L

Laurel

The models are the product. Laurel's home and legal pages describe agents that automatically capture chargeable and non chargeable digital work across the applications a timekeeper uses, group the activity into time entries, write the billing narratives, and check entries against each client's billing rules before the bill goes out. The privacy policy describes the mechanism: captured activity is transformed into abstract mathematical representations so that self hosted AI models can suggest work clusters.

The further products, Signal for project profitability and Work Studio for mapping how work is split between people and AI, run on the same captured data. Remove the models and there is no product left to sell, because the manual time entry sheet that remains is what every practice management system already provides and what Laurel exists to replace. The legal page's promise to capture up to 30 minutes of missed billable time a day depends entirely on the capture and clustering working, and the rebrand post describes the rebuild around improved AI accuracy. What a firm buys is the capture model.

Accuracy is asserted without a measurement. The product does not cite law; it asserts how long a timekeeper worked on what, and what the narrative for that work should say. The rebrand post says the rebuilt platform's AI accuracy is four times better than before, with no baseline, test set, metric or date. The legal page quotes a firm chair saying that if someone challenged her time she could say it is literally the AI saying how long she worked, which presents the capture as evidence without any published error rate.

Entries are built from captured activity a timekeeper can see in Timeline and Timesheet views, per the Tonkon Torp case study, so a narrative can be traced to the activity behind it. Not published is how often work is assigned to the wrong matter, how often the narrative misdescribes the work, or how accurate the billing rule checks are.

The review point is described: timekeepers see captured work and assign it before anything is released. The Tonkon Torp case study says Laurel shows captured activities in Timeline and Timesheet views, where timekeepers review, group and assign entries to client matters before release. The legal page says the billing agents catch issues against each client's billing rules before they reach the bill. That is a real review surface with a defined point of release, keeping a person between the capture and the client.

Not published is the control structure around it: whether an entry can be released without being opened, whether delegates or administrators can release for a timekeeper, what happens when an entry the agents questioned is released anyway, and whether any step runs on a schedule without a person acting. The agents capture continuously and unattended by design, so oversight sits entirely at the release step.

Named law firms are published with figures, without dates or methods. The Tonkon Torp case study names the Portland firm and its managing partner, Kurt Ruttum. It reports 21 more billable minutes a day per timekeeper, about $20,000 more annual revenue per timekeeper, a 20 percent faster collection cycle and 18 percent more time entries a day. A partner is quoted on making an honest accounting to clients. The legal page quotes a firm chair, Cynthia Kuehl, and states results of a 4 to 11 percent profit increase, 2.88 more time entries a day and up to 30 minutes recovered daily, without attributing them to a firm.

Further stories cover an Am Law 200 firm, Obermayer, Lerners and the accounting firm GHJ, and Laurel's home page shows logos including EY, Aprio, Saul Ewing, Watson Farley & Williams and FBT Gibbons. None of the stories says when the figures were measured, over what period, against what baseline or for how many timekeepers.

The confidentiality position is stated in product terms, and the agreement that would make it binding is not published. The home and security pages say Laurel uses no public large language models, its AI is single tenant and firm specific, firms can bring their own encryption keys and revoke access, data can be stored in region, and data is encrypted in transit and at rest. The legal page adds that a firm's AI gets smarter while its data stays private.

The privacy policy (effective 14 August 2026) does not govern the desktop and mobile apps that capture work; processing through the apps is governed by agreements with business customers, none of which is published. So the captured record of what a lawyer worked on for which client, which can itself reveal privileged matters, is protected by commitments a firm cannot read before signing. Nothing published addresses privilege or work product, how long captured activity is kept, or what the firm specific model retains of one firm's work.

The review step is described in a customer story, and nothing connects the product to a lawyer's billing obligations. The Tonkon Torp case study says timekeepers review and assign captured entries to matters before release, which places the decision about what is billed with the lawyer. That is the position that matters for a timekeeping product, and it sits in marketing rather than in published terms, product documentation or anything a firm signs.

The legal page goes the other way in one quotation, presenting the AI's record as the answer to a client who challenges the time. Nothing published addresses a lawyer's duty to bill only for work actually performed, how generated narratives are checked before they describe work to a client, how AI assisted drafting time is recorded, or bar opinions on reasonable fees for AI assisted work. The audience is clear: timekeepers and firm leaders at law and accounting firms. The rules that govern what goes on a client's bill are left to the firm.

Laurel holds an AI management system certification, and publishes no testing results or named owner. Laurel's press release announces ISO 42001 certification, the international standard for AI management systems, and says every AI risk is documented, tracked and managed through a defined process and that accountability for AI decisions is clear. The security page shows ISO 42001 beside SOC 2 Type 2. A certified management system means an external auditor has examined how Laurel identifies, assesses and controls AI risk.

Not published are the certification body, the date or scope of the certificate, who inside Laurel is accountable, what is tested before a model change ships, or any finding on how capture and narrative accuracy vary across practice areas, roles, languages or applications. The privacy policy's description of turning activity into mathematical representations for self hosted models is architecture, not governance. A firm can point a client to an ISO 42001 certificate and has nothing to show about results.

Encryption, key control and storage in region are published, and what happens to captured activity over time is not. The security page says data is encrypted in transit and at rest and that firms can bring their own keys and revoke access at any time. It also says storage in region keeps data within a geographic boundary, and that single sign on is supported. The product records what every timekeeper does across their applications.

Nothing published says how long that record is kept, who at Laurel can see it, which subprocessors handle it or how a breach is reported. The privacy policy (effective 14 August 2026) covers the website and excludes the apps, whose processing is governed by unpublished customer agreements, and it keeps data as long as necessary for its purposes. No subprocessor list or incident practice is published.

Nothing is published on who bears the loss when a captured entry or a generated narrative is wrong. Laurel's sitemap lists no terms of service page, and the usual address for one on laurel.ai does not exist. No customer agreement, terms of service, warranty, indemnity, limitation of liability or insurance statement is published. The privacy policy (effective 14 August 2026) says the apps are governed by agreements with business customers that are not published.

A wrong matter assignment or an inflated narrative that reaches a client bill exposes the firm to a fee dispute or a complaint. The legal page suggests the AI's record answers a challenge to a lawyer's time, which raises the stakes of an error. Whatever allocation of that risk Laurel offers sits in a contract a firm sees only in a sales process.

The integrations a law firm needs are named on both sides of the product, without documentation. The legal page lists practice management and billing systems the entries flow into: Aderant, Clio, Elite, ProLaw, SOS, SurePoint, Tabs3 and Juris, with a flat file option for others. On the capture side it lists the applications Laurel reads from: Outlook and the calendar, Microsoft Teams, Zoom, Edge, Chrome, Gmail, Adobe, Workday and the document systems NetDocuments, iManage, Box and eDocs.

Single sign on is supported. That covers the systems where a firm's time and documents already live. What is not published is how each billing integration works: whether entries are pushed or pulled, at what point, which fields map, how matter numbers are matched, and what happens when a matter is closed or an entry is rejected in the billing system. The help center has collections for firm administrators and IT teams. So a firm on Aderant or Elite can see that Laurel connects, not how.

The tenancy model of the AI is stated and residency is offered without the regions being named. Laurel's home page states single tenant AI with no public large language models, and its legal page describes firm specific models, so each firm's capture and narrative models are its own. The security page offers in region data storage to keep data within a specific geographic boundary, without naming the regions available or saying whether processing by the models happens in the same region as storage, and customer managed encryption keys.

The privacy policy, which covers the website, says information may be transferred to, processed and stored in the United States and other jurisdictions, with standard contractual clauses for transfers from the EU. The desktop and mobile apps that capture activity run on timekeepers' own machines, and nothing published says what is processed locally before data leaves the device. A firm in Europe or Canada can have its data kept in region, but the regions on offer and where the models run are not published.

SOC 2 Type 2 and ISO 42001 are stated, and no auditor, scope, date or report route is published. The home and security pages list SOC 2 Type 2 and ISO 42001 with GDPR, CCPA and HIPAA compliance, and the ISO 42001 press release announces that certification. No auditor or certification body, no report period or certificate date, no scope statement and no trust portal or route to request the SOC 2 report is published on the security page, the home page, the sitemap or the press release, and no ISO 27001 certification is claimed.

The security page describes controls in brief: encryption in transit and at rest, customer managed keys, in region storage and single sign on. The site also publishes a responsible disclosure policy for security researchers. The two standards are named, and no route to the evidence behind them is published.

The architecture is described and no provider or model is named. Laurel's home page says it uses no public large language models and that its AI is single tenant, and its legal page describes firm specific models. The privacy policy (effective 14 August 2026) says captured activity is turned into abstract mathematical representations so that self hosted AI models can suggest work clusters. So clustering runs on models Laurel hosts itself, and each firm's model is its own.

Not said is which models those are, whether they are Laurel's own or open models it hosts, whether narrative writing uses the same models, where they run, and whether any third party model is called. No subprocessor list naming model or hosting providers is published, and no commitment to notify customers when a model changes. A firm can tell a client that its time data does not go to a public chatbot, and not what does process it.

No pricing information is published at any level. The sitemap has no pricing page, and no price, plan, unit of charge, minimum commitment or implementation fee appears on the home, legal, accounting or security pages or the ROI calculator page. The routes to buying are a demo request and the ROI calculator, which estimates recovered revenue rather than cost. The rebrand post says implementation time fell from four weeks to one, without saying what implementation costs.

For a product sold on recovering billable time worth a stated amount per timekeeper, a firm can estimate the upside from Laurel's own figures and cannot set it against the price. Whether Laurel charges per timekeeper, per firm or by value recovered is not stated.

Who the product serves is described with substance across professions, and the boundaries within law are not stated. Laurel publishes pages for legal, accounting, global accounting and Big Four firms including PwC, Deloitte and KPMG. Its customer stories range from regional law firms such as Tonkon Torp in Portland and Obermayer to an Am Law 200 firm and international firms such as Watson Farley & Williams. Accounting firms such as GHJ, Aprio and EY also appear.

The listed practice management integrations, from Clio and Tabs3 to Aderant and Elite, span small and large firms. The rebrand post describes a move beyond legal timekeeping toward all knowledge work, and the named users are timekeepers, delegates, firm administrators and IT teams. Not stated is where it stops in law: whether flat fee, contingency and in house teams are served, which practice areas the narratives handle well, and which countries' billing conventions the rule checks support, beyond EU data transfer terms and customers on three continents.