This article was originally published in Legaltech News, here.

 

Every firm has done something about AI by now. A committee was formed. A pilot was run. Vendors came through and impressed the room. Somewhere in the budget is a line item leadership can point to and say, yes, we are investing in this.

And yet the sentence I hear most often from firm leadership has not changed: “I have spent money on AI, and I am not getting a return on the investment.” The hype cycle has a name for where that leaves us — the trough of disillusionment, where inflated expectations meet unrealized results. But the gap between what was spent and what can be shown is not a technology problem. The technology works. The failure is operational — it lives in everything that happens after the purchase: the data the tool runs on, the workflows it has to fit inside, the governance that says who may use it for what, and the day-to-day management that turns a license into a capability. This is a deployment problem, and deployment is operational work.

Clients have noticed. Deloitte Legal’s June 2026 survey of 121 senior in-house legal leaders found that 58 percent of general counsel say their outside providers rarely or never proactively raise the benefits of AI with them, and only 4 percent report having directly experienced a benefit from a provider’s AI use. From the client’s side of the table, a firm that cannot demonstrate deployment is indistinguishable from a firm that never started. The committee, the pilot, the platform — none of it registers.

What clients want is not vague. In the same survey, 78 percent named cost reduction as the leading benefit they expect from providers’ AI use, and 57 percent named improved quality of legal services. Those are the two things a firm that has genuinely deployed can evidence — and the two things a firm that has merely purchased cannot.

The tool was never the hard part

The tools are extraordinary. That is precisely the problem. When everyone can buy the same platform, owning it confers no advantage — differentiation moved somewhere else, into the operational layer, and most firms have not followed it there.

Adoption is not integration. Most AI use inside firms is individual and informal: a lawyer running a draft through a public tool between calls, a partner testing a summarizer out of curiosity. That is experimentation. It is not workflow, and it does not compound.

MIT’s Project NANDA put a number on what happens next. Its 2025 report, The GenAI Divide: State of AI in Business, reviewed more than 300 AI initiatives and found that roughly 95 percent of enterprise generative AI pilots delivered no measurable impact on the P&L. The researchers were explicit that the cause was neither model quality nor regulation.

It was what they call the learning gap — tools that never get integrated into a workflow, never retain context, and never improve. Their phrase for it is worth borrowing: high adoption, low transformation.

That is the legal market in one line. Generic tools demo beautifully and stall the moment a real workflow demands context. Moving from a pilot that worked to production that holds requires a foundation underneath it: clean and accessible data, security that survives scrutiny, governance that tells people what they may and may not do. Without it, deployments stay small and tightly controlled — not for lack of ambition, but because scaling would expose what is missing.

In the field, firms treat the platform like a retail transaction: you buy it, you install it, you wait for it to do something — and when nothing changes, you blame the tool. The work is in implementation, in the right data, and in managing it every day after go-live. That work does not come in the box.

From doing innovation to deploying it

Innovation is not a product you buy; it is a process you lead. The firms pulling ahead did not adopt more, they deployed better. McKinsey’s QuantumBlack team reviewed more than 50 of its own AI builds and led with exactly this point: it is not about the tool, it is about the workflow. Efforts that focus on the technology produce impressive demos that do not improve the work. Efforts that reimagine the whole workflow — people, process, and technology together — are the ones that deliver value.

Clio’s data makes the divide concrete from the revenue side: firms that adopted AI widely were nearly three times more likely to report revenue growth than firms merely dabbling. Both groups have the technology. The gap is depth of deployment.

Which brings me to the principle I would ask every firm leader to sit with: diagnosis before prescription. You cannot deploy responsibly onto an unstable foundation. Before layering AI on a firm, you have to know what the firm is standing on — the real state of your data, your security posture, your governance. Most firms skip this because it is unglamorous. It is also the best predictor of whether a pilot becomes production or becomes another line item.

Scoring a firm against a readiness model, and benchmarking it against peers, tends to surface something uncomfortable: the assumption that the foundation is fine is usually the one nobody actually tested.

What deployment actually looks like

Deployment is visible, and it shows up in the workflow, not the toolbar — embedded where the work already happens, not in a side window a handful of enthusiasts open occasionally. If using it requires remembering to use it, it is not deployed.

McKinsey’s legal example makes that concrete. At an alternative legal services provider modernizing contract review, every user edit in the document editor was logged and categorized, creating a stream of feedback the team used to teach the system and enrich its knowledge base. The workflow taught the tool, not the reverse.

It is also measured — and measured at each step, not just at the outcome. In another McKinsey example, a legal services provider saw accuracy drop suddenly on a new set of cases. Because the team had built step-level observability into the workflow, they traced it quickly to lower-quality data coming from certain users rather than guessing at the cause. Firms that track only outputs cannot diagnose their own failures — and cannot report progress to a board in terms that survive questioning.

None of this displaces the lawyer. In McKinsey’s legal workflows, the technology organized claims and recommended approaches, but people reviewed, adjusted, and signed — underwriting the decision with their own license. Deployment changes what the work is, not who is accountable for it.

Diagnosis first, then deploy

The firms that win the next phase will not be the ones that bought the most — they will be the ones that deployed on a foundation built to hold. The move that separates them is not another pilot. It is diagnosis first: know your readiness, fix the foundation, then deploy.

And the commercial clock is running faster than most firms have priced in. In the same Deloitte survey, 85 percent of general counsel expect AI to change how firms price work, and they project the share of work billed hourly to fall from 72 percent today to 44 percent within two to three years. Forty-two percent believe AI-generated savings should be split evenly between firm and client. Those conversations are already happening. A firm that has deployed will enter them with evidence. A firm that has not will negotiate on the client’s terms with nothing to point to.

The honest first step is not a purchase. It is a clear-eyed assessment of what your firm is actually ready for — before you build anything else on top of it. That is unglamorous work. It is also the only thing that makes the rest of it worth doing.