OnitX ELM vs Provakil: how they compare in 2026
Invoice exceptions are the main AI workflow in OnitX ELM's Spend Agent, while Provakil applies generation to case material and contracts. OnitX joins invoice review to budgets, accruals, vendor scorecards and configurable matter approvals. Provakil combines counsel fees with court tracking, order summaries and regional language translation. The AI Legal Index identifies Spend Agent's explanations for invoice findings as a distinction for billing reviewers. Its surrounding platform also gives firms submission status and feedback through a vendor portal. Within Provakil, the AI Legal Index identifies contract playbooks and traceable proposed clause changes as another use of legal context. Both provide legal departments with work and cost records, but the automated output differs. OnitX helps reviewers act on billed charges; Provakil supplies case preparation material, contract revisions and collections workflows.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. Hover a grade to see what the letter means on that axis.
AI Centrality
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
OnitX ELM combines matter, spending and vendor administration with Spend Agent as an optional invoice-review module. The agent interprets natural-language billing guidelines, examines invoice lines, accounts for engagement terms and recommends financial adjustments. Administrators can apply guidelines at client, vendor and matter-vendor levels. Conventional invoice validation, approval chains, budgets, accruals and reporting continue to operate separately. Counsel Exchange supplies the vendor submission and feedback workflow, while the AI adds contextual explanations and review decisions. The platform remains a configurable legal system of record without Spend Agent. Its AI contribution is concentrated on a core spending task rather than replacing every matter-management or financial function.
AI summarizes court orders and petitions, translates regional language material and produces case progress reviews inside the litigation workspace. An in case chatbot answers from the orders, hearing history and files linked to that matter. Contract Intelligence extracts more than 30 metadata fields, identifies obligations and proposes clause rewrites against company playbooks. Teams use those results in contract review and case preparation while continuing to manage assignments and approvals in the surrounding software. The underlying case repository, notice delivery, billing controls and approval routing would still operate without generative models. Borrower segmentation and bulk notice templates also use configurable rules.
Citation Accuracy and Hallucination Disclosure
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
Spend Agent connects findings to applicable billing guidance and opens the affected invoice line, showing the original amount, reason and recommended correction. Matter-level terms can supplement general guidelines, and reviewers can inspect the supplied engagement material. The frequently cited 92% result comes from a 2025 study of six general-purpose models on 50 invoices and 492 lines, not a validation of the current configured Spend Agent. That study used anonymized and synthetic invoices and found considerable disagreement among experts about dispute reasons. Current output therefore has inspectable business sources, but no corresponding reproducible product-wide accuracy measure. The workflow evaluates billing compliance, not judicial authority or subsequent case history.
Court summaries draw on orders and petitions held in the case workspace, and the matter chatbot uses the linked case history rather than an unrestricted knowledge base. Provakil reports 99.5% accuracy for court tracking, without a test set, measurement period or error breakdown. That operational figure does not measure generated summaries, translations or clause rewrites. Contract Intelligence similarly claims accurate extraction from millions of contract data points without reporting evaluated results. The service agreement warns that third party public information may be inaccurate and tells users to consult independent sources. A demonstrated citation link, retrieval evaluation and unsupported answer path are not provided.
Autonomy and Oversight Model
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
Spend Agent offers Sentry, Auto and Disabled modes for each vendor. Sentry leaves approval or rejection with a person; Auto can approve or reject invoices itself. A separate adjustment control can prevent monetary recommendations and automatic corrections while retaining violation analysis. Reviewers inspect red-marked lines, reasoning, proposed adjustments and totals, then accept, reject, alter or reapply recommendations. Client-wide guideline changes require an administrator, and matter-vendor guidance can narrow the applicable terms. The configurable appeals process returns adjusted invoices to firms, with response deadlines, maximum appeal rounds and a designated reviewer. Ordinary approval chains can escalate according to monetary authority. These controls distinguish AI actions, human decisions and routes for reconsideration.
Lawyers review contract changes in a collaborative editor with version history and clause playbooks. Approval workflows route agreements according to contract type, value, risk and business rules. Litigation workspaces separately assign tasks and collect counsel updates. AI redlines and risk flags remain traceable, with human oversight throughout the workflow. Provakil's legal operations guidance treats model output as draft material requiring verification, and the subscriber agreement requires independent checks of public information. The product does not specify a model confidence boundary that blocks an answer, a mandatory approval before every generated communication, or a recovery procedure after an incorrect recommendation.
Operational and Outcome Evidence
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
Corteva's 2021 to 2025 transformation combined Onit workflows, managed services, internal AI expertise and consulting partners. Its case describes invoice processing, supplier-record migration and multilingual assistants, with more than 1,000,000 tasks streamlined and nearly 1,000,000 hours saved across the program. The calculation method and the contribution of OnitX ELM alone are unspecified. Separately, Spend Agent's August 2026 customer aggregate identified 18% of reviewed spending for potential adjustment, with a range of roughly 10% to 45%. That measure concerns recommendations rather than money actually saved, and it lacks customer names and a sample size. Neither result isolates the current agent's causal effect on a department's final costs.
Asian Paints tracks more than 2,500 cases in Provakil, and Bajaj Finserv automates 550,000 notices each month. The contract management business names Domino's with more than 10,000 digitized contracts. CorpLit Consultants reports saving more than 40 hours of administrative work per week across its docket. Government and public sector users named by Provakil include IOCL, the Government of Tamil Nadu and the Election Commission of India. These customers use different parts of the software, from litigation tracking to notice delivery and contract storage. No dated evaluation ties the volume figures to an AI accuracy result, a comparison baseline or a specified measurement method.
Privilege and Confidentiality Posture
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
Customer Data is confidential under Onit's agreement, with purpose limits, reasonable care and access restricted to personnel or agents who need it. Product-improvement and benchmarking rights nevertheless permit aggregated, anonymized customer-data use, and the third-party training restriction contains a deidentified or aggregated exception. Spend Agent also describes learning from billing patterns and enforcement history. File sharing inherits access from vendor records; vendor users receive selected shared files and cannot access internal AI configuration or audit logs. The contract provides advance notice of compelled disclosure when lawful and a post-termination export window. Privilege preservation, model-provider retention and segregation of derived learning across customers remain unspecified.
The case management AI statement says client files and case records do not train underlying models and remain within the firm's workspace. The EULA limits access and use to providing or maintaining the service, improving the user's experience and legal compliance, but does not expressly prohibit training. It also covers vocational information such as cases, tasks and attached material. Restricted counsel access and role permissions limit who can reach a case, while public guidance discusses privilege and ethical walls. The model processors, their retention terms and a technical explanation of matter isolation during retrieval remain unspecified. The explicit training promise is a policy commitment.
UPL and Professional Responsibility Posture
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
The services agreement defines Onit as a technology provider rather than a law firm and disclaims legal advice, recommendations and representation. Customers remain responsible for reviewing AI output and obtaining professional advice where needed. Spend Agent's Sentry setting supplies a human approval route, while Auto mode can make invoice decisions under configured guidance. AI features are restricted to legitimate internal business use and cannot be used for high-risk decisions materially affecting individuals. The intended buyer is a corporate legal department with legal operations and finance participants. The terms do not identify licensing jurisdictions, assign every automated billing decision to a licensed lawyer or map configuration choices to specific professional supervision duties.
Provakil presents AI as assistance within legal operations and contract review, with lawyers responsible for checking generated material. Its legal operations guide engages ABA Formal Opinion 512 and keeps the verification duty with the person signing a filing. The EULA cautions subscribers to corroborate public information independently before relying on it. Case summaries and translations support preparation while practitioners still decide how to use the result. The services are available for lawful professional and educational purposes. An authorized user boundary for each AI feature, and jurisdiction specific supervision requirements across India, the Gulf and Southeast Asia, are not specified.
AI Governance and Bias Disclosure
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
Spend Agent has operational controls for review modes, vendor enablement, guideline selection and action logs, and Onit's AI contract identifies biased and inaccurate output as risks. Its marketing also makes EU AI Act and ISO 42001 compliance claims. A current model-governance framework connecting those claims to release approval, accountable owners and ongoing subgroup testing is not specified. Onit's AI Center of Excellence conducted the 2025 invoice benchmark with explicit consent and anonymized data, but it deliberately tested general-purpose models rather than specialized legal products. That research does not establish the release controls or bias results of today's Spend Agent across firms, currencies, billing practices and matter types.
Contract Intelligence follows stated principles of accountability, explainability and human review. Redlines and risk flags are traceable, and user defined playbooks shape clause analysis. Provakil says models are regularly validated and fine tuned, without specifying the testing method or results. Its October 16 governance guide concerns the customer's own program: recording AI suppliers, tracking complaints as matters and preparing board reports from spend and task data. It does not identify the person responsible for Provakil's models or their release criteria. Testing frequency, failure thresholds and findings about uneven results across languages or contract types remain unspecified.
AI Safety and Data Stewardship
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
Onit's data processing addendum covers personnel confidentiality, differentiated access, secure transfers, logging, audits and prompt notification of a personal-data breach. Its subprocessor register identifies service providers and countries by product, with a subscription-based notice process and an opportunity to object to new providers. Customer records remain available for export for 30 days after termination before deletion under the current retention policy, subject to backup and lawful recordkeeping exceptions. OnitX also supports encryption of selected fields and incoming email content, with historical-data conversion handled separately. Precise purge periods for all AI inputs, derived data and provider logs are unspecified, and the breach commitment has no universal deadline in hours.
The EULA limits access to service delivery, improvement of the user's experience and legal requirements. It also limits the data a subscriber can receive after service ends to specified case identifiers. Product controls include encrypted storage, role permissions, restricted outside counsel access and activity trails. A case repository holds petitions, orders and other uploaded material, while AI produces summaries and contract analysis within those workflows. No application deletion deadline, backup expiration, model processor list or incident notification procedure is stated. The separate privacy policy covers website visitors and correspondence, without setting a retention period for client matter content or AI conversations.
AI Liability and Recourse
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
The subscription warranty promises functional conformity and bars material reductions during the term. Customers must report a discovered performance failure within 30 days, after which the remedy process can provide correction or termination with an unused-fee refund. Onit defends qualifying intellectual-property claims involving the service, excluding customer data, certain combinations, customer designs and continued use after a stop notice. The ordinary cap is 12 months of fees under the applicable order, with indemnification obligations excluded. AI output is supplied as is, and standard AI uptime commitments do not apply unless the order adds them. No specific warranty pays for an incorrect Spend Agent adjustment or guarantees the validity of its billing judgment.
Provakil's subscriber EULA provides the service as is, disclaims accuracy and other warranties, and excludes indirect, special, incidental and consequential losses. It expressly places reliance on third party public information with the subscriber and asks that discrepancies be reported. The indemnity runs from the subscriber to Provakil for misuse, overuse or contractual breach. No reciprocal vendor indemnity, AI output warranty, stated insurance protection or defined remedy for an incorrect summary is included. Subscription fees are described as nonrefundable, while a separate cancellation page invites requests about remaining monthly or annual periods. Neither page promises compensation for a legal error.
Practice Systems Integration Depth
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
Accounts-payable integration sends fully approved invoices out as payment instructions and imports payment confirmations. The standard SFTP approach exchanges separate invoices-to-pay and invoices-paid files; APIs can operate in either direction. Integrations must use the Invoice Payment child application rather than directly changing invoice records, preserving synchronized payment phases. The REST API supports records, attachments, approvals, history and actions, using a designated API user and defined request limits. Document-system connections include iManage, NetDocuments and SharePoint. OnitX business rules can also map Unity request and RFP events into records and send approved updates back. The separate MCP beta provides read-only external-assistant access, with writes still planned.
Microsoft 365 supports contract editing in Word, while calendar connections send hearing dates, deadlines and reminders to Google Calendar, Outlook and iOS calendars. Provakil also connects to SAP, Oracle and Salesforce in the enterprise workflow. Legal collections accepts borrower data from banking and debt systems, and the litigation service uses secure APIs and configurable connectors for enterprise repositories. Signing and stamping integrations complete the contract process, and identity connections support user access. The customer configures workflows during onboarding and can migrate historical files. Field mappings, synchronization directions and a public configuration procedure for each enterprise connection are not specified.
Deployment Model and Data Residency
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
OnitX's core legal-management platform uses AWS hosting with a customer choice of the United States or Germany; the latter has backups in Ireland. Counsel Exchange is hosted in the United States regardless of that core choice, and Affinda performs portal PDF conversion in Australia. The Spend Agent register assigns Google Cloud, including Gemini, to the United States and OpenAI to the United States and worldwide processing. Regional storage therefore does not imply that invoice inference stays in the chosen hosting country. The data processing addendum supplies contractual transfer arrangements. The exact tenancy architecture, customer-operated deployment options and a model-by-model regional restriction for every AI request remain unspecified.
Enterprise deployments can run in the cloud or on the customer's own servers. Contract management names AWS and Google Cloud as hosting providers and offers local storage and processing in India, the UAE, Saudi Arabia and Malaysia. The collections product also expressly supports on premises deployment for regulated institutions. They do not identify which AI provider executes each summary or redline, whether its processing always remains in the selected country, or where backups and support access occur. No public plan matrix maps deployment choices, isolation arrangements and regional availability to a subscription tier.
Security Certifications and Trust Center
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
Onit describes independent audits against SOC 2 Type 2 and other security standards, alongside encrypted transport, daily backups, restricted access and monitoring. Its data processing addendum provides an annual audit right and permits recent assurance reports to satisfy that review in specified circumstances. OnitX adds field-level encryption for selected text and incoming email, excluding encrypted fields from retrieval indexes; existing historical content requires a separate conversion process. The trust center offers assurance material through an access request. An OnitX-specific auditor, coverage period and full system boundary remain unspecified in the accessible assurance details. Neither a provider's certification nor a compliance claim for AI establishes the scope of Onit's own examination.
Contract Intelligence is described as ISO 27001 and SOC 2 Type II certified, and the broader contract product also names ISO 9001:2015. The litigation FAQ describes SOC 2 aligned controls alongside ISO 27001 and GDPR principles. The AI product page adds regular third party audits, role permissions and data isolation. An auditor, current certificate period, covered entity schedule and SOC examination period are not supplied with these claims. The navigation includes an AI Trust Center label without a reachable collection of reports. The available material therefore leaves the certification scope and period for each product unresolved.
Model Supply Chain Disclosure
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
Spend Agent's provider register identifies Google Cloud, including Gemini, in the United States and OpenAI with United States and worldwide processing. AWS hosts the underlying Onit platform, while separate services handle portal PDFs, email and support. The current production model versions and allocation of invoice tasks between providers are not specified. Six named models appear in the 2025 Better Bill GPT experiment, but that research explicitly excludes specialized legal products and does not identify today's deployed configuration. Customers can subscribe to subprocessor notices and object to a new provider. Onit separately reserves the ability to replace models or prompts without materially reducing core functionality, leaving version-level notification unresolved.
The AI explanation identifies large language models operating over court feeds, matter repositories and contract content. Contract Intelligence adds supervised learning and company playbooks to clause recognition and extraction, while the in case chatbot limits its subject matter to the associated case record. These descriptions explain the intended source material but do not name the model families, model providers or hosting route for inference. AWS and Google Cloud are named for application hosting without being assigned particular AI workloads. A promise not to train on case content likewise leaves provider retention unspecified. No model substitution notice, version history or customer model selection is described.
Commercial Transparency
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
OnitX ELM and the optional Spend Agent module require a quote. Standard product limits include a 250 MB attachment maximum, an 80,000-row dashboard export limit and 1 TB of storage across production, development and test, with additional storage at $600 per month per TB. Annual Legal Vendor Billing charges can apply to firms submitting invoices through the portal, based on annual invoice activity; those payments can affect the customer's subscription terms. Implementation and additional configuration training are governed by statements of work. Default renewals require 45 days' notice, and sustained excessive AI use can incur overages after warning. The core license rate, setup amount and included agent allowance remain unspecified.
Provakil sells access through subscriptions and directs pricing inquiries to its sales team. The cancellation policy distinguishes monthly and annual subscriptions, while the EULA refers to a separate schedule of plans and rates. Case Management serves individual advocates, and Practice Management adds client handling, billing, team work and dashboards. Enterprise deployments combine selected litigation, contract and collections functions. No recurring price, per user or per matter unit, minimum commitment or implementation charge accompanies these options. The EULA permits additional fees for features introduced after a subscription begins. It treats subscription fees as nonrefundable, while the cancellation policy invites requests about unused periods.
Firm and Practice Coverage
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
OnitX ELM targets corporate legal and legal operations teams managing matters, outside counsel, invoices, budgets and reporting. Its configurable approval phases and financial integrations fit departments with established workflows and multiple participants. Counsel Exchange gives submitting firms access to assigned matters, invoices, budgets, timekeepers and rate requests, rather than supplying a firm's complete practice-management system. Onit markets across agriculture, pharmaceuticals, finance, insurance, technology and government, with regional hosting and international service teams. Matter administration can span practice areas, while Spend Agent concentrates on billing compliance against supplied guidelines and engagement terms. Exact language coverage, jurisdiction-specific billing boundaries and supported currency combinations for the AI module remain unspecified.
Corporate legal departments use Provakil for litigation, outside counsel oversight, contracts and intellectual property, while banks and nonbank lenders have dedicated legal recovery workflows. Indian court connectivity spans the Supreme Court, High Courts, district courts and specialist tribunals. Practice and Case Management editions also serve law firms and individual advocates. The vendor names government and public sector deployments and offers enterprise hosting across India, the Gulf and Southeast Asia. Localized notices and signing support serve regional workflows. Coverage does not establish every country's litigation connectivity or professional requirements, and an exhaustive list of unsupported jurisdictions and deployment sizes is not provided.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
Onit's agreement permits aggregated and anonymized customer-data use for product improvement, analysis and benchmarking. Its restriction on third-party models training for others contains an exception for deidentified or aggregated input needed to operate the service. Spend Agent also describes learning from historical billing patterns, firm performance and enforcement history. These provisions do not amount to an unconditional ban on training or improvement use.
Customer ownership and confidentiality duties remain, but no standard Spend Agent opt-out or complete explanation of how derived learning is separated between customers is specified.
The case management AI policy says client files and case records do not train underlying models, and account data stays within the firm's workspace. The EULA permits use to provide or maintain the service, improve the user's experience and comply with legal requirements. It contains no express training prohibition. Contract Intelligence also learns from millions of contract data points, whose origin is not identified.
Its supervised learning and playbooks support extraction and clause analysis without specifying whether subscriber contracts enter a shared training set.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Spend Agent stores commentary and review outcomes in Onit's systems, with action history available for audit. The agreement gives customers 30 days after termination to export their records, then applies the current retention policy rather than a fixed deletion deadline. Backup copies and legally required or reasonable internal records may remain under continuing confidentiality restrictions. AI output need not be retained or recreated unless the feature expressly supports it and retention is enabled.
No uniform duration is specified for invoice-processing prompts, provider logs, derived learning or every retained backup.
Case repositories retain uploaded files, hearing history and contract material used by the AI features. Ending a subscription removes the subscriber's access to the service. The EULA limits the subscriber's data return entitlement to forums, case types, case numbers, years and file numbers entered into the account. No duration is stated for prompts, generated summaries, contract analysis or backup copies. Website visitor information and correspondence have separate retention language. A customer setting for AI history or zero retention is not described.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
OnitX file-sharing permissions inherit from the vendor record for internal users, while external users with permission to invoice the client can access the files shared through that relationship. This general sharing facility is distinct from internal matter files. Spend Agent restricts external vendors from AI configuration and internal audit logs, and its review context uses supplied client, vendor and matter-vendor guidance.
The MCP beta executes as the authenticated Onit user. These are native Onit permissions; automatic enforcement of an external document-management system's ethical walls during every AI operation is not specified.
An in case chatbot works from the files, orders, hearing records and history attached to that matter. Account data stays in the firm's workspace, and external counsel receive restricted access to submit updates without entering the entire system. Case owners can share files and tasks with internal stakeholders through role permissions. The technical rules for enforcing matter permissions when the model retrieves content, including how access changes affect indexed material, are not specified. The platform does not describe inheriting an external document system's permissions.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Onit's confidentiality clause requires reasonable advance notice of compelled disclosure when legally permitted, plus cooperation at the customer's expense to seek confidential treatment. The obligation covers Customer Data because the agreement classifies it as confidential. The data processing addendum separately requires prompt notice of government or third-party requests concerning personal data unless prohibited.
That contractual route can give the customer an opportunity to respond before production. The broader privacy policy's discretionary language does not replace it. Request counts, challenge outcomes and legally prohibited notification rates remain unspecified.
Section 17.2 of the subscriber agreement allows use of vocational information to comply with law or a binding government order. The following sentence says information will not be disclosed to a government or third party, without reconciling those provisions. The agreement contains no customer notice commitment for a compulsory demand. A separate website privacy policy allows disclosure when law mandates it. No request transparency report or procedure for notifying a subscriber after a legal prohibition ends accompanies the service terms.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Spend Agent's governing material consists of customer-supplied billing guidelines, invoices and engagement terms. OnitX can attach guidance at the client, vendor and matter-vendor levels, and the current matter-vendor upload permits one combined file. These business records determine financial compliance rather than establish the current law of a jurisdiction. Customer terms require rights to submitted content and permit certain improvement uses.
No primary-law inventory, publisher licensing schedule or legal-source update cadence is specified for this workflow. The historical invoice benchmark is a separate research dataset, not a research corpus supplied to customers.
Court tracking receives causelists, orders and status updates from the Supreme Court of India, High Courts, district courts and tribunals such as NCLT, NCLAT and DRT. IP India supplies registry information for intellectual property tracking. The EULA describes aggregation from multiple public sources and reserves rights in Provakil's aggregation. A license or public domain basis for each underlying source is not stated.
Uploaded contracts and case files provide additional material for summaries and analysis within the customer's own workspace.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Spend Agent associates an invoice issue with billing guidance and the affected charge. Reviewers can inspect the rule, engagement context, proposed adjustment and financial effect. That source relationship supports assessment of the bill, but it does not determine whether a judicial authority remains good law. No case-treatment service, reversal flag or subsequent-history procedure is specified for the module. An OnitX matter record may concern litigation without turning the invoice reviewer into a legal-research system. Authority verification remains a separate step when a team prepares legal advice or court submissions.
The intellectual property workspace shows timestamped histories for filings, renewals, oppositions and ownership changes. IP India integration supplies registry updates, while expiry and renewal alerts help users follow changes affecting registered assets. This status history concerns patents and trademarks. The litigation workspace separately follows hearings and case progress, without checking how later decisions treat a cited judgment.
Public information remains subject to the EULA's independent verification requirement. No commercial case law citator or registry update service level is named.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
The Spend Agent workflow includes a missing-guidelines path: when required billing guidance has not been uploaded, the invoice is returned with a message that processing did not occur. Sentry mode also presents potential violations to a reviewer, who can dispute, approve or adjust the invoice instead of allowing an automatic decision. These controls address missing governing input and review responsibility. They do not specify calibrated confidence levels, a universal low-confidence abstention threshold or an explanation of uncertainty for every ambiguous billing narrative. An ordinary billing-rule rejection is distinct from an AI refusal.
Contract Intelligence assigns severity and risk scores to clauses and obligations, while litigation AI summarizes case progress and next actions. Contract scores concern legal or commercial exposure, rather than confidence in the model's answer. Playbooks guide proposed wording, and approval workflows give people a review stage. No explicit abstention message or unsupported answer path is described for the case chatbot, translation or contract analysis.
The stated 99.5% court tracking accuracy concerns operational updates, without measuring model confidence or the reliability of generated recommendations.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
The reviewed court-decision collection contains no case identifying OnitX ELM or Spend Agent as the source of fabricated legal authority. Loose matches to the owner's name appeared in unrelated text rather than as tool identifications. Spend Agent's normal output is a billing explanation and adjustment, not a brief or judicial citation. That limits what a legal-citation incident would measure, without proving the absence of unsupported billing conclusions.
Onit's AI agreement warns about inaccuracies and hallucinations, and incidents involving unnamed systems cannot reliably be assigned to this product.
The AI Hallucination Cases database maintained by Damien Charlotin has no entry naming Provakil or PVKL in connection with fabricated legal citations. The product summarizes orders and petitions, translates court material and answers questions from a matter's files and history. Contract Intelligence also extracts fields and suggests clause changes. Its subscriber agreement requires users to check public information independently before relying on it, and its legal operations guidance keeps verification responsibility with the lawyer.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Onit's community commentary addresses verification duties, candor to courts, unauthorized AI adoption and cautions from Britain's Bar Standards Board about agentic systems. The services agreement separately disclaims legal advice and makes customers responsible for reviewing output. This provides general professional-responsibility context, rather than a mapping of Spend Agent controls to particular ethics opinions. No jurisdictional matrix connects automatic invoice decisions, confidentiality, customer-data improvement rights and client consent with the requirements of named bar opinions. Sentry and Auto settings describe operational choices without certifying compliance with a lawyer's individual duties.
Provakil's legal operations guide names ABA Formal Opinion 512 when discussing confidentiality, privilege and verification of generated work. It places the verification duty with the person signing a filing and treats model output as draft material. The guide describes role permissions, ethical walls, data residency and audit trails in the legal workspace. It does not map individual product features to professional conduct requirements across multiple jurisdictions. The separate EULA also requires subscribers to corroborate public information independently before relying on it.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
OnitX ELM manages legal spending from invoice submission through approvals and payment handoff. Spend Agent reviews the actual charges against supplied billing terms, explains violations and can propose or apply monetary adjustments. Reviewers can distinguish agent, user and conventional-rule adjustments, and firms can receive reasons and appeal changed fees. The financial record is therefore the work product of the system.
It does not establish how much generative AI outside counsel used while performing each billed task, or automatically convert that firm's time savings into an appropriate customer charge.
Enterprise legal departments track outside counsel budgets, court fees and invoices in Provakil. Stage based billing ties payments to litigation phases, and approval workflows control disbursements. The collections edition similarly lets lenders monitor advocate costs alongside recovery activity. Practice Management also supports law firms' own billing and advertises administrative time savings. No per task record connects those firms' AI use to time actually worked, and no guidance explains how saved time or software charges should appear on their clients' bills.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
The public Spend Agent datasheet explains review modes, role-based permissions, vendor visibility and exportable audit history. Combined with the product-specific subprocessor register, AI contract and data processing addendum, it gives a client information about the service's providers, operating controls and terms before an agreement is signed. That material can support a firm's response to an AI diligence request.
It does not itself obtain client consent or establish compliance with every outside counsel clause. Enforcing the department's billing rules is a separate function from satisfying restrictions on AI use.
Application hosting uses AWS and Google Cloud, with regional deployment choices and role based access. The case AI policy prohibits training on client files, and the subscriber agreement restricts the purposes for which vocational information can be used. Neither identifies the models or processors receiving content during inference. A current subprocessor register, model change procedure and client consent or notification pack are not available. The publicly named infrastructure providers do not give a complete account of the AI processing chain.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Spend Agent action logs carry a timestamp, vendor, outcome and review mode, with export by status, date or vendor. Reviewers can inspect invoice lines and explanations, distinguish adjustment sources and preserve changes through the invoice workflow. This supports reconstruction of financial review decisions and identifies AI involvement in the billing record. The export does not promise a model version, complete retrieved-source set and human verification for each generated passage.
A filing disclosure or a regulator's request concerning later use of those conclusions would require additional information beyond the operational audit trail.
AI assisted output is logged in audit trails and limited to authorized users. Contract Intelligence retains traceable redlines and risk flags, and litigation records keep the court orders and uploaded files behind case work. These features let a team revisit activity alongside source material. No per file export combines the model version, retrieved sources and human verification, and no court disclosure form is generated from the recorded events. The model identity itself is also unspecified in the product material.
Which one fits
Choose OnitX ELM if
- The priority is explaining suspected billing violations during invoice approval.
- Legal operations needs configurable matter approvals and portfolio cost dashboards.
- Outside counsel should receive invoice status and feedback through a dedicated portal.
Choose Provakil if
- AI output is primarily needed for orders, petitions and contract clauses.
- Spend controls need to sit alongside automated Indian court tracking.
- The department manages bulk legal notices as well as case and contract work.
In summary
OnitX ELM
OnitX ELM combines configurable matter processes with budgets, accruals and outside counsel administration. Spend Agent is an additional invoice review module. Findings include explanations that help staff address charges before completing approval.
Provakil
Provakil keeps case activity, legal expenses and counsel updates together. Its AI summarizes orders, reviews progress and proposes contract wording. The collections edition extends the same operational scope to notices and lender recovery.
Questions buyers ask
Is Spend Agent part of every OnitX configuration?
Onit presents Spend Agent as an additional module for the ELM platform. Core matter and financial administration remain distinct from that AI review capability. From the AI Legal Index, based on each vendor's own published materials as of October 16, 2026. No vendor pays for placement.
Does Provakil also track counsel costs?
Its enterprise product follows budgets, court fees and advocate invoices, including stage based billing. The described capability does not establish the same explanatory invoice analysis as Spend Agent. From the AI Legal Index, based on each vendor's own published materials as of October 16, 2026. No vendor pays for placement.
What do users review in the AI workflows?
OnitX reviewers receive invoice findings with reasons. Provakil users work with generated case summaries and traceable contract suggestions in the context of their files and playbooks. From the AI Legal Index, based on each vendor's own published materials as of October 16, 2026. No vendor pays for placement.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything on it comes from public material on the dates shown. How the index grades.
OnitX ELM is the configurable Onit platform covered here. Separate products in Onit's portfolio are outside this comparison.