Our recap of last year's ILTACON described a conference where AI had already moved from risk to results. Firms were past asking whether AI was safe and into embedding it in high-impact workflows, with implementation, more than tool selection, as the real differentiator. That same recap flagged one area still underexplored: commercial models.
"As technology changes delivery, pricing models and value propositions will have to adapt, and that shift is coming," we wrote one year ago.
ILTACON 2026 is where that shift arrived. Across sessions, vendor floors, and hallway conversations, the center of gravity moved from "are you using AI?" to a harder question: what is it doing for you, and can you prove it? Usage is up industry-wide, and so are AI budgets, but few firms or legal departments have a credible way to connect it to outcomes.
This tension ran through the whole conference, from operating-model discussions to renewed debate over the billable hour, and it framed one of the week's more candid conversations: why legal teams have gotten good at using AI, but almost no one has figured out how to prove it's working.
The measurement gap is real
Firms and legal departments can track usage. What they can't yet show is value. Everyone can see people using the tools. The question is what value is being generated, internally, in how firms practice and deliver work, and externally, in what clients are getting from the output.
That gap shows up in the numbers. Usage is climbing and so is cost, but no one has seen costs come down as a result, a dynamic where lawyers report feeling more efficient with AI even as billing data shows them logging more time per matter than before.
Harbor's own research puts a hard number on the gap. Our newly released report, "Operationalizing AI: Transforming AI investment into business and client value," found that 0% of participating top-tier law firms and global in-house departments have a mature framework for measuring AI's business impact, despite a 41% increase in average annual law firm software spending between 2021 and 2025. Investment has moved fast; accountability for what it produces has not kept pace.
Part of the problem is that firms are measuring the wrong things.
In one session Q&A, an IT leader at a smaller firm asked how to demonstrate technology ROI to partners, and the answer cut to the heart of the gap. Stop reporting logins and usage rates. They don't answer the question partners are asking. The metrics that matter are growth metrics tied to production and practice, connected to how a practice performs, how many more engagements it has won, how much revenue has grown. Smaller lawyer populations, if anything, make this easier to quantify. The larger point applies beyond smaller firms. Work backwards from the business result you're solving for, then choose metrics that ladder up to it.
The fix isn't a single metric. It requires tracking outcomes on two levels: the quantitative (is a legal or business problem actually being solved faster, better, or cheaper) and the qualitative (how is the organization changing, how are people's habits and day-to-day work adapting).
AI value comes from the operating model
The idea that AI tools are failing when firms don't see ROI is faulty logic. The technology largely works; the question is whether the operating model around it is built to capture the value it creates. Harbor's Enterprise AI Operating Model framework is built on five capabilities firms must develop to convert AI investment into measurable outcomes: AI strategy, measuring what matters, investing with intention, changing behavior, and building trust at the speed of innovation.
Skip any of those, and the organization around the tool becomes a bottleneck. Data is where that bottleneck most often starts. A recurring theme across ILTACON's marketing technology and knowledge management sessions was that clean, connected data is the precondition for AI producing anything useful. Give AI a trustworthy, connected data set and it accelerates reporting and surfaces opportunities faster; point it at a broken one and it will still return a confident answer, just a wrong one. This is structural transformation debt in its purest form, years of disconnected systems and deferred governance decisions, surfacing the moment AI is asked to reason over what's underneath.
The sequence that came up repeatedly: connect the systems so tools can talk to each other, govern the data and establish trust in it, then apply AI for precision.
Skip straight to the last step and firms end up with a shiny tool sitting unused on a shelf. One practical way firms are managing this without waiting a decade for a firm-wide data initiative: govern only the data behind a specific, recurring deliverable, such as a report produced regularly or a request fielded constantly, and expand from there. It’s slower to feel comprehensive but it delivers value continuously instead of on some distant future date.
Governance itself is still catching up. A recurring frustration raised in roundtable discussions was fragmented ownership of AI security and governance decisions across IT, Risk, the General Counsel's office, and knowledge management, with client requirements increasingly arriving through outside counsel guidelines faster than internal policy can keep pace. This is exactly what Harbor's "build trust at the speed of innovation" capability is meant to address. Governance needs a clear owner and a defined, visible conversation.
This is also not a one-size-fits-all problem. AI's impact varies by practice, shaped by the specific type of work being delivered. Blanket adoption of a single tool doesn't tell you much; what matters is what it does for a specific practice and the services that practice provides.
The commercial model is the next pressure point
One example from ILTACON 2026 makes the tension concrete. The CIO of a 100-lawyer insurance defense firm described his managing partner ordering Copilot and other AI tools shut off after billing came in down, a decision framed around security concerns but driven by revenue.
Corporate legal clients have told Harbor directly that they're willing to pay for the cost of AI use, provided it's tied to outcomes and value, rather than passed along as another line item. Combined with a coming reckoning over rising token/inference costs layered on top of AI-driven billing increases, that sets up a pricing conversation the industry hasn't fully had: if usage and cost both rise while billed time doesn't fall, something in the commercial model has to give.
Adoption is a change management problem
Implementation success will come from embedding forward-deployed engineers directly alongside the teams doing the work, people who understand both the technology and the legal or business process, placed close to the point of delivery. The logic follows classic change management: people who feel like participants in a change are more likely to adopt it than people who have a tool handed to them.
The same logic showed up throughout the conference's project leadership and knowledge management conversations too. One recurring observation: capable, well-intentioned professionals still resist well-designed rollouts when adoption is treated as a post-launch activity rather than a design requirement from the start.
A separate session on project leadership for legal tech made a related point about how success itself gets measured; go-live should be treated as the start of adoption rather than the finish line, and real success criteria extend past an on-time launch to usage, satisfaction, and benefit realized weeks later, when nobody's watching the rollout anymore. Legal tech environments rarely move in straight lines; priorities shift mid-project and integrations surface dependencies no one planned for, which argues for treating implementation plans as living documents with built-in checkpoints rather than fixed waterfall milestones.
Consumer AI has also reset expectations. Because tools like ChatGPT and Gemini have redefined the term user-friendly, enterprise legal tech is now held to a consumer-grade bar: familiar, frictionless, with no training required. The risk is that frictionless gets mistaken for trustworthy, making it easier to skip the verification step that still matters most in legal work. The strongest interfaces and workflows build in easy, well-timed opportunities for human intervention, even when the output feels convincing.
The closing theme, and a look ahead to 2027
ILTACON 2026 ended in both excitement about the possibilities and a caution that ran through the whole conference: AI is a prediction engine that supports, rather than replaces, human judgment. As Above the Law's Joe Patrice put it, AI can't invent the equivalent of a plane if it's never seen one. Trained purely on existing patterns, it will often produce answers that are technically not wrong, but not necessarily right for a nuanced legal situation either. Patrice goes deeper on this with Legaltech Hub's Cate Giordano in the season two premiere of Legal Soundings, Harbor's podcast.
If 2026 was the year legal caught up to consumer AI expectations and started asking more complex questions about ROI, 2027 is likely to be the year those questions get answered, one way or another. Expect the 0%-mature-framework statistic to become the baseline every firm is measured against, as clients and finance leadership push for a real accounting of what AI spend is producing. The commercial model conversation will likely move from theory to practice, with more firms piloting hybrid or outcome-based pricing where it's easiest to prove value, even as the billable hour holds on elsewhere.
Expect the verification conversation to extend to autonomous agents as well. Where a chatbot's output invites a human to check it, an agent that acts rather than suggests raises a different question, how much oversight and accountability is required before agentic work is trusted anywhere near a client matter. That trust threshold, rather than the underlying capability, will likely determine how quickly agents move from pilot to practice in 2027.
Talent and governance will keep converging, too: the hiring pace behind Harbor's 266 AI-focused professionals suggests firms are treating AI oversight as a permanent function, rather than a project team that winds down once the pilot ends. And as the forward-deployed engineer model spreads beyond early adopters, embedded, practice-specific implementation will likely become a baseline client expectation.
The throughline for next year's ILTACON: less conversation about whether AI works, and a lot more about who can prove it.
- AI
- Analytics
- Change management
- Show all 6
