A lot of in-house legal teams have been buying AI tools for two years. One for contract review. One for legal research. One for document drafting. Perhaps an AI assistant embedded in the document management system, or a chatbot bolted onto the matter intake process.
Each tool made a compelling case at procurement. Each one demonstrably speeds up the task it was designed for. And yet for many GCs, the overall experience of AI in the legal function feels underwhelming relative to the time, money, and change management effort invested.
The reason is not that the tools are bad. Most of them work as advertised. The reason is structural, and understanding it changes how you think about what you should be building.
The architectural choice that determines whether AI compounds
The distinction between a task and a workflow sounds simple. In practice, it is the single most important variable in predicting whether a legal team's AI investment generates compounding returns or stays flat.
A task is a discrete piece of work with a defined input and output: summarise this contract, find the relevant clause, draft a response to this query. Point tools are designed for tasks. They accept an input, process it, and return an output. They are often accurate, frequently impressive, and genuinely useful. And after twelve months of continuous use, they are no better at your organisation's specific legal work than they were on day one.
A workflow is an end-to-end sequence of connected steps: matter intake, triage, assignment, drafting, review, approval, execution, obligation tracking, renewal. A platform designed for workflow automation does not just process the immediate task. It accumulates context across every matter that passes through it. It learns the organisation's standard positions. It builds institutional memory of how different types of work have been handled, what risks were flagged, what escalations were made, and how matters were ultimately resolved.
After twelve months, the 500th contract review is not just faster than the first. It is smarter. It knows what the first 499 taught it.
The point-tool trap is easy to fall into because the procurement case for individual tools is much easier to make than the case for a platform. A contract review tool has a clear use case, a clear demo, and a clear before-and-after metric. "Contract review used to take four hours. Now it takes forty minutes." That is a real saving, easy to present to a CFO, and entirely accurate.
The case for workflow automation is harder to present because the compounding benefits appear over time rather than immediately. The ROI of the 500th smarter contract review is significant, but it requires twelve months of consistent use to materialise. CFOs are calibrated for near-term payback. Legal teams, which historically lack operational dashboards and struggle to quantify their output, often cannot make the longitudinal case persuasively.
So teams buy tools, one at a time, each individually justified, and end up with a collection of fast task processors that share no context with each other. The legal function is faster in isolated moments. It is not smarter over time.
Map the full workflow before selecting any tool. Document every step in the most important legal process the team handles, from the moment a request enters legal to the moment it is resolved. Identify where AI can assist at each step, and evaluate platforms against the full workflow rather than individual steps
Ask vendors the compounding question. After twelve months of use, what does the system know about our organisation's legal work that it did not know when we started? How is that knowledge applied to new matters? Point tools cannot answer this question. Workflow platforms can.
Set a threshold for tool proliferation. Some legal teams have established a policy that no new AI tool can be added to the stack unless it integrates with the primary platform and contributes to the shared context layer. This prevents the accumulation of productive but isolated point tools.
The organisational dynamics that keep teams in point-tool mode
The compound zone is the operational state where an integrated AI platform is improving with every matter. It is where the institutional capital from twelve months of use is being applied to the next contract review, the next advice query, the next compliance assessment. Teams in the compound zone are not just faster than teams in point-tool mode. They are pulling ahead at an accelerating rate.
Research from Thomson Reuters' Future of Professionals 2025 report puts the operating advantage of workflow automation over point-tool approaches at 3.5 times. This is not a speed differential on individual tasks. It is the cumulative effect of institutional learning applied consistently across an entire legal function.
The compound zone requires organisational patience that most procurement processes are not designed to support. Benefits are not visible at month one or month three. They become clear at month six, and compelling at month twelve. Budget cycles, leadership changes, and the pressure to demonstrate near-term ROI all work against the patience required to reach the compound zone.
There is also an internal coherence challenge. Moving from a collection of point tools to an integrated platform means retiring tools that teams have learned to use and often like. Change resistance at the tool level is a genuine implementation risk, separate from the organisational change management required for AI adoption more broadly.
Set a twelve-month minimum evaluation horizon. When presenting the business case for platform-level AI, include outcome projections at three, six, nine, and twelve months. Make the case that the compounding benefits are real but require time to materialise, and agree that evaluation will be based on the twelve-month trajectory, not the month-three snapshot.
Track institutional capital explicitly. Measure not just task speed but the improvement in AI output accuracy over time. If the 200th contract review is materially more accurate than the 50th because the system has learned your standard positions, that improvement is quantifiable and should be documented.
Consolidate before expanding. Before adding any new AI capability, audit what the existing platform can do that is not yet being used. Teams that have moved to integrated platforms consistently find that the platform's existing capabilities are underutilised before they reach for additional tools.
The technical reality that prevents accumulation
The compounding advantage of workflow automation is technically dependent on a shared context layer. Every matter that passes through the platform contributes to a common institutional memory that informs the next matter. This requires that all work flows through, or is connected to, a single system.
Point tools, by design, do not share context. The contract review tool does not know what the matter management tool learned. The legal research tool does not know what positions the drafting tool has established. Every tool starts each task from a generalised baseline rather than your organisation's accumulated knowledge.
The practical consequence is that adding a fifth point tool to a stack of four does not produce a compounding effect. It produces a fifth isolated speed improvement, useful in itself but generating no institutional capital.
The integration problem is often discovered after procurement rather than before it. Teams buy tools sequentially, each individually evaluated and approved, and only later recognise that the collection does not add up to a system. Retrofitting integration between tools that were not designed to share context is technically complex, expensive, and often unsuccessful.
Treat integration architecture as a first-order procurement criterion. Before any tool evaluation, document the integration requirements: which systems must the new tool connect to, what data must flow in both directions, and who owns the integration maintenance. Tools that cannot meet these requirements without significant custom development should be disqualified.
Evaluate platforms on their data model, not just their features. A platform's ability to accumulate and apply organisational context depends on how its underlying data model is structured. Ask vendors to demonstrate specifically how context from past matters is surfaced in current work.
Accept that consolidation requires short-term disruption. Moving from five point tools to one integrated platform involves transitional pain. Teams that reach the compound zone have typically gone through that transition and considered it worthwhile within six months of doing so.
Source: Plexus Future-Ready General Counsel 2026 Survey, n=150 General Counsels, January 2026. External citations: Thomson Reuters Generative AI in Professional Services Report 2025; ACC/Everlaw GenAI Survey 2025, n=657; Gartner Legal and Compliance Leader research 2025.
Ready to find out where your team sits on the maturity spectrum? Take the AI maturity assessment or explore the Plexus platform.