“Stronger by Design” is CLOC’s theme for this year’s Global Institute in Chicago, and their own 2026 State of the Industry Report makes the case for why legal departments had better take that phrase literally. Based on Harbor’s survey of 135 organisations, the report describes a structural productivity gap where demand is surging and every traditional lever for handling it has been cut. The question the data raises is: design with what?
To set the scene with the report data: regulatory compliance demand is up 63%, cybersecurity demand is up 58%, and yet expectations for increased outside counsel spending collapsed 21 points in a single year, from 58% to 37%. Inside legal spend expectations dropped from 65% to 47%.
Of the departments surveyed for the 2025 CLOC SOTI report, 83% expected demand to keep rising. The budget to handle it has been cut, and the gap between what legal is being asked to do and the resources available to do it continues to widen.There is no obvious way to close it using the usual playbook.
We spend a lot of time working with corporate legal teams on exactly this problem. There are three responses which we see come up regularly. All three having some merit, but none of them, taken alone, is sufficient. A fourth approach is emerging from the organisations that have moved furthest, and it looks structurally different from the first three.
This option is usually a total non-starter. CLOC’s penultimate State of the Industry Report found that the median legal operations team has remained at just 5 FTE since 2023, with legal ops professionals comprising roughly 5% of total department staff. The ACC 2025 Chief Legal Officers Survey, covering 772 CLOs across 48 countries, named understaffing as the single biggest barrier facing legal departments, with 41% still operating under cost-cutting mandates. An Axiom report from December 2025 surveying over 500 senior in-house leaders across 8 countries, found that 97% say hiring quality talent is extremely difficult, 77% report that both workload and complexity are increasing simultaneously, and nearly half of in-house legal professionals are actively or passively seeking new jobs.
There is enough data to tell us what is already obvious to legal operations leaders with budget visibility – you cannot hire your way out of a structural demand problem when the budget for hiring has been cut at the same time. CLOC’s 2026 report found that 39% of legal ops teams plan to handle rising demand by increasing the workload of existing staff. It is, we think, an uncontroversial view that systemic burnout is not a valid strategy.
AI copilots or assistant tools are gaining adoption with a Clio report in late 2025 finding 79% of legal professionals now use AI in some form. Harvey’s seat counts reportedly double within 12 months of deployment, which speaks to genuine day-to-day satisfaction.
But the aggregate picture is harder to reconcile with all of the enthusiasm. Clio’s same report found that legal-specific AI tool usage actually declined from 58% to 40%, with lawyers increasingly defaulting to general-purpose tools like ChatGPT. The ACC Report on Generative AI’s Growing Strategic Value for Corporate Law Departments which polled 657 in-house professionals across 30 countries, found 59% report no cost savings from their outside counsel’s use of AI, and only 12% track technology ROI.
We know from personal and customer experience that Copilots can accelerate individuals. We also know from direct experience that they do not capture or scale institutional expertise. Research published in Harvard Business Review based on data from 10,000 users at 115 companies, found that employees spend an average of 21% of their work time searching for knowledge and another 14% recreating information they could not find. For a Fortune 500 company, that translates to billions in lost productivity annually. Copilots and well-curated SharePoint environments can help on the retrieval side, but finding a relevant precedent and knowing how to apply it to new facts, with the right exceptions, escalation logic, and audit trail, are completely different problems. In our experience working with corporate legal teams, the demand problem requires something that scales the institution’s collective reasoning, not just the throughput or the search results of whoever happens to be at their desk that morning.
It has not happened at any real scale, but trusting generative AI alone to make substantive legal decisions is where the keenness to scale fewer team members outpaces the evidence most dangerously. Stanford researchers have produced the most rigorous work on legal AI reliability. Their April 2025 paper tested some premium specialised products and found that Lexis+ AI hallucinated over 17% of the time, and Westlaw AI-Assisted Research over 34%. General-purpose LLMs performed far worse when measured by the same researchers, with hallucination rates between 58% and 88% on verifiable legal queries.
Research out of the University of Maryland ran the same legal question 20 times with the same model at temperature zero. GPT-4o gave different answers 43% of the time. The ABA’s Formal Opinion 512 last year characterises AI output as work from “an inexperienced or overconfident nonlawyer assistant.” Damien Charlotin’s website has identified hundreds of documented court cases involving AI-generated fabrications in legal filings, with the rate accelerating through 2025.
Samuel Dahan, Megan Ma, Xiaodan Zhu and Chu-Fei Luo writing in the National Law Review at the end of 2025 put it pretty simply: “Large language models (LLMs) are optimized to produce plausible answers, not to follow legally valid reasoning paths.” The important thing for Legal Ops teams is to recognise that this is an architectural limitation, and something that scale will only amplify. You cannot build enterprise legal services on a probabilistic foundation where your profession and your regulators require deterministic accuracy.
Even the newer wave of AI tools that let individual experts encode personal playbooks into reusable custom skills still depend on the person who built them to pilot the interaction, judge the output, and maintain the configuration over time. When that person moves on, the carefully tuned prompts become orphaned artifacts with no audit trail, no confidence gating, and no guarantee that the same input will produce the same answer twice.
If hiring is constrained, copilots at best only scale individuals, and pure LLMs cannot be trusted for deterministic legal work, what is left?
Look at what has actually worked at enterprise scale. Microsoft’s CELA team built a centralised data infrastructure before layering any AI on top. Allen & Overy Shearman combines LLMs with curated “gold standard” precedent libraries against which AI cross-verifies its suggestions. JPMorgan’s COIN platform uses a hybrid of ML, NLP and rules-based classification to process 12,000 commercial credit agreements annually. Thomson Reuters grounds CoCounsel outputs in Westlaw’s authoritative content and 4,500 subject matter experts. Neota Logic customers are able to implement a neurosymbolic method which achieves up to 100% accuracy on legal queries in contract and privacy and compliance matters, where LLMs alone consistently produce work with errors.
The consistent pattern: none of these approaches relies on generative AI alone. Every success story places AI within a deterministic framework. Rules engines handle decisions that must be right every time. AI handles the unstructured work it genuinely excels at: extraction, summarisation, drafting, document classification. Human oversight sits at structured intervention points for the genuinely difficult calls.
CLOC’s theme prompted us to pose the question: design with what? If the CLOC data is right that the shift is structural, then the answer must be structural too. Not another assistant that makes one lawyer faster, but an architecture that turns what your team collectively knows into services the whole organisation can rely on. That is what “Stronger by Design” should mean in practice.
At Neota Logic deterministic logic has always been our backbone. Now, we apply that deterministic rigour to our AI Task Agents, which operate within governed workflows, with confidence gating that routes uncertain outputs to human review rather than letting them pass silently. Verification guardrails check AI outputs against source material before anything reaches a user. Every decision path produces an audit trail. The practical effect is that your team’s collective expertise gets encoded into digital services that deliver reliably, whether your best specialist is available that day or not.
The Neota Logic team will be at CGI in May. If you are trying to move beyond individual AI productivity and into the governed delivery of a substantive legal service, we would very much like to hear how you are thinking about it.
