AI Will Not Save You From Your Data
Law firm IT is forced to live with a level of customization that most other industries do not tolerate. Every practice has its own workflow. Every attorney has a preferred way of working. Every client brings requirements that must be accommodated.
To support the business, IT does what it has to, which means getting creative with details like creating custom fields on standard systems, purpose-built scripts to fill the gaps between them, one-off integrations to make them talk, and workarounds layered on top of workarounds.
The result is the environment most of us operate today. Systems that mostly talk to each other. Data that mostly agrees; processes that mostly line up. For years, “mostly” was good enough, because humans filled in the gaps and everyone worked to satisfy each client’s requirements.
AI changed all that. At its core, AI is still just the execution of small programs against data. The models are more sophisticated, but the fundamentals have not changed. Automation depends on predictable inputs, consistent structure, and repeatable processes. When those hold, it scales. When they do not, the output is inconsistent, unreliable, or wrong.
That is what most law firms are running into now. The tenant is configured, the licenses are assigned, and the pilots looked clean. Yet the automations that worked in a vendor’s demo are hard to reproduce across the firm. The problem is rarely the AI. It is the data and processes underneath it, which were never designed for consistent execution.
Fixing that is not solely a technology project. It is an organizational one, and IT cannot solve it alone. It requires convincing the business that the next decade of value will come from consistency, not customization. That is a hard sell in a law firm. It is also the conversation we need to start driving.
Automation Depends on Predictable Data
Every automation, whether it is a script, a workflow, or an AI agent, makes the same basic assumption. The inputs will look the way they looked last time, the fields will be in the same place, and the process will run the way it ran yesterday. When those assumptions hold, the automation is boring and reliable. When they do not, it breaks in ways that are hard to see until something downstream goes wrong.
Law firms often struggle with predictability at the data layer. Information about a single matter typically lives in multiple systems. The DMS, practice management, CRM, SharePoint, Teams, and email were each implemented at a different time, by a different team, with a different set of priorities. Each has its own idea of what a client is, what a matter is, and how the two relate.
The same concept ends up represented differently depending on where you look. One system calls it a Client ID, another a Matter Number, and a third stores it as free text in a description field. Humans navigate this without much thought, but automation cannot. An agent asked to pull together information about a matter spends most of its effort reconciling identifiers instead of doing the work you really wanted it to do, ad this drives up the cost of the automation while lowering its effectiveness.
And this is why so many AI pilots look impressive in a demo and struggle in production. The demo used a clean data set built for the demo; production never had that luxury. The gap between the two is not a modeling problem or a prompt engineering problem: it is a consistency problem, and it existed long before AI was added into the equation.
Operational Maturity Is the Real Differentiator
A pattern shows up quickly once a firm moves beyond the pilot phase. Two firms deploy the same technology, licensed the same way, configured against the same guidance. One starts finding real opportunities within weeks. The other spends months struggling to get consistent value out of the same tools. The technology is not the variable. Operational maturity is.
Mature environments share a few characteristics. Naming conventions are enforced, metadata is populated because the process requires it, and systems of record are clearly defined so everyone knows which one wins when they disagree. Access is governed by policy rather than by whoever asked most recently. None of this is glamorous. Most of it was built quietly over years by people who cared about doing things the right way.
Less mature environments look functional on the surface but fall apart under scrutiny. Two groups store the same information in different places, fields that should be required are optional, and repositories that were supposed to be temporary have quietly become permanent. Nothing is broken in a way that anyone would escalate, but nothing is clean enough to automate against either.
AI does not create these differences, it exposes them. A law firm that has invested in operational discipline finds its data is ready to be worked with. And a law firm that has deferred that work finds every automation project starts with a cleanup project. The tools are the same; the starting line is not.
This is worth naming plainly when discussing automation at your firm. AI does not raise every boat equally: it rewards the organizations that did the unglamorous work, and it makes the cost of skipping that work newly visible.
Governance Is Consistency at Scale
Governance is usually framed as a control function: protect the data, manage the risk, and satisfy the client surveys and insurance requirements. Those things matter, but they are not the reason governance is suddenly showing up in conversations about AI. The reason is simpler. Governance is the mechanism by which an organization produces consistency at scale.
A sensitivity label applied to one document and not another is not just a protection gap. It is an inconsistency that any downstream automation will inherit. A retention policy enforced in one repository and ignored in another produces two different realities in the same technology stack. A DLP rule that fires reliably for some content types and unpredictably for others teaches users and systems to distrust the output. Every governance gap is also a consistency gap, and consistency is what automation runs on.
When IT asks the business to invest in governance, the response is often polite resistance. Governance sounds like overhead. It sounds like slowing things down. It sounds like something that benefits the compliance team and inconveniences everyone else. But when IT asks the business to invest in consistency instead, the conversation should change. Consistency is what allows automation to work. It is what makes AI outputs trustworthy. It is what turns a promising pilot into a production capability.
The controls are not any different. The labels, the policies, the retention rules, and the access reviews are all still tools of the trade. What changes is why the firm does them. Governance stops being a defensive posture and starts being a prerequisite for everything the firm wants to do next.
IT Cannot Win This Argument Alone
None of this gets fixed by IT alone. The systems that IT own are only part of the environment. The rest is owned by the practices, the internal business functions (HR, Accounting, Business Services), and the attorneys who built their workflows around the way things are today. Any real move toward consistency requires those groups to change how they work, and that means IT has to get good at a kind of conversation it has historically avoided.
The instinct is to lead with the technology, explaining the architecture, showing the diagrams, and walking through the policy model. That approach lands with peers in IT and almost no one else. The business does not care how the automation works. It cares what the automation makes possible and what it will cost to get there. The case for consistency has to be made in those terms.
Consistency is what allows the firm to trust the output of an AI tool in front of a client. It is what turns a one-off script into a repeatable service an AI agent can perform reliably. It is what keeps the cost of every new automation from starting at zero. Each of these is a business, not technical, argument and each is true.
Expect resistance and expect it from people who are not wrong to push back. A partner who has run their practice a particular way for twenty years is not going to change because IT sent a memo. A practice group that has built its reputation on a bespoke process is not going to abandon it because a governance committee asked nicely. The work is slower than that. It looks like picking the right first use case, showing a small concrete win, and building enough credibility that the next conversation is easier than the last.
IT will not win this argument by being right. IT will win it by being patient, being specific, and translating a technical problem into a business one. Then repeating that process until the business starts making the argument on its own.
The Firms That Did the Work Will Move Faster
The firms that get the most out of AI over the next several years will not be the ones with the biggest license counts or the most aggressive rollout plans. They will be the ones that quietly did the work underneath. Cleaner data, consistent identifiers, enforced metadata, and governed repositories where processes run the same way every time for every matter and every client. The unglamorous work that never made anyone’s roadmap look exciting is about to become the work that determines what the firm can and cannot automate.
That is a hard message to carry into a firm that has spent decades rewarding customization. It is also an honest one. AI is not going to make an inconsistent environment consistent, it is going to make the inconsistency more expensive, more visible, and harder to work around. The sooner IT starts having that conversation with the business, the more runway there is to do something about it.
Ready to connect at ILTACON 2026?
If this is a conversation you are having at your own firm, or even one you are trying to start, come find us onsite in Nashville for ILTACON 2026. Dan Paquette will be presenting on this topic with Faith Drewry of Bryan Cave Leighton Paisner and Rich Lilly of Refoundry on Tuesday, August 25 at 10:00 AM Central. Their session will discuss how firms are approaching AI and data in practice: what is working and where the hard conversations tend to happen. Come, ask questions, and be part of the session!
Visit Kraft Kennedy’s ILTACON 2026 event page for registration details, agenda information, Golf Outing tickets, and much more.