Stuck in the messy middle: why so many firms can’t seem to operationalise AI
Laura Wenzel, global marketing and insights director at iManage, explores why many law firms stall in their AI adoption journeys and details how stronger governance, clearer workflows and the right skill sets can help them move from experimentation to real, operational use.
At first glance, a key figure from our recent global knowledge work benchmark report seems like a triumph: 85% of knowledge work organisations, including law firms and corporate legal departments, are doing something with AI.
But look a little closer, and the story changes. Only 17% have truly operationalised AI within their workflows. The rest are somewhere in between: experimenting with AI, piloting and, too often, stalling out.
That 68 percentage point gap between organisations that have started their AI journey and those that have successfully operationalised it isn’t just a statistic. It’s a symptom of something we might call ‘the messy middle’ — and it’s where much of the legal industry is currently stuck.
The middle is struggling
Looking deeper into the data reveals which types of firms are feeling the brunt of the pain.
Those 17% of organisations who have truly operationalised AI tended to have the greatest level of knowledge work maturity, meaning they have made significant foundational investments in their approach to knowledge work. This cohort might occasionally pause an AI initiative, but they rarely abandon one.
The firms with the lowest level of knowledge work maturity are also highly unlikely to abandon AI projects or pause them – but that’s because this cohort is mainly in the planning phase. In other words, there’s nothing to abandon because nothing has officially been launched yet.
That leaves that middle cohort — the ones with a medium amount of knowledge work maturity — as the group pushing hardest to adopt AI but also abandoning it at the highest rate: some 30% of respondents in this category hit a dead end with their AI projects and then simply move on.
What are the key challenges and how can they be overcome?
When survey respondents were asked to put a finger on what was causing these AI failures, most pointed to people and process, while very few blamed the tools themselves.
“People” really comes down to skills. According to the report, 28% cite skills and adoption gaps as a key challenge, which is understandable: it takes a special skill set to truly integrate AI into your workflows, rather than just offering it as an ad hoc tool for users to leverage.
This is why we’re seeing the rise of the legal engineer – typically a former practicing lawyer who understands the nuances of legal workflows and can map them out as repeatable, improvable processes.
As part of mapping out their processes, firms need to ensure they’re taking a comprehensive approach to AI-related data governance. The research report shows that 36% of organisations have experienced documented security violations related to AI use – a clear indicator that governance in many firms is not keeping pace with AI adoption. And when security issues arise without the skill set or resources to address them, many organisations find themselves at a loss and abandon their AI initiatives entirely – which is why governance and the way it’s actually implemented is so critical.
Rethinking what governance looks like
Too many organisations approach AI governance by writing a policy, distributing it, and expecting employees to self-govern. That approach will permanently limit how far AI can scale within the organisation because it bakes in a dependence on human discretion at every step.
The organisations that have truly operationalised AI have done something different: they’ve embedded the governance into the workflow itself.
Consider a lifecycle governance process for a specific practice area. Rather than expecting a lawyer to recall a policy before every AI-assisted action, the guardrails are defined in advance: which document types are in scope, what disposition rules apply, when human review is required, and what happens next. The AI operates within those rules, reducing the need for human intervention except in high‑risk, high‑profile moments where judgment truly matters.
This approach also solves a problem that’s increasingly on every firm’s radar: auditability. When a regulator or client asks how AI-driven decisions were made, they can point to a documented, rules-based workflow. The AI didn’t act autonomously, it acted within boundaries the firm’s team deliberately set.
When AI is thoughtfully integrated into workflows in this way – rather than leaving professionals to make individual decisions based on a policy they once read – the guardrails are automatically baked in. The workflow becomes the guardrails around the AI.
The path forward
The legal industry spent 2024 and 2025 racing to adopt AI. In 2026, with many firms stuck in the messy middle, they’re posing a different question: How do we get unstuck and successfully move beyond AI pilot projects?
Success with AI doesn’t stem from the number of end users or licenses that have been rolled out within an organisation. It comes from fundamentally rethinking the relationship between AI and workflow, especially as it relates to governance. Until firms make that shift, most will remain exactly where they are: experimenting with AI but not truly leveraging it — stuck in that messy middle.
