Knowing Where AI Changed the Work: The Next Measure of Law Firm Maturity
Fusion Legal & Tax · October 30, 2026Thought Leadership7 min read
The most consequential finding in the State of Generative AI in Legal 2024 may not be how many firms were experimenting with generative AI. It may be how difficult firms found it to explain what that experimentation changed.
In PERSUIT’s 2024 survey report, hosted by Faegre Drinker, most firms able to estimate GenAI’s impact on billable work reported that 5% or less of billables were affected. An even larger portion could not determine the impact. The report suggested that the low reported impact could indicate firms were applying GenAI in lower-risk areas outside billable work. For the larger group unable to determine the impact, it pointed to challenges quantifying AI’s influence or a nascent stage of AI integration.
That distinction matters. A firm that deliberately confines an emerging tool to internal knowledge work is making a strategic choice. A firm that can see exactly where the tool is being used can explain and govern its own work; one that cannot yet do so lacks that visibility. The reported percentage may look the same from a distance, but the firms are at very different stages of maturity.
What does “5% or less” actually tell law firms?
PERSUIT collected responses from more than 70 law firms between the end of November and December 2023. The respondents ranged from Global 200 organizations to smaller boutiques, providing a cross-market view of adoption, utilization, talent, expertise, strategy, and outlook. The billable-impact finding should not be read as proof that GenAI was economically insignificant in late 2023.
“Billables affected” can conceal several different developments:
- An AI-assisted task may take less time while producing the same billed amount under a fixed-fee arrangement.
- Time saved on a first draft may be redirected to factual investigation, strategic analysis, or client communication.
- A lawyer may use AI for an internal administrative task that never would have appeared on a bill.
- The firm may write down AI-assisted time without separately identifying the reason.
- Lawyers may use an approved tool but fail to record the task in a way that permits firm-level analysis.
- A pilot may be intentionally restricted to a small group, practice, or workflow.
These possibilities do not carry the same implications for staffing, pricing, quality, or client communication. Counting affected hours without defining the underlying workflow gives firm leaders a number, but not necessarily an answer they can use.
The report’s introduction framed GenAI adoption as a client–firm issue, asking how corporate legal departments can determine whether outside counsel is both legally capable and adept at using technology to deliver better service. In his signed foreword, LawVision principal Mark Medice identified an opportunity to reconsider “pricing approaches and strategies” while emphasizing that the industry remained in the “nascent stages” of the transition. Those observations belong together: firms cannot responsibly redesign pricing around AI efficiencies they have not learned to measure.
What should a law firm measure about legal AI?
A useful measurement model should distinguish four concepts that are often collapsed into a single adoption statistic.
1. Access
Who can use an approved tool?
License counts and enabled accounts measure availability. They do not establish whether lawyers use the tool, whether use is concentrated among a few enthusiasts, or whether it affects client work.
2. Utilization
Who actually uses it, how often, and for what category of task?
Utilization can be measured through system metadata collected under clear governance, lawyer reporting, matter codes, or structured pilot records. Firms should avoid collecting more client information than the measurement objective requires. In many cases, task category, practice group, matter type, user role, and frequency may be more useful than retaining complete prompts or outputs.
3. Workflow impact
Which step changed?
“Drafting” is too broad to be a reliable measurement category. A firm needs to know whether AI supported issue spotting, document comparison, first-pass summarization, authority retrieval, chronology development, substantive drafting, editing, or quality control. Each function presents a different review burden and a different opportunity for improvement.
4. Economic and service impact
What happened after the workflow changed?
Potential measures include elapsed time, lawyer time, write-offs, rework, realization, cycle time, staffing mix, and the treatment of the work under the applicable fee arrangement. None should be assumed to improve merely because AI was present. The purpose of measurement is to test the effect, not declare it in advance.
This distinction becomes more pressing as firms move beyond isolated experimentation. Once a tool becomes part of everyday infrastructure, anecdotal success stories are not enough. Firm leaders need operational evidence they can understand, test, and discuss with clients.
How can law firms build an AI impact ledger?
A practical next step is an AI impact ledger: a structured record connecting authorized AI use to a defined workflow and outcome. It need not reproduce confidential prompts or become a surveillance system. Its purpose is to create a consistent vocabulary for evaluating changes in legal service delivery.
For each measured workflow, the ledger could capture:
| Field | Question it answers |
|---|---|
| Matter and phase | Where in the engagement did use occur? |
| Task category | What specific work was assisted? |
| Tool and approved environment | Which system supported the task? |
| Human role | Who operated the tool and who reviewed the result? |
| Source-validation method | How were authorities, quotations, facts, and citations checked? |
| Baseline | How was comparable work performed before the pilot? |
| Time and cycle effect | Did lawyer time or elapsed time change? |
| Rework or escalation | What additional review was required? |
| Fee treatment | How was the work priced, recorded, or written down? |
| Client-facing effect | Did the workflow change timing, scope, staffing, or deliverables? |
The ledger should be proportional. A small firm does not need an enterprise analytics program to begin. It may need only a controlled pilot, a stable task definition, a baseline sample, and disciplined after-action reviews. In fact, boutiques may have an advantage: fewer layers can make it easier to identify who performed the work, how the workflow changed, and what the client received.
What should corporate legal departments ask outside counsel?
The introduction to PERSUIT’s 2024 survey positioned its findings as a basis for clients to question outside counsel about the depth of their technological engagement. Those questions should move beyond “Does your firm use AI?”
More informative questions include:
- Which matter tasks are approved for AI assistance, and which are excluded?
- Can the firm identify where AI was used in our work without exposing other clients’ information?
- What human review and source-validation steps apply to each use case?
- How does the firm determine whether the workflow improved speed, quality, consistency, or cost?
- How are AI-related efficiencies treated under hourly, fixed, capped, or other fee structures?
- What evidence would cause the firm to stop, narrow, or redesign a use case?
- How does the firm preserve junior-lawyer involvement and substantive learning when a task is accelerated?
These are not procurement traps. Asked well, they create a shared language for protecting the relationship while exploring better ways to deliver legal work. The goal is not to force every firm into identical technology or pricing. It is to make expectations, controls, and claimed benefits visible enough to evaluate.
How can a firm start measuring AI impact in 90 days?
Firms still building this capability can begin without waiting for perfect financial data.
Days 1–30: Define. Select one recurring, bounded workflow. Document the starting process, participants, typical review points, billing treatment, and available baseline data. Define what would count as improvement—and what would count as unacceptable rework or risk.
Days 31–60: Measure. Run a controlled cohort using an approved tool. Capture task-level metadata, review steps, elapsed time, lawyer time, and exceptions. Do not rely solely on user impressions.
Days 61–90: Compare and decide. Review the AI-assisted and baseline workflows. Decide whether to expand, modify, pause, or retire the use case. Record what changed in staffing, supervision, quality control, and pricing assumptions.
The exercise may reveal that a tool adds little value to a particular task. That is still useful intelligence. A measured decision not to automate can be as mature as a successful deployment.
What does legal AI maturity look like now?
In his foreword, Medice described the 2024 report as a “roadmap” for law firms and clients navigating GenAI adoption together. One of its clearest directions is implicit: firms need to know not merely whether AI is available, but where it changes the work.
The next generation of credible legal AI claims will therefore be narrower and more testable. Instead of saying that a platform makes the firm more efficient, leaders should be able to identify the workflow examined, the baseline used, the review applied, the result observed, and the conditions under which the conclusion holds.
That level of precision does not diminish innovation. It makes innovation easier to trust, improve, price, and discuss with clients. For law firms of every size, the emerging maturity test is straightforward: Can the firm explain where the work changed—and show how it knows?