Deloitte just published its survey of AI adoption in the legal industry and the headline numbers are interesting:
61% of legal departments are now in active AI deployment (p.13), whereas two years ago, 76% reported no adoption at all (p.4). AI agents are projected to handle 30% of in-house legal work within three to five years (figure 13, p.22). External legal spend could fall by 20–40% over the next three years (p.28).
These are the numbers that will circulate on LinkedIn, appear in board presentations and find their way into budget conversations. They are also, if taken at face value without the surrounding context, misleading.
Because buried in the same report is a finding that General Counsels should be taking far more seriously than the projections.
Deloitte surveyed 121 senior legal leaders: General Counsels, Heads of Legal and Legal Operations Directors across nine industry sectors and four regions. The AI investment picture they describe is one of rapid acceleration: 79% of legal departments increased AI spend year-on-year (p.8), with average budgets rising 67% (figure 1, p.8).
But follow that money and a problem emerges. Technology and infrastructure attracted the highest share of investment from 67% of respondents (figure 2, p.8). Training and transformation, the work required to change how people operate, ranked considerably lower. An updated AI data strategy is in place for just 36% of all respondents, and quality assurance frameworks exist in only 24% of organizations (figure 3, p.9).
“The technology is ahead of the organization. We have tools our lawyers aren’t yet fully using, and processes that haven’t been redesigned to take advantage of them.” - General Counsel, Financial Services (Deloitte Legal AI Survey, 2026)
The root cause is an information architecture problem, and it has a name: “IA before AI.”
The principle is straightforward, and it predates the current AI cycle by decades: before you automate a process, you have to understand it. Before you train a model on your data, that data has to be structured, governed and trustworthy. Before AI agents can autonomously handle contract review, obligation tracking or regulatory compliance work, the contracts, knowledge assets and operational data those agents will rely on need to be in order.
This is what we mean by IA before AI. Information Architecture before Artificial Intelligence.
The argument here is about sequencing, not skepticism about AI investment. And the Deloitte data makes the case for it more clearly than any vendor presentation could.
“Large-scale technology transformations fail most frequently not due to technology choices, but because of insufficient investment in the people and cultural change required to embed new capabilities.” - Deloitte Legal AI Survey, 2026 (p.8).
The legal functions that realize the greatest value treat transformation as a first-order priority, not as something to address once the platform is live.
The organizations that stall, the report observes, share a common pattern: “investing heavily in technology and lightly in data, people, process redesign and cultural change.”
Nowhere is this tension more visible than in contracts and commercial work, which the Deloitte survey identifies as the leading practice area for AI adoption (figure 6, p.14) and the area with the highest automation potential, at 34% (figure 9, p.17).
Contract Lifecycle Management is, in theory, exactly where AI should deliver the most measurable value. Contracts are structured documents. They have defined fields, standard clauses, obligation triggers, renewal dates and counterparty data. They contain the commercial terms that govern how a business operates. And for most organizations, they are also the most comprehensively mismanaged data asset legal teams own.
Contracts sit in shared drives, email chains, legacy repositories and a patchwork of platforms that don’t communicate with each other. The metadata is incomplete. Templates are inconsistent. Obligations tracking is manual, where it exists at all.
AI cannot fix this. It will amplify it. A model trained on poorly structured, inconsistently maintained contract data will surface unreliable results. An AI agent tasked with obligation tracking cannot do its job if the obligations were never properly extracted and stored. Operational Intelligence, the ability to surface meaningful insight from legal and contract data, only works when the data itself is fit for purpose.
This is why the finding on data strategy, i.e., only 36% of legal departments have an updated AI data strategy in place (figure 3, p.9), is the most consequential number in the Deloitte report.
Not the 30% AI agents projection. Not the 67% budget increase. The 36%.
The Deloitte report describes a three-stage transformation roadmap: Foundations, Embedding and Scaling. The Foundations stage, it notes, is one organizations “spend longer in than they might have expected.”
This is the work. And the General Counsels who will see the most from their AI investment over the next three years are the ones who treat the Foundations stage with the same discipline they currently apply to platform selection.
In practical terms, that means auditing the contract data you have and its current state. It means standardizing intake processes before automating them. It means building a knowledge management approach that treats your legal data (contracts, precedents, opinions and playbooks) as a structured asset rather than a filing system.
It also means being honest about sequencing. Deploying AI on top of processes that have not been redesigned to support it does not accelerate transformation. It accelerates the wrong things.
The report raises three questions it doesn't fully answer.
Governance without guardrails. Only 24% of organizations have quality assurance frameworks in place for AI use cases (figure 3, p.9). As agentic AI scales, and it will, given that 61% of respondents are already experimenting with it (figure 11, p.20), the cost of that shortfall grows significantly. An AI agent acting autonomously on poorly governed contract data without a QA framework is a source of risk exposure, not a productivity tool.
The Jevons’ Paradox question. Will AI generate more work and keep headcount broadly stable, or genuinely reduce it? Richard Trowman at Artificial Lawyer raises this point in his reading of the report1. The answer likely depends on whether organizations use AI to do the same work faster or to do fundamentally different work. The former is efficiency. The latter is transformation. Only the latter changes the economics.
Who owns the contract data? The report addresses the relationship between legal and external providers but does not fully resolve the internal question of where contract data governance sits. Legal, procurement and finance all have a legitimate claim on contract data. Without clear ownership, the information architecture problem does not get solved. It gets deferred, and it becomes someone else’s problem to fix when the AI deployment underperforms.
The Deloitte Legal AI survey is a substantive piece of work. The projections are bold, the case studies are useful and the transformation roadmap is practical. What it confirms, between the headline numbers, is that the legal functions best placed to capture value from AI are not those investing most heavily in technology. They are the ones who understood, early on, that AI is only as good as the information it runs on.
Get your information architecture right first. Everything that follows depends on it.
Co-Flo’s CLM for Legal is built on iManage Work, the platform most legal teams already use. It extends iManage with structured intake, automated workflows, real-time dashboards, and AI-powered obligation tracking, covering the full contract lifecycle from the first request through to post-signature. For legal teams working through the Foundations stage, it is designed to do exactly what the Deloitte report recommends: put the data and process infrastructure in place before the AI layer goes on top.