Technologies and tech-powered solutions have lifespans and with the advent of AI-based large language models (LLMs), CLM systems’ use of fixed metadata may be nearing the end of its lifetime. What does this prospect mean for the future of CLM systems – and those looking to acquire, deploy, or upgrade a CLM?

Why Metadata Matters

For CLM companies, fixed metadata serves as the backbone of effective contracts management. It provides a uniform structure to capture critical details such as contract parties, effective dates, renewal terms and payment obligations. This structured approach ensures data accuracy, which is essential for reporting, compliance, and analytics. Moreover, fixed metadata is critical for seamless integration with other platforms like CRM, ERP, and financial systems, enabling organisations to streamline workflows and maintain consistency across systems.

Metadata is not just about storage – it is about ensuring that accurate, actionable data is always available. Structured fields make it easy to filter or sort through large volumes of contracts without needing to process the full content each time. This reduces dependency on manual reviews by highlighting key details at a glance, enabling quicker decision-making. For instance, if a company wants to review uplift clauses or price increase language to prepare for contract renewals, it would need fixed metadata to determine which contracts:

Along Comes LLMs

AI solutions for legal, including tools incorporating LLMs, are increasingly demonstrating they can go head-to-head with human practitioners. In February, the first in-depth GenAI benchmarking study for legal LLM tools arrived, courtesy of Vals AI, an independent evaluator of LLMs. The study benchmarked tools from four vendors against the performance of lawyers in handling seven tasks, one of which was data extraction. Scores included a ‘plus/minus’ number reflecting the margin of error.

Two tools scored higher on data extraction than humans. However, it is worth noting that, in both cases, the tool’s score did not exceed the margin of error of the lawyers’ score. This suggests that the difference in scores may not be statistically significant. Moreover, two tools scored lower than humans, and both had scores exceeding the margin of error. The bottom line is that LLMs for legal are approaching the point where they consistently match or even exceed human performance – but they aren’t there yet.

LLMs for CLM

With the arrival of LLMs capable of dynamically analysing and extracting data in real time, the role of fixed metadata is now ripe for scrutiny. LLMs offer unparalleled flexibility, adapting to queries on the fly, identifying patterns and providing deeper insights. But can they fully replace the reliability and foundational consistency of structured metadata?

Not entirely—at least not yet. Fixed metadata, while rigid and limited in scope, provides consistency and reliability that are crucial for compliance, integrations, and mission-critical processes. LLMs, on the other hand, can enhance the functionality of CLM systems by offering dynamic capabilities, such as identifying risks, generating insights or filling gaps in metadata. But LLMs are computationally intensive and might miss nuances in complex legal language.

Are We There Yet?

Returning to the question at the outset of this post, the advent of LLMs is not an endpoint for the use of fixed metadata in CLMs. Rather, a hybrid approach will take hold, one that combines the best of both worlds: the accuracy and consistency of structured data with the flexibility and adaptability of dynamic analysis. For CLM users, there’s no need to panic; the results of the Vals AI study indicate that, while attention to CLM LLMs is warranted, a wait-and-see stance is prudent, at least for now. The future of CLM isn’t about choosing between metadata and AI but about synthesizing them to build smarter, more responsive systems.