Australian legal teams are enthusiastic adopters of AI

Australian legal teams are enthusiastic adopters of AI. Unfortunately, enthusiasm is outpacing strategy and compromising return on investment. Thomson Reuters' 2026 Future of Professionals Report, a survey of 1,816 professionals across 62 countries, found that adoption is no longer the barrier, with 74% of professionals now using AI at least several times a week. This aligns with what we see both at MinterEllison and the in-house legal functions that we work with. The gap now is one of value, with 91% of survey participants saying their organisations are falling short of what the technology could deliver, a shortfall the report calls the “AI value gap”.

However, where a considered strategy is in place, 66% of professionals say AI is meeting or exceeding expectations on value; where a strategy is not in place, that figure falls to 22%. Strategy is a clear differentiator, and a framework for deciding where AI fits your service delivery model, which work it touches, and with what data, is what turns adoption into value.

That decision framework matters even more now that the legal industry has become such a focus for AI innovation, and the number of vendors looking to serve the market has exploded. The major AI labs, including those behind ChatGPT and Claude, have shown a particular interest in legal work and are building capabilities specifically for it. OpenAI has hired the co-founder of contract management platform Ironclad to lead a new legal team, and Anthropic has built a dedicated legal version of Claude. Amazon has now joined them, launching Amazon Quick for Legal, and Microsoft has rolled out early access to a new Legal AI agent for Copilot. AI is also being built into the core products that legal functions already run, such as triage and matter management, research tools, document management, contract lifecycle management and spend management. And a wave of well-funded new entrants have burst onto the scene with the promise of addressing almost every legal function workflow in one place.

The challenge for General Counsel is deciphering which products fit their function, and just as importantly, how they articulate (and justify) their choices to those who control the budget and the technology environment.

This is a strategy problem before it is a procurement one. As we set out in our series on the in-house legal operating model, technology must serve the legal function strategy, not the other way round. In this article, we set out the three broad categories of AI now available to legal functions, the factors that should drive the choices between them, and identify how General Counsel can build a more defensible narrative to help unlock investment and bring stakeholders with them.

An AI tool is not an AI strategy

The market tends to frame choices in this arena as somewhat binary. Build or buy? General-purpose or specialist tool? That framing is far too blunt for the decisions a General Counsel actually faces.

An AI strategy for a legal function does not exist in a vacuum and generally draws from two sources: the organisation's broader AI strategy (and governance parameters) which sets the boundaries and risk appetite, and the legal function's own strategy and operating model which defines the work to be done, by whom and how (including enabling technology). The General Counsel needs to ground their AI choices in these sources.

In practice, the most mature legal functions are not picking a single product, but rather combining a broad, general-purpose backbone for everyday work with specialist tools layered in where precision and depth matter most. The question is not simply "which AI tool should we buy?", it is "which kind of AI fits with which part of our work?" Treating the decision as a single, permanent purchase is the most common way to overspend, create duplication, or stall in perpetual pilots.

The three broad categories

1. General-purpose enterprise AI

These are the horizontal AI assistants and productivity-suite copilots, such as ChatGPT, Claude and Microsoft Copilot, that your organisation may already be paying for across the business.

Their strengths are real and attractive:

• they work in natural language, the core medium of legal work;
• they carry little or no marginal cost where the enterprise already licenses them, a fact that tends to resonate with CFOs and with CIOs who would rather leverage existing infrastructure than onboard another system; and
• they are familiar and quicker to adopt.

Their limitation is that they are not built specifically for legal work. They are not designed to be robust document or matter management systems, they lack legal-specific workflows, and their outputs can lean on general web sources and overseas-centric reasoning rather than authoritative, jurisdiction-specific material. They can also produce confident, well-structured answers that are wrong – a risk that is acute for junior lawyers who may lack the experience to spot the error. We still see in-house lawyers persist in using them for legal research because that is all they have access to, which tends to produce frustrating and unreliable results. Where the enterprise already licenses these tools broadly, proper permissions management is essential. Without active configuration, lawyers may inadvertently expose privileged or confidential material to the wider organisation, or access information they should not.

The natural home for these tools is horizontal work, such as summarising, first drafts, brainstorming and internal communication, and building AI literacy across the team. The value of these tools is directly proportional to the AI maturity of your people and their ability to configure them for legal work rather than general use. In our experience, the real gains come from function-specific training on these enterprise tools, aimed at legal use rather than a generic rollout, which lifts both adoption and the quality of what the tools produce. For many legal functions they are the sensible bridge, a low-risk way to build confidence and evidence demand before committing to specialist spend.

2. Point solutions: native or layered on existing technology

These go deep on a single workflow or problem. Some are AI-native tools built to do one thing exceptionally well, such as reviewing and redlining contracts. Others are AI layered onto an existing legal system of record, such as matter management or contract lifecycle management tools, embedding intelligence where the work already happens rather than in a separate window. By concentrating on a single, significant problem, they deliver a focused and material return.

The embedded solutions extend this advantage, operating within existing workflows rather than merely responding to queries. The corresponding risks are narrowness and tool sprawl. A proliferation of single-purpose tools can duplicate one another and compound the integration and governance burden. Their best fit is a clearly defined workflow, or a desire to bring AI into the system where your legal work is already managed. Some point solutions also serve work that spans functions, which means their cost and value can be shared. A tool that supports both legal and marketing workflows, for example, benefits two teams and can be funded from a broader base than the legal budget alone, which strengthens the business case for each of them.

3. Purpose-built legal AI platforms

These are broad tools designed specifically for legal work, with the legal-grade citation and jurisdictional coverage that general-purpose tools do not offer. They can lift a wide range of legal tasks at once. The main trade-offs are cost, procurement effort and greater change management overhead that the legal function has to fund and drive itself (since these platforms sit outside the environment where work already happens). Many of these tools were also built first for law firms rather than in-house teams, so their default workflows may not map neatly to how an in-house function operates. Access is also typically limited to the legal function itself, which means the broader business does not see or benefit from the investment directly.

There is no doubting the power of these tools, and their best fit is primarily a legal function with genuine breadth of legal work and the maturity and budget to embed a platform and drive adoption across it. An interesting angle for General Counsel is accessing some of these platforms' capabilities through "portals" offered by their panel law firms, which may be a smart play.

What should drive the choice

This is the decision framework we foreshadowed in the introduction: the factors that should drive the choice between the three categories outlined above. Armed with a view of the full range of technologies and options available to them, a General Counsel should weigh the following:

• Start with your strategy. Before looking at tools, be clear on what the legal function is there to do and how it has decided the sum of legal work will get done, whether internally, through panels, or via self-service.
• Build a business case based on the work. The most persuasive business cases begin with a clear-eyed view of where the legal function's effort and cost actually go, where the legal function may be unnecessarily standing in the way of business self-service, and which work is worth solving for. As we have set out in our work on building data-driven AI investment cases, this baseline is what turns a request for "productivity tools" into a proposal to solve a quantified problem.
• Weigh total cost of ownership. The subscription is the visible, and often smaller, part of the cost. The real economics sit in the hours spent checking and correcting outputs, in training, integration (especially with knowledge sources) and change management, and in the duplication that comes from overlapping tools. AI pricing itself is also shifting, as vendors move from flat per-seat subscriptions towards consumption and usage-based models. As costs scale with usage it can become harder to forecast the total cost of ownership. That is precisely why finance and technology leaders now expect a clearly defined business case and a measurable return before they commit. This is the number a CFO cares about, and the one that separates a tool that pays for itself from one that quietly costs more than it saves.
• Treat security, privacy and data residency as threshold questions. Legal work runs on confidential and privileged information. Where is data stored and processed? Is it used to train the vendor's models? For Australian functions, data residency and the AI-specific risks that traditional IT controls do not fully address are not details; they are the questions your CIO will ask first, and the ones most likely to decide whether a tool is viable at all. It is also worth remembering that the legal function is itself subject to the organisation's AI governance framework, often with additional restrictions arising from legal professional privilege, information barriers, and professional obligations that do not apply to other business units. Bring IT, security, procurement and finance in early to understand their priorities and processes, allowing you to shape the business case around them.
• Check the fit to Australian law and your own data. Availability is rarely the constraint; suitability is. A tool that has access to predominantly overseas material, or that is unable to draw on your own precedents and playbooks, will disappoint in ways that are hard to see until you rely on it.
• Be honest about maturity and change capacity, and cost it in. The best tool your team will not adopt delivers nothing. Adoption is not free: it consumes training time, process redesign and management attention, and every tool you introduce draws on the same limited pool of change capacity. Match the ambition to the function's readiness, budget realistically for the change effort, and prefer consolidation over sprawl wherever you can.
• Expect a blend and sequence it. Use general-purpose AI to build literacy and evidence demand; add a legal platform or point solution as the case is proven. Starting broad builds the evidence and internal confidence that justify heavier specialist spend later, and a staged rollout keeps cost and change effort manageable at each step.
• Revisit as the market converges. The lines between these categories are already blurring; today's decision is not permanent and should not be treated as such. As general-purpose tools add legal features and specialist platforms broaden their reach, set a regular cadence to review your stack against how the market and your own needs have shifted, and be ready to rebalance.

These factors line up with the four dimensions we set out in our guide to evaluating legal AI: data and model architecture; security and governance; accuracy and reliability; and integration and support, which is where to turn once you are assessing a specific solution.

A quick note on what is not the differentiator. Most tools across all three categories are built on the same handful of commoditised foundation large language models. Even at the frontier of legal AI, the durable advantage now lies in workflow depth, data and integration, not in which model sits underneath. This means evaluation should look past the model and the demo and focus instead on how it addresses your function's specific needs. There is also a practical cost angle here. Some architectures let administrators set default models by task, which curbs the tendency for users to reach for the most powerful and expensive model when a lighter one would suffice. As AI pricing shifts towards tokens consumed, and because not every tool lets you set a cheaper default model for a given task, the same work can cost very different amounts depending on the tool you reach for. Using the right tool for the job is therefore an ongoing discipline, and one that is easier to hold with a point solution, where usage and return on a specific problem can be easier to measure than with general-purpose AI spread thinly across everything.

Where to from here: turning optionality into investment

The reason all of this matters is that the choice rarely rests with the General Counsel alone. Unlocking budget means convincing a CFO. Deploying safely means satisfying a CIO. Landing the change means bringing procurement and internal clients or stakeholders with you.

Each of those stakeholders is asking a different question. The CFO wants to know what problem this solves, what it truly costs, and how you will measure the return. The CIO wants to know how it handles data, where that data lives, and how it fits the security environment. The business wants to know that working with legal will get faster, not slower.

Optionality is an advantage only when it is paired with a rationale. "We chose this kind of tool, for this work, because the numbers and the risks point this way" is a far stronger position than "everyone is buying this AI." It is our observation that the functions moving fastest are not the ones with the biggest budgets. They are the ones that can explain their choice in the language each stakeholder understands.

There has always been more than one way to bring technology into a legal function. The task for General Counsel is not to find the single right tool, but to choose deliberately among good options, and to make the case for that choice with the clarity their stakeholders expect.

MinterEllison's Legal Optimisation Consulting team helps in-house legal functions map the legal technology and AI landscape to their strategy, build the business case, and choose and implement with confidence. Contact us to start the conversation.