Novo vs Precedent: how they compare in 2026

The artificial intelligence is the product and nothing survives its removal. Novo sells two outputs, a medical chronology and a demand letter, both generated from uploaded records, and the whole estate is organized around exactly those two things: the navigation has two product pages and no others. There is no case management system, no document repository, no intake pipeline and no workflow layer underneath that a firm would still pay for; a buyer uploads records and receives drafts. The terms confirm the architecture rather than leaving it to marketing, defining an AI Feature as any component of the services incorporating AI Technology and defining that as machine learning, deep learning, large language models and neural networks generating content from user prompts, and then defining the demand letters and chronologies created under the agreement as Outputs of an AI Feature.

The models drive a core capability on a service platform that would still work without them. Claim setup with the carrier, policy verification and medical record retrieval are operational services with a workflow and human core. The vendor presents them as the first three of four steps before AI enters, and a firm could buy them alone and receive a working service. The models power the fourth step, which holds the flagship. Demand Composer drafts the demand package from the case documents, MedChron builds medical chronologies, and Global Demand produces packages for several plaintiffs. Other tools sort incoming files, benchmark demands and anticipate adjuster tactics, and a voice assistant called Ask Casey sits alongside them.

Completeness is asserted on the product pages and disclaimed in the agreement, and nothing is measured. The marketing claim is specific: the chronology captures every injury, treatment and medical event so a firm never misses a critical detail. No accuracy figure, test set, evaluation, error rate or statement of failure modes accompanies it, and no accuracy or benchmark page exists on the estate. Nothing describes grounding either. No statement says whether an entry in a chronology cites the page of the record it came from, which is the control that would let a paralegal verify the output quickly. The agreement runs against the marketing and governs where they conflict. It provides the services as is with all faults, disclaims any warranty that they will achieve intended results or be free of errors, and makes it entirely the user's obligation to review, revise, edit and proofread all outputs for accuracy and completeness. A vendor cannot both promise that nothing is missed and require the buyer to check everything.

Grounding is real, documented and checkable. In what the vendor calls a Closed Loop, every medical fact, diagnosis and billing code in a demand letter is linked directly to its source page in the raw medical record. The vendor tells buyers they need not trust the AI, because they can click the citation and see the document. The same page carries a heading that reads Zero Hallucinations, yet the vendor also has reviewers check factual integrity before delivery, which a system free of hallucinations would not need. Accuracy therefore rests on source links and human review rather than on model behavior. That is a sound design, and it is not what the headline claims. No measured accuracy figure, test set or method is published, and no failure mode is named.

The review point is written into the agreement rather than implied by marketing, and the rest of the control structure is missing. Section 16 states that it is entirely the user's obligation to review, revise, edit and proofread all outputs for accuracy and completeness. The user shall not act on an output, or decide not to act based on one, without first doing so and seeking counsel from an attorney licensed in the applicable jurisdiction. It adds that outputs are for convenience and information and cannot be relied on to guarantee any outcome. That is a written commitment placing a supervising lawyer between the model and the client, and the review surface is real, since the deliverable is an editable draft the firm finishes. Nothing is published on the system's own side. Nothing describes what runs unattended, no confidence signal or uncertainty indicator is mentioned, nothing states how a firm should verify a chronology against the underlying records, and nothing addresses what happens when an output is wrong. The obligation is allocated, and the means of meeting it is not described.

Human review is written into the workflow at two stages, and the controls around it are not published. The vendor's own trained expert reviewers check each draft for narrative logic and factual integrity before the firm sees it, an oversight step inside the vendor rather than an obligation placed on the buyer. The attorney then receives what the vendor calls a draft ready for review, with every medical fact traceable to its page in the record, so checking is a designed workflow rather than an instruction. No mode is described, no threshold is stated at which the system stops or escalates, and nothing says what happens when output is wrong. The source linking architecture is named rather than described as a control with stated limits.

There are named customers without figures and figures without a named customer. Three testimonials carry full attribution: Tanya Hardin, a paralegal at Johnson Law, who reports the program cut her time by half; Chris Carsten, a partner at Armada Law; and Daniel Swenson, an attorney at Robert Wilson, who says settlement offers came back higher after adoption. None carries a measured figure beyond the halving claim, and none is dated. Separately, the home page publishes five and a half hours saved per case. The company's February 2025 announcement states that early adopters cut total time on demand packages by more than 50 percent, with no firm named against either. One display element does not hold up. A logo wall says the product is trusted by lawyers and recognized by six organizations. They resolve to Best Lawyers in America, the Million Dollar Advocates Forum, Lawyers of Distinction, Premier Lawyers of America, the Top 100 National Trial Lawyers and Best Law Firms 2024. Those are attorney award and directory marks, not recognition of this product. A row of Fox, CBS and BBC logos under a featured on heading links to no coverage.

Deployment evidence is real and partly unverifiable, with one headline figure out of step with its source. A named case study with Hines Law Firm reports settlements up 16 percent and the likelihood of tenders up 71 percent, measured across hundreds of personal injury cases. Named customer logos include Catania, Ramos Law and Sand Law. Testimonials carry a person and a role, among them a Personal Injury Director at Hines Law who describes drafting thirteen demands in a day. Aggregate figures are more than 100 law firms, more than 50,000 demands sent, a 95 percent retention rate and a 4.7 star rating, none with a stated basis. The case study page was not opened, so the method, comparison basis and period behind its figures are not known and the page may say more. The home page banner claims 25 percent higher settlements while the case study it links reports 16 percent, and nothing reconciles the two.

The privilege question is expressly addressed, and two other parts are missing. Section 14 defines confidential information to include, by name, any information protected by attorney client privilege and any sensitive or protected health information. It commits Novo to protect that information with at least a reasonable degree of care, to use it only to perform its obligations, and to restrict disclosure to those who need it. The privacy policy repeats the commitment, stating that Novo will use industry standard security measures for privileged information and protected health information. Section 16 also has the customer warrant that it holds the rights and consents needed for privileged client data it uploads, so the agreement engages the question head on. The other side of the same document cuts against it. Section 7 reserves an unlimited license to use anonymized outputs to improve the services and to create training data, so privileged material is protected in one clause and used in anonymized form in another. Nothing describes separation between firms or walls at matter level, or what the named model providers retain.

Confidentiality is covered in contract and in architecture, while privilege and work product are not addressed anywhere. The Master Service Agreement puts Protected Health Information inside its definition of Confidential Information, binds both parties to strict confidence that survives termination, and requires subcontractors to be bound. It also commits to handling consistent with HIPAA. Behind the contract sit isolation between firms, encryption, and a bar on using case files for training or sharing them across clients. The firm is said to own its data. Nothing addresses what an underlying model provider may retain from the medical files and demand strategy that pass through the system.

A real published position, with no engagement with the professional rules themselves. Section 16 states that the services and outputs are for convenience and information only, do not constitute legal advice, are not an offer to represent the user or their clients, and do not create an attorney client relationship. It goes further than a disclaimer by imposing conduct. The user must review, revise, edit and proofread every output, and must not act on an output or decline to act on one without first seeking counsel from an attorney licensed in the applicable jurisdiction. It closes by stating that outputs cannot be relied on to guarantee any outcome in a matter. That is a coherent account of where the product stops, published where a buyer reads it before signing. What is absent is the professional layer. No bar association, rule of professional conduct or ethics opinion is named, nothing addresses the supervision and competence duties that attach when a machine drafts a demand, and nothing addresses what a client is told.

The advice line is never drawn, though the vendor writes about professional conduct elsewhere. The Master Service Agreement has no statement that the service is not legal advice. Marketing presents the product as a drafting and workflow service rather than in terms of advice, and it is sold only to plaintiff firms. Every draft passes through the vendor's reviewers before it reaches the attorney, so lawyer review is built in and stated. A vendor article on billing for AI powered services tells firms their use of technology must not compromise duties of loyalty and communication. Nothing states what the product is and is not, who may rely on it, or how it supports competence and supervision, and no jurisdiction limits are named.

No governance position of any kind is published. There is no responsible AI page, no principles statement, no accountable owner or function named, no testing regime before release, no management system and no certification. Nothing addresses uneven output, and the gap has a concrete shape on this product rather than a formal one: a demand letter is an argument about how much an injury is worth, and nothing published states whether the drafting has been examined for systematic variation across injury types, treatment histories, or the demographic characteristics that appear throughout a medical file. The estate is small and was inventoried in full on 4 September 2026, comprising two product pages, a blog, a service providers page, terms of service and a privacy policy; none of them addresses governance.

No governance position is published for a system that drafts the document a claimant's recovery is negotiated from. Nothing identifies who inside the vendor is accountable for the models. No testing or evaluation before release is described, no results are disclosed, no policy or principles statement exists, and no certification such as ISO 42001 is claimed. Nothing addresses uneven output across injury types, jurisdictions or claimant demographics. The security page does carry a section headed AI Safety with real content, but what it describes is the product's quality controls, source links and expert review. Those govern the quality of a draft. Governance is a different matter, namely who answers for how the system is built, tested and corrected, and that is not published.

Most of the ground is covered, and the retention period is the notable gap. Storage location is stated: the privacy policy confirms that data is held on Novo's servers in the United States and names what is stored, including account information, resultant data and training data. Rights of access, rectification, erasure, restriction, portability and objection are published with a process and a clock, exercised by email, with a best efforts response in forty five days that can be extended to ninety. Incident practice is committed rather than implied, the policy stating that in the event of a breach Novo will notify by email or by a notice on the services. The supplier picture is disclosed by name, with a published service providers list naming eight companies and components. Two limits remain. No retention period is stated at all, the policy saying only that data is kept as long as necessary and that the length may vary. And the commitment to anonymize data before it reaches the AI providers is qualified as best efforts rather than absolute.

Data handling is documented in substance, and incident response is the gap. Encryption is AES-256 at rest and TLS 1.2 or better in transit with perfect forward secrecy. Access inside a firm runs on roles, retention is set by the customer rather than the vendor, and permanent deletion is available. Client files are kept out of model improvement, and health data is handled under a design built for HIPAA. All of that is published policy rather than marketing. No incident or breach notification practice appears on any page, on a platform that takes in complete medical files and holds health records for every case it works.

The allocation of loss is published, readable before signing, and runs one way throughout. Section 16 provides the services as is and with all faults, disclaiming all warranties including merchantability, fitness for purpose and noninfringement, and expressly warranting nothing about meeting requirements, achieving intended results or being free of errors. Section 17 caps direct damages at the total amount paid in the six months before the event giving rise to the claim, and excludes lost profits, data loss and consequential damages entirely. Section 18 runs the indemnity from the user to Novo, covering use or misuse of the services and breach of the terms; there is no indemnity from the vendor anywhere, not even for intellectual property. No warranty on output, no service credit, no insurance position and nothing addressing a chronology that omits a treatment or a demand that misstates damages is published. The exposure the product creates is addressed through a standard limitation clause that disclaims it rather than left unstated.

A real liability position is published, and it stops short of wrong output. The Master Service Agreement sets out mutual indemnification against third party claims arising from negligence, material breach, legal violations and other listed causes, with prompt notice and cooperation required. Aggregate liability is capped at amounts paid in the preceding twelve months, infringement claims excepted, and neither party bears indirect, consequential or punitive damages. The vendor warrants its authority to perform and that it does not infringe registered US copyrights, patents and trademarks. Nothing a buyer can invoke applies when the output itself is wrong. No warranty covers the accuracy or completeness of a demand package, clause 6.3 disclaims all warranties beyond the limited one, and no insurance is referenced. One clause says the vendor has no liability for failing to identify or notify the client of any demand, on a service that sends demand letters to carriers for the firm. The published agreement is an abridged summary. Fuller clause language appears in search results, and the full limitation wording, which could not be read, may say more.

Four case management systems are named on the vendor's own estate, and nothing published describes what moves between them. The home page carries CASEpeer, Smokeball, SmartAdvocate and MyCase under a heading identifying them as case management integration partners, presented as logos with no description, and a blog post announces integrations with MyCase and CASEpeer by title. That is a relevant set of systems for a plaintiff product. What is missing is everything an implementer would use. There is no statement of what syncs, in which direction or on what trigger, no field mapping, no configuration guide, no API or developer documentation, and nothing about what a firm must do to connect the two systems. No document management, electronic signature or billing integration is named. The integrations announcement post was not opened and may say more.

Integrations reach the case management systems this buyer runs, and their depth is not documented. Three are named, each with its own page. In SmartAdvocate, Demand Composer is said to run directly inside the platform. In Clio, demand creation is described as streamlined within Clio Manage, and in Litify demands are generated and managed in the firm's matter system. For plaintiff personal injury, those three cover most of the case management market. The vendor was a finalist for Best New App in the 2025 Clio Integration Awards, an assessment of the integration itself, and a dedicated integrations index page exists. The three integration pages were not opened, so what syncs, in which direction and what a firm must configure are not known, and the pages may say more. No API or developer documentation was found, and no document management or electronic signature connection is named.

The residency question is answered plainly and the tenancy question is not. The privacy policy states that personal data about end users and visitors is stored in the United States, that the services are intended for use only inside the United States, and that a user in the European Economic Area or elsewhere consents to transfer of their data there. The terms reinforce it with a geographic restriction section stating that the services are based in the United States and provided only to persons located there. That is a stated storage location, published where a buyer can read it. Against it, nothing states whether the platform is single tenant or multi tenant, no dedicated or isolated option appears at any tier, and no region choice is offered. The named infrastructure providers, Railway and Supabase, are listed on the service providers page, but no statement ties them to a region or describes the hosting arrangement.

Residency is stated at country level and nothing below it. The vendor states that all data processing and storage occur only in the United States and that client data never leaves US soil, framing this as data sovereignty. Tenancy is stated as logical isolation between firms. No cloud provider, region or data center is named, and no failover or secondary location is described. There is one cloud service, with no private, dedicated or on premises version. Residency is the vendor's architecture rather than a customer choice, and a firm has no way to require or verify placement.

One badge, no scope, no date, no report. The home page displays a HIPAA compliance mark as an image, with no statement of what was assessed, by whom, over what period or against which controls, and the underlying asset filename misspells the statute. HIPAA is in any case a statutory regime under which a business associate attests to its own compliance, not an independent attestation. No business associate agreement is offered or mentioned anywhere, despite the product being built to ingest medical records. No SOC 2 of either type, no ISO certification, no penetration test summary and no named auditor appears on the estate, and there is no trust center or security page, so there is nothing gated to request and this is an absence rather than a retrieval limit. What the vendor does publish in this territory is described in general terms only, the privacy policy referring to physical, electronic and managerial control procedures and to vetting of outsourced providers, without naming a single control. The estate was inventoried in full on 4 September 2026.

SOC 2 Type II is claimed, and nothing lets a buyer see the evidence. The vendor states it is SOC 2 Type II certified on the strength of annual audits by independent third parties covering security, availability and confidentiality. Its architecture is said to be designed for HIPAA compliance, and it offers to sign a business associate agreement for firms handling Protected Health Information. The security page is the vendor's own and is substantive. No auditor is named, no audit period or report date is given, and the scope is not described. There is no download, no stated request path and no security review process. No other attestation is claimed, and no penetration testing or vulnerability program is described.

The supply chain is disclosed by name on a dedicated published page, and stops short of the models themselves. The service providers page, incorporated into the terms by reference and reachable without an account, lists eight suppliers with links to each one's terms. GPT by OpenAI and Claude by Anthropic are the AI services used in the platform, Google OCR and the open source Tesseract engine handle optical character recognition, Railway provides infrastructure and deployment, Supabase is the backend platform, Gotenberg handles PDF conversion and Redis handles caching. Both AI providers and the hosting layer are named. Three limits remain. The models are identified only by product family, as GPT and Claude, with no version stated, so a buyer cannot tell which model reads a medical file. The page states that the list is not exhaustive and may be updated as the platform evolves, with no notice commitment. And section 13 of the terms authorizes Novo to let third party providers process user data whether or not those providers are disclosed to the user or white labeled.

Models are referred to generically, and nothing underneath them is identified. The one trace of the supply chain is the security page's training statement, which mentions public foundational models and so shows that foundational models are in the architecture. It names no model, provider or version, and nothing states whether the generative layer is built by the vendor or supplied by a third party. The Closed Loop is described without saying what generates inside it. No commitment to notify customers of a model change is published, and with nothing named, there is nothing such a notice could cover.

No pricing information is published at any level, including the unit of charge. The site navigation carries only the two product pages and a blog, with no pricing page, and the two calls to action are a free draft and a booked demo. The only commercial statement on the marketing estate invites a buyer to start on a free plan and upgrade when ready. It names no paid tier, rate, unit or included volume, and says nothing about what the free plan contains or where it stops. The terms describe payment mechanics without figures. Portions of the service may be offered free and certain features are charged, fees are payable in US dollars within thirty days of invoice, and late payment accrues interest at one and a half percent per month. Payments are final and not refundable, and Novo reserves the right to set, remove and revise prices at any time. The amounts themselves live in an order form the buyer sees only through a sales process. A buyer cannot establish whether the product is charged per user, per case, per document or per page.

Pricing is public for the flagship only, and the marketing and the agreement disagree about commitment. Demand Composer's rate, trial and fee terms are all published, so one demand can be priced up front, while the rest of the platform has no published rate. The home page states there are no contracts, and the security page that there are no long term contracts. Clause 9.2 of the Master Service Agreement provides for automatic renewal for a further twelve months with a 7 percent fee increase, unless the agreement is canceled in writing thirty days before the service end date. Both statements are published, and the agreement is the binding text.

The segment is stated without ambiguity and one real boundary is drawn. Every surface addresses personal injury plaintiff practice, with the home page headlining the AI built for PI. The scope is narrower still: the stage before litigation, built around the demand package rather than the whole matter. The company's own announcement frames the market as the roughly eighteen million demand packages US personal injury firms send each year. The terms name the users as law firms, attorneys and legal staff. A limit is published: the services are based in the United States and offered only to persons located there, so a buyer outside the country is told plainly that the product is not for them. The rest of the boundary is absent. No firm size is addressed, and nothing states whether the product suits mass tort, medical malpractice or workers compensation work as against motor vehicle claims. No variation in demand practice by state or jurisdiction is discussed, and nothing says which record types or matter types it handles poorly.

Coverage is deep within plaintiff personal injury and narrow by design, and its limits are not stated. The segment is plaintiff personal injury law firms, positioned explicitly against the carrier side. The vendor says it serves all firm sizes and handles small and large cases alike. Workflow coverage runs across the work before litigation, from claim setup and record retrieval to demand drafting and adjuster preparation. Named customers range from regional practices to firms that advertise heavily. Nothing says which case types, injury categories or jurisdictions fall outside coverage, or whether the product suits work once litigation starts. A testimonial refers to requirements specific to a jurisdiction, and no jurisdictional scope is published.