The second wave of legal AI is beginning to focus on something different. Beyond helping firms perform existing work more quickly, AI makes it practical to capture and structure information that was previously too costly or burdensome to collect. As more legal work is transformed into standardized, analyzable data, firms gain the ability to identify patterns, measure performance, improve forecasting, support more informed pricing models, and generate insights that were previously unavailable. The opportunity extends beyond efficiency by creating new analytical capabilities built on richer and more comprehensive datasets (Bedford et al., 2025; Alshak & Grady, 2026; CLOC, 2019).
What Should Firms Actually Capture?
That shift raises an important question. If the second wave of legal AI is about creating richer and more analyzable datasets, what data should firms capture alongside time entries, and in what form? More importantly, what new insights become possible when legal work is consistently structured and classified?
For structured billing classification, the legal industry has commonly relied on coding standards. Among them, the Uniform Task-Based Management System (UTBMS) provides standardized task and activity code sets, while LEDES provides formats for exchanging e-billing and related data. UTBMS codes are commonly included in LEDES-formatted invoices when clients require coded billing. UTBMS gives firms a shared vocabulary for litigation, transactional, and counseling work, allowing legal spend to be compared across firms and matters. Applying UTBMS coding universally, rather than only when a client requires it, can provide one useful answer to the question of what firms should capture.
But it is not the only answer, and it may not be the most complete one. UTBMS code sets and LEDES data-exchange formats were developed principally to standardize legal billing information and facilitate its exchange and review between law firms and their clients (Thomson Reuters, 2025). Although firms may also use the resulting data for internal analysis, these standards were not designed as comprehensive operating models for understanding all aspects of a firm’s work.
A more complete answer is this: firms considering AI-assisted activity capture should evaluate whether it can support defined operational and strategic use cases as a byproduct of the work itself. They should also assess whether it can categorize that activity across a range of dimensions the firm considers operationally or strategically important, rather than limiting classification to the dimensions provided by a billing standard. UTBMS coding can, and in many cases should, be one of those dimensions. But it should not be the only one, and it should not define the limits of how a firm can use its data.
The Case for Passive Capture
“Passive capture” means work activity is observed and recorded automatically in the background rather than reconstructed later from memory or entered manually as a discrete step. That activity can include emails, documents, meetings, calls, calendar events, and other work performed through connected applications (Billables AI, 2025). AI-powered time capture systems already do this to generate draft time entries. Depending on the system and the firm’s configuration, some of the underlying activity data may be available for additional governed uses once a narrative and duration have been produced. That data is what makes richer categorization possible without materially increasing a timekeeper’s workload.
Passive capture should not be understood as unrestricted or invisible monitoring. Firms should define which activity sources may be accessed, what information may be retained, who may review it, and the business purposes for which it may be used. Attorneys and staff should also be informed about how the system operates and how captured activity data will be accessed, retained, reviewed, and used.
This matters because one persistent constraint on structured legal data has been the time and attention required to apply a taxonomy consistently. In practice, manual coding can add significant effort to time entry (Mitratech Staff, 2026). Timekeepers must locate the appropriate category, understand the classification structure, and apply it consistently. Because of that administrative burden, firms have often concentrated structured coding on matters for which clients require coded invoices, while much of their remaining work has not been categorized in the same way.
When categorization occurs automatically at or near the time the activity is generated rather than after the fact, that constraint can be substantially reduced. A firm may be able to obtain more comprehensive data without imposing an unreasonable administrative burden because AI may propose classifications based on available activity data, with timekeepers or designated reviewers confirming or correcting the results.
Beyond a Single Taxonomy: Categorizing on the Firm's Own Terms
Passive capture substantially reduces the administrative burden, but it does not, by itself, solve the flexibility problem. This is where relying solely on UTBMS coding falls short for many internal management purposes. UTBMS offers standardized code sets, but no individual code set captures every dimension a firm may need for internal management. LEDES, by contrast, is a data-exchange format rather than a taxonomy. A firm that wants to understand its own operations needs the ability to define categories that reflect its practice structure, staffing model, and strategic priorities, rather than only the categories included in a billing code set.
AI-powered activity capture makes this possible because AI may make it easier to map activity to multiple approved taxonomies rather than relying on only one external code set. The same underlying activity, whether an email, draft, call, or research session, can be tagged simultaneously using multiple classification schemes. These may include the firm’s internal taxonomy, client-specific billing requirements, matter-management rules, approved coding requirements, a client’s outside counsel guidelines, UTBMS where relevant, and other governed frameworks approved for the relevant use case. This approach potentially requires less manual effort, depending on system integration, accuracy, and the required level of review. UTBMS becomes one lens among several, useful for client-facing billing, rather than the sole structure within which the firm must organize its operational data.
Dissecting Work Activity Across Multiple Dimensions
Once activity is captured passively and classified flexibly, firms can analyze their work across dimensions that narrative time entries or a single billing code were never built to reveal, including:
● Type of work performed: The substantive nature of the task, such as drafting, review, negotiation, research, or strategic advice, independent of any particular coding standard.
● Source of the activity: Whether the work originated in email correspondence, meetings and calls, document drafting, or web-based research, giving firms visibility into how time is actually being spent, not just what it was spent on.
● Stage of the matter lifecycle: Whether the activity occurred during intake, discovery, negotiation, trial preparation, closing, or another phase, allowing firms to see how effort shifts as a matter progresses.
● Timekeeper role and experience level: Which categories of work are concentrated among partners, associates, paralegals, or other staff, supporting more informed staffing and leverage decisions.
● Client and matter type: How work patterns differ across industries, practice areas, and matter categories, enabling more precise benchmarking between genuinely comparable engagements.
● Any combination of the above: Because the underlying activity is tagged across several dimensions at once, firms can slice the same dataset differently depending on the question being asked, rather than being limited to whatever cross-section a single code set happens to support.
This provides a materially broader capability than universal UTBMS coding alone: a multidimensional, passively captured dataset shows not just what task category was billed, but what was done, how, at what stage of the matter, by whom, and for what kind of client and matter, all from the same underlying activity.
From Consistent Classification to Better Decisions
Metrics and analytics are most useful when information is collected consistently and comparably across the organization. Without some shared structure, meaningful analysis becomes difficult, and narrative time entries make this especially visible: two attorneys performing the same underlying task might describe it very differently, one writing "review correspondence," another writing "analyze client email regarding settlement strategy." Both descriptions are accurate, but they are difficult to compare at scale using text alone (CLOC, 2019).
A consistent classification layer, applied automatically across several dimensions at once, reduces that variation. It can group potentially comparable work, subject to validation and appropriate treatment of matter-specific differences regardless of how individual timekeepers describe it. This makes benchmarking possible by allowing firms to compare staffing models, task allocation, and effort levels across comparable engagements, identify best practices, and better understand the factors that influence matter cost (CLOC, 2019; Kelly, 2024).
What matters most, though, is not the classification itself but what it enables a firm to do (Wolters Kluwer, 2025). Applied at scale, the data can supplement estimates with observed effort, cost, and selected outcome indicators. Firms can use those observations to assess which matters may require different staffing models, which clients or practice areas are the best candidates for a given fee structure, and whether proposed alternative fee arrangements (AFAs) appropriately account for the expected work, required resources, and allocation of risk.
There is also an important timing advantage. In practice, once time entries have moved through billing and financial workflows, retrospective reclassification can become more difficult, less efficient, and less reliable than classifying the work when it is first captured. Capture and classification at or near the point of work facilitates richer classification when the underlying work activity is still fresh, detailed, and available for review.
Why This Matters for Alternative Fee Arrangements
The importance of this kind of data grows as legal pricing continues to evolve. Hourly billing remains dominant in many practice areas, but clients continue to seek greater predictability and value-based pricing. At the same time, the use of AFAs continues to expand as firms and clients look for ways to align fee structures with the scope of work, expected outcomes, risk allocation, and business objectives. Successful alternative pricing depends on data: firms need to understand how work is actually performed, how much effort specific kinds of tasks require, and what resources different types of matters consume (Kelly, 2024).
Classified activity data is an important component of AFA pricing, but it is not ordinarily sufficient by itself. A useful AFA pricing dataset typically combines activity and work-effort information with the anticipated scope and volume of the engagement, matter complexity and client requirements, staffing and cost assumptions, historical financial performance, budget and cycle-time results, risk and dependencies, scope changes, and client-defined measures of value or success where reasonably measurable.
Better data does not necessarily result in lower legal fees. Its primary value is to support pricing arrangements that more accurately reflect the anticipated scope of work, required expertise, resource demands, value delivered, and risks assumed, while giving firms and clients greater predictability and transparency.
A multidimensional, passively captured dataset is particularly well suited to this. Because activity is tagged by type of work, source, lifecycle stage, role, and client and matter type, firms can analyze thousands of activities across genuinely comparable matters and identify recurring patterns, typical effort levels for specific kinds of tasks, and underlying cost drivers that extend well beyond what a single billing code could reveal. Together with scope, complexity, cost, realization, margin, and risk data, this foundation can support a range of alternative pricing models, including fixed, recurring, and capped-fee structures, that are better aligned with expected work, value, resource requirements, and risk (Kelly, 2024; Alshak & Grady, 2026).
Client Service and Governance
Comprehensive, multidimensional data can also improve client service. Even when clients do not require coded invoices or submit bills through an e-billing platform, firms with structured internal data may be better equipped to answer questions about staffing, budgeting, matter progress, and efficiency. If a client later asks for more detailed reporting, including UTBMS-coded invoices, the underlying activity data may enable the use of existing UTBMS classifications and related reporting, subject to client requirements, technical feasibility, and appropriate validation (Mitratech Staff, 2026).
Firms should explain what activity data will be captured, how it will be reviewed and protected, and how it will be used so that attorneys and staff understand that the objective is to improve matter management, pricing, and client service rather than serve primarily as a tool for individual monitoring.
Firms should also assign responsibility for taxonomy maintenance, exception handling, version control, and periodic review to an appropriate cross-functional group so that classifications remain aligned with evolving practices, client requirements, and pricing strategies.
As with any AI-assisted classification, governance matters. AI-generated categorizations across all dimensions should be monitored and periodically reviewed for accuracy and consistency. Appropriate controls should address transparency, reliability, accountability, privacy, and human oversight (National Institute of Standards and Technology, 2023). Controls specific to legal practice should be designed to protect attorney-client privilege, client confidentiality, ethical walls, matter-level permissions, and applicable data-retention requirements. Legal professionals remain responsible for the quality of services delivered to clients, and AI should function as an assistive technology rather than a replacement for professional judgment.
Conclusion
The legal profession has spent decades collecting time data, yet relatively few organizations have been able to fully leverage that information for strategic decision-making. AI creates an opportunity to change that, not primarily by making a single coding standard easier to apply, but by making it possible to capture a richer, more structured picture of legal work and classify it across a governed set of dimensions tied to defined business purposes and data-minimization principles.
The next evolution of legal time capture is not simply capturing more hours, or even applying one industry code more consistently. It is building a dataset that reflects how a firm actually works: what type of work is being performed, where the activity originates, what stage of the matter it belongs to, who is doing it, and for which clients and matters, without materially increasing the timekeeper’s administrative workload.
Firms that build this kind of foundation will be better positioned to analyze profitability, improve operational decisions, and develop AFAs that more accurately reflect expected work, value, and risk across flat-fee arrangements, subscription models, capped-fee engagements, and other non-hourly structures.
Attending ILTACON? Don't miss this related session, AI-Powered Time Capture (Session #4611), on Monday, 24 August, from 2:30–3:30 PM CDT.
References
Alshak, R., & Grady, W. (2026, April 16). How AI Is Transforming Time Capture in Law Firms. International Legal Technology Association (ILTA). https://www.iltanet.org/blogs/william-grady/2026/04/16/how-ai-is-transforming-time-capture-in-law-firms
Bedford, S., Brito, N., Von Busekis, K., Cassidy, J., Catanzaro, J., Ritchie, E., & Thomas, J. (2025). From data to wisdom. In KPMG (No. 139930-G). KPMG. https://assets.kpmg.com/content/dam/kpmg/lv/pdf/2025/GM-TL-01764-From-data-to-wisdom_V9-High.pdf