U

Unity ELM

Marketed as AI native, the product is structurally an ELM system with two AI capabilities inside it. The operational core is conventional enterprise legal management: matter records, spend tracking, budgets, approvals and billing guideline configuration, which the vendor calls the foundations a team starts with. The AI sits on top in two named pieces, Spend Agent for automated invoice review and Ask Unity for querying matter and spend data in plain language.

Remove both and a working system of record remains. The vendor presents Unity ELM as the automate layer, with Ask Unity extending it.

Each Spend Agent finding is tied to the billing guideline it rests on, and the one accuracy figure has no stated method. Onit's Spend Agent page says the agent connects each finding to the applicable billing guidance, calculates a recommended adjustment and takes the reviewer to the affected invoice line with an explanation. The same page claims more than 92 percent accuracy at invoice level, against 72 percent for experienced lawyers, without saying who measured it, on how many invoices or over what period.

Section 12.3 of the services agreement says output may be incomplete, inaccurate or biased and cannot be guaranteed free of hallucinations. Ask Unity answers from the department's own data, with no retrieval method described. The product cites no legal authority, so the risk is a missed charge, or a valid one wrongly disputed.

An autonomous function runs, and almost nothing is published about how it is supervised. Spend Agent is said to identify invoice errors, discrepancies and departures from billing guidelines automatically, and to work with legal providers to adjust invoices with minimal human intervention. That is an agent transacting with a department's outside counsel. No page read says what runs unattended and what a person approves, what thresholds trigger escalation, where a reviewer sees the agent's decisions, or how a disputed adjustment is reversed.

The agreement only puts review and reliance on the customer. Ask Unity is a query tool a user starts, so the question arises less sharply there. The phrase minimal human intervention leaves any approval step undescribed.

No named customer of this product and no figure attached to it are published. The March 2026 release says several clients have implemented Spend Agent and more are preparing to deploy, with no organization, deployment date or measured result named. Onit's figure of more than 3,000 customers covers its whole portfolio of brands, not this product. A case studies index on the site was not opened.

A real confidentiality regime sits beside a training position that stops short. Clause 8 of the services agreement makes confidentiality mutual and designates customer data as the customer's confidential information, and clause 5.1 leaves the customer owning its data. The AI terms thin out: outside model providers are restricted from training on customer input for others' benefit, with an exception, and clause 5.2 reserves aggregated anonymized customer data for product improvement and benchmarking.

Nothing addresses privilege or work product, and no segregation by matter or tenant is documented. Clause 4.10 offers only commercially reasonable safeguards.

A clear position on advice is in the agreement, with a limit on high risk use and no treatment of supervision or jurisdiction. Section 4.14 of the services agreement, last revised 19 August 2026, headed No Legal Advice, says Onit is not a law firm and does not offer legal advice. Section 12.7(d) forbids using the AI features in a way that makes, or substantially contributes to, a decision with a material legal or similar effect on an individual, and section 12.3 makes the customer responsible for reviewing output.

The users are lawyers and legal operations staff in corporate legal departments. Nothing says who in the department should approve an adjustment Spend Agent proposes to a law firm's bill, how a lawyer's supervision duties apply, or which jurisdictions are covered.

No governance position is published. No responsible AI page, principles, accountable owner, testing regime before release or ISO 42001 or equivalent is published across the release announcements, the Ask Unity material, the security page and the services agreement. Bias is mentioned once, as a disclaimer: the agreement warns that output may be biased, which places the risk on the customer rather than describing anything the vendor does about it. The agreement also lets Onit change models or prompts at will, a change right with no governance statement beside it.

A subprocessor list with a change process and a breach commitment are published, and a retention period is not. The subprocessor list names the AI providers with their locations, alongside AWS hosting and Onit's own affiliates for support, and customers who subscribe to updates can object to new subprocessors. The data processing addendum, revised 20 April 2026, commits Onit to notify the customer promptly of a security breach, with its nature and the data affected, and allows one audit a year or reliance on SSAE 18 reports.

No breach notice period in hours or days is stated, and customer data is deleted after termination under a retention policy that is not published.

Onit's services agreement publishes a cap, an uncapped infringement indemnity and a performance warranty, and then takes the AI outside them. The agreement, last revised 19 August 2026, caps each party's liability in section 10.2 at the fees paid or payable in the twelve months before the event, and section 10.3 lifts indemnities out of that cap. Section 11.1 commits Onit to defend the customer against claims that the subscription services infringe intellectual property rights, and section 9.2 warrants that the services will operate substantially as documented.

Section 12.5 then provides AI features and their output as is, with every warranty disclaimed, and section 12.8 says no service level applies to AI features unless an order says so. A wrong adjustment from Spend Agent, or a wrong answer from Ask Unity, is therefore the department's loss.

Integration is asserted at platform level and not documented at product level. The release material makes integration across Onit products the headline, with products, agents and workflows, including LSR and the forthcoming Unity RFP, working together on the unified platform without extra configuration, and Spend Agent built into Unity ELM. Those are internal Onit connections rather than links to other practice systems.

On outside systems, clause 4.4 says Onit APIs exist and are licensed on fees and restrictions in an Order, and clause 4.13 disclaims third party applications entirely. No named integration with an electronic billing feed, matter management, document management or finance system, and no public developer documentation, is published.

Hosting and AI processing locations are published in the subprocessor list, and the tenancy model is not stated. The list, dated 17 August 2026, shows AWS hosting in the United States for all Onit platforms, with a United States or Germany option for its legal management and contract platforms. Google Gemini and OpenAI serve Spend Agent and the Unity platforms in the United States and worldwide, so invoice content sent to the AI is processed in the United States.

The data processing addendum covers transfers under standard contractual clauses and the UK addendum. The security page describes an enterprise class hosting environment without saying whether customers share infrastructure, and no private cloud or on premises option is described. Whether Unity ELM itself can be hosted in Germany is not stated.

Certifications are named by standard, with none of the particulars. The security page says Onit audits to SSAE 18 SOC 2 Type 2, SOC 1 Type 2 and HIPAA standards, verified by an independent auditing firm, and that physical security is audited separately, with the data center described as SSAE 18 Type II SOC II. A trust center at trust.onit.com has a controls section. No auditor, trust services criteria, scope, coverage period, report date or route to obtain a report is given in the agreement or on the page.

The security page describes 256 bit SSL and data center controls and says nothing about the model layer. The trust center was not opened.

The model providers are named for this product, with a change notice process, and the agreement reserves model changes without notice. Onit's subprocessor list of 17 August 2026 names Google Cloud, including Gemini, and OpenAI for Spend Agent and the Unity platforms, and section 12.9 of the services agreement confirms that third party model providers are used. Customers who subscribe to list updates can object to a new provider.

Section 12.4 lets Onit improve, replace or deprecate models or prompts without notice, as long as core functionality is not materially reduced. No model or version is named, and nothing commits to notice of a change of model within a provider.

No pricing information is published at any level. There is no pricing page, every commercial route is a demo booking, and no tier, package, unit or figure is published. The agreement shows that a charging structure exists without disclosing it. Clause 3.1 refers to a Subscription Fee and Professional Fee set in an Order and makes all fees subject to annual increase. Clause 4.7 describes usage tiers, with automatic moves up to the next tier and invoicing for the difference when usage exceeds a tier maximum, and clause 12.6 permits overage fees at current rates for sustained AI overuse after notice. What any tier costs cannot be learned.

The buyer is identified clearly, and the boundaries are not. Unity ELM is sold to corporate legal teams to manage matters, outside counsel spend, invoices, budgets and approvals. Onit publishes role pages for legal operations, general counsel and contract managers, and industry pages for sectors such as pharmaceuticals, financial institutions and insurance. No department or company size is stated, nothing addresses law firm use, and no segment or jurisdiction is named as out of scope.

The one limit is contractual: the AI features may not be used in decisions with significant legal effects on individuals.