claimed, nobody inside the vendor is identified as accountable for AI, nothing describes what is evaluated before an agent ships, and no finding from any evaluation has been disclosed. The bias limb is unaddressed and it is not theoretical on this estate, which is why the note says so plainly. The vi suite ships AI-powered employee performance reviews, sentiment analysis and auto-summarisation, and the Talent Agent drafts evidence-based associate assessments and identifies development themes from performance history and feedback.

That is a model writing evaluations of named individuals inside a firm, feeding decisions about advancement and work allocation, in a jurisdiction where employment discrimination is actionable. Nothing published addresses whether those assessments are even across the people being assessed, what the model was trained on for that purpose, or how a firm would audit the output. Verified 12 September 2026.

Access is addressed and what happens to data after processing is not, which is the C band, and here the gap has a documented cause rather than being simple silence. What is published on the protection side: SOC 2 Type 2 examinations completed across named products, graded on the certifications row; a statement that agents follow product permissions and firm controls; a statement that MADDI aligns with each firm's existing permissions, configurations and governance; and, on the input side, a specific and unusual commitment that all input to the AI architecture undergoes masking, anonymisation and filtering, alongside the statement that confidential, intellectual property, personally identifiable and sensitive data are not used for model training.

What is absent is the entire data lifecycle. No retention period is published for prompts, outputs, agent working records or customer data generally. No deletion or return commitment on termination was located. No subprocessor list exists on any surface. No incident notification commitment to customers was located. The reason is structural and is recorded as a finding rather than a retrieval limit: there is no published customer agreement, and the privacy notice, last updated March 2024, expressly excludes personal data processed on behalf of customers through the products, so the one published instrument that might carry these commitments removes itself from the question by its own terms. Verified 12 September 2026.

Nothing published on who bears the loss when the system is wrong was located, which is the D band. No warranty, indemnity, liability cap, exclusion, service level, service credit or insurance position appears on any surface read. The cause is that no customer agreement is published at all. The complete legal inventory in this vendor's footer is a privacy notice, an API terms of use, a modern slavery and human rights statement and a code of ethics; there is no master subscription agreement, no terms of service, no data processing addendum and no order form template.

This is an enterprise vendor selling to the Am Law 200 under negotiated contracts, so an agreement certainly exists, but the buyer cannot read any of it before entering a sales process, which is precisely what this axis measures. The API terms of use is published and was not opened; under R25 it is named and not treated as load-bearing, because it governs consumers of the application programming interface rather than the platform and its AI, and a liability regime for API access would not settle what the vendor stands behind when an agent misprices a matter or a general ledger answer is wrong.

The exposure is worth stating because it is concrete rather than abstract: the AI here drafts appeals against e-billing deductions, prioritises collections, recommends billing codes and forecasts revenue, and every one of those errors has a direct financial consequence for the firm or its client. Verified 12 September 2026.

Real integrations, named and dated, with the direction described on the one that matters most, short of documentation an implementer could work from. Three named partnerships are published as announcements on the vendor's own newsroom. The Harvey integration is the deepest and the most specific: it connects Harvey's AI-performed legal work to iTimekeep so that completed work becomes draft time entries which lawyers review and submit through the firm's standard approval process, which states both what moves and in which direction.

The First AML integration brings automated anti-money-laundering and client due diligence checks into Expert and Expert Sierra, described as enabled through Aderant's professional services team. The LawPay partnership was renewed on a multi-year basis for payments. Beneath the partnerships sits genuine internal interoperation: sixteen products across financial management, work-to-cash, docketing and talent, with Stridyn as the common cloud platform and MADDI spanning them, and the Agent Center's stated design is that agents draw authorised billing, receivables, time, rates, talent and financial data from whichever Aderant products the firm licenses.

An API exists and is real enough to have its own published terms of use. What holds it off A is documentation: no public API reference, developer portal or connector register was located, no configuration detail is published for any integration, and the First AML route is stated to run through the vendor's own services team rather than through self-service configuration. Verified 12 September 2026.

Cloud delivery is stated plainly and neither the tenancy model nor the region is published, which is the C band. The delivery model is not in doubt and is described repeatedly: Stridyn is the vendor's AI-driven cloud platform, Expert Sierra is the cloud successor to the on-premises Expert product, and the current marketing is built around cloud migration, with a webinar series addressed to firms moving from Expert to Expert Sierra.

So a buyer knows the product is cloud-delivered and knows there is a migration path from an installed predecessor. What a buyer cannot learn is anything below that. No region, country or data centre location is named anywhere on the surfaces read. No cloud provider is identified. No residency option is offered or refused. No tenancy model is described, so nothing states whether a firm's data sits in a shared or isolated environment.

And nothing distinguishes where data is stored from where it is processed, which is a live question on this record for a specific reason: the AI layer's own hosting is never addressed, and the vendor's account of what sits behind MADDI is inconsistent, so a buyer cannot tell whether inference happens inside the same environment as the firm's financial data. The gap is more consequential than it would be for a domestic vendor, because Aderant reports clients in 26 countries and offices across North America, Europe and Asia-Pacific, and European firms have residency obligations that nothing published addresses. Verified 12 September 2026.

Certification is real, current and stated with its scope named product by product, and no evidence behind it is reachable at all, which is the B band. The scope naming is what distinguishes this record and it is better than most: rather than a single unexplained claim, the vendor publishes dated announcements of completed SOC 2 Type 2 examinations identifying which platforms each covers, one for Expert Sierra, vi by Aderant and iTimekeep, and a separate one for Onyx.

A buyer can therefore tell which products sit inside the audited boundary and which do not, which is the question the scope section of a SOC 2 report exists to answer and which most vendors leave open. C does not fire: these are announcements in the vendor's own voice with named subject matter, not badges on a page. What is missing for A is the whole evidentiary apparatus and any route to it. No auditor is named on any first-party surface read.

No report period or observation window is published, no certificate or report number, and no statement of which trust services criteria are covered. There is no trust centre or security page of any kind on the estate: the footer's complete inventory is a privacy notice, an API terms of use, a modern slavery statement and a code of ethics, so R5 is not reached because there is no access flow to grade, not even a sales-gated one.

Nothing is published on penetration testing, vulnerability remediation or incident response. Verified 12 September 2026.

The vendor describes what sits underneath in two ways that contradict each other and identifies nothing in either, which is the C band. The first account is that the technology is proprietary: MADDI is stated to be developed entirely in-house by Aderant's research, development and engineering teams, purpose-built, and not retrofitted from general-purpose AI models. The second account appears on the same page: MADDI is stated to combine enterprise-grade model platforms with Aderant's legal-domain expertise and proprietary AI capabilities.

Those cannot both be the whole truth, and R37 rule 2 governs, so neither is picked as the favourable one. What the conflict reveals is the thing the band measures: no model is named, no version is given, no provider is identified, and a buyer cannot tell whether their firm's financial data is processed by a model Aderant built or by a third-party platform Aderant licenses. Nothing states where inference runs and no commitment to notify customers when the model or provider changes was located.

There is no subprocessor list anywhere on the estate that would answer the question by another route. One adjacent disclosure is recorded and expressly not credited to this vendor's own supply chain, because it is a partner's model rather than a component of MADDI: the published Harvey integration brings Harvey's AI-performed work into iTimekeep, and Harvey is a named third-party AI whose own disclosures are its own. Verified 12 September 2026.

No pricing information is published at any level, including the unit of charge, which is the D band. There is no pricing page on the estate, and this is a page-inventory finding rather than a retrieval limit: the full navigation was read and it carries Solutions, Learn, Careers, Service and Support, About Us, Product Login and Momentum, with no pricing tier anywhere, and the single call to action across every page is to request a demo.

Nothing states whether the platform is licensed per user, per timekeeper, per firm or per module, nothing indicates a minimum, and no figure, band or term appears for any of the sixteen products. Under R10's closing discipline an estate that only invites a sales conversation is an absence and belongs in this note alone, so no VendorPricing row is written for this record. Two things are recorded and neither is credited.

The vendor publishes a marketing blog answering the objection that its cloud product costs two to three times the incumbent one, framed as a full cost comparison; it was not opened, and it is named here rather than treated as load-bearing because a blog answering a migration objection is not a published price and could not supply a unit of charge. And the agreement that would carry commercial terms does not exist publicly either, there being no master subscription agreement or order form template on the estate.

The absence is worth weighing against the buyer: these are Am Law 200 firms making eight-figure platform decisions with no published starting point. Verified 12 September 2026.

Segment coverage is described with real substance and evidenced by scale, with the boundaries left open, which is B. Who this is for is unambiguous and is demonstrated rather than claimed: law firms, and predominantly large ones, with published figures of more than 2,500 client firms, 98 per cent of the Am Law 200, 86 per cent of the Global 100 and 26 countries served. Functional coverage is enumerated product by product across four named groups, financial management, work-to-cash, docketing and calendaring, and people management, so a buyer can see exactly which parts of a firm's operations the estate reaches.

Roles are addressed by implication throughout, the buyer being finance, billing, operations, talent and compliance leadership. R15 applies to two limbs of the A band and the note says so rather than penalising silently: practice areas are not applicable, because this is business-of-law infrastructure that does not vary by whether the firm does patent litigation or trusts and estates; and in-house and government use are neither claimed nor relevant to a product built around law firm timekeeping, realisation and outside counsel guideline compliance.

What holds it off A is that the limits are not stated. No firm size floor is published, so a small firm cannot tell whether it is a customer; nothing identifies a segment the product is not for; and, most concretely, agent availability is stated to depend on which Aderant products a firm licenses without publishing which agent requires which product. Verified 12 September 2026.