The Challenge
The Risk That Clause-Level AI Cannot See
Complex commercial contracts contain two distinct categories of risk. The first category is clause-level: standard clauses that are poorly drafted, unusual liability caps, missing boilerplate that should be present. This is the risk that existing contract AI tools — trained on legal databases and clause libraries — identify competently. It is also the risk that a reasonably diligent junior lawyer will catch on a careful read.
The second category is structural. It is the risk that emerges not from individual clauses but from the combination of clauses — the interplay between a limitation of liability provision, a force majeure definition, and an indemnification clause that, taken together, create a dispute pattern that a first-year associate would not recognise but that an experienced litigator would flag immediately. This is the risk that experienced practitioners recognise because they have litigated the consequences. It is the risk that lives in the pattern, not the clause.
Generic contract AI is built on legal databases: it is trained to recognise clauses, compare them to market standards, and identify textual anomalies. It performs this function well. But it was not built by litigators, and it does not reflect a litigator's knowledge. A litigator's knowledge is retrospective: it is built from the disputes that resulted from contracts, not from the contracts themselves. The risk signatures that experienced litigators recognise are signatures derived from outcome data — from knowing what the disputed contract looked like three years before the dispute was filed.
That outcome-derived knowledge is what no legal database contains, and what no generic contract AI encodes. It exists in the heads of experienced litigators. James Whitfield had been accumulating it for 28 years.
The Approach
Building a Risk Framework from Litigation Experience
The Expert's Contribution
James Whitfield is a commercial litigator with 28 years of practice in complex disputes across construction, technology, and infrastructure contracts. His contribution to this project was not his knowledge of contract law — that knowledge is codified and accessible. His contribution was his risk stratification framework: the mental model he had developed across more than 200 major commercial cases for identifying which contract structures correlate with dispute, and why. This is a different kind of knowledge. It is empirical, pattern-based, and not derived from any written source. It exists because he has spent 28 years watching which contracts end up in litigation and reading backwards to understand what the warning signs were.
Mapping the Risk Signatures
Knowledge-encoding sessions with James were structured around his active case history and his portfolio of resolved disputes. Rather than working from contract templates or legal principles, sessions worked backwards from outcomes: for each major dispute category in his practice — payment disputes, scope disputes, liability disputes, force majeure claims — the question was: what did the contract look like? What clause combinations, what jurisdictional factors, what counterparty behaviour patterns were present in the contracts that generated dispute versus the contracts that did not?
This produced a fundamentally different output than legal-database training. James's risk signatures are not derived from what clauses are "standard" or "unusual" in isolation — they are derived from what clause combinations he has seen, repeatedly, in the months before a dispute is filed. These are the signatures that appear in contracts before they become problems, and they are the signatures that generic tools, trained on static legal databases, are structurally unable to identify.
Structuring the Framework Across Contract Categories
James's risk knowledge was not uniformly distributed across all contract types. It was concentrated in the categories where his practice had been most active: technology services agreements, construction contracts, infrastructure concession agreements, supply chain contracts, joint venture agreements, and project finance documentation. Sessions were structured category by category, mapping the specific risk signatures relevant to each — including jurisdictional factors (the risk pattern in a NSW construction contract differs from the pattern in a Queensland contract in specific, documented ways) and counterparty type (government counterparties, international contractors, and domestic SME suppliers each generate different dispute patterns).
Validation Against the Case Record
The encoded risk framework was validated by applying it retrospectively to a sample of James's resolved matters. For 45 contracts that had proceeded to dispute and 30 that had not, the encoded system was used to generate a risk assessment — and compared against the actual outcome. Cases where the system failed to flag the dispute-generating risk patterns were examined in detail with James, and the framework was refined to address the gaps. This iterative validation process is what produces a system grounded in actual litigation outcomes rather than legal theory.
The Product
A Contract Risk Report Built on Litigation Logic
The resulting product reads commercial agreements and produces a structured risk report aligned with James Whitfield's litigation-derived risk framework. The report is not a clause-by-clause audit — it is a risk analysis that operates at the structural level, flagging the specific patterns that James's 28 years of practice have identified as predictors of dispute.
Each risk flag in the report links to the underlying litigation logic: not "this clause deviates from standard form," but "this combination of clauses — specifically the interaction between the payment milestone structure and the variation mechanism — matches a pattern present in 14 of the 23 construction disputes in the encoded case base, and here is why that matters." The report explains the risk in terms a junior lawyer can act on, grounded in reasoning that a senior partner would recognise.
The system covers 83 distinct risk pattern signatures across the six contract categories in James's practice. It does not attempt to cover all commercial contract risk — it covers the risk that James Whitfield's practice has made him unusually qualified to identify. This is a feature, not a limitation: a system built on one expert's 28 years of litigation experience in specific domains is more useful to practitioners in those domains than a system trained on every commercial contract ever written.
The system is explicit about its own scope. When a contract falls outside the encoded categories, it says so. When a risk pattern it identifies is at the edge of the validated case base, it flags that too. Epistemic honesty is built into the output design — the system communicates the confidence and the limits of its analysis, not just its findings.
28
Years of litigation judgment — systematically mapped for the first time
83
Risk pattern signatures encoded across 6 commercial contract categories
200+
Major commercial cases underpinning the risk framework — each flag linked to documented case precedent reasoning
"I've seen the same mistakes in contracts for 30 years. Now there's an AI that sees them too — and explains why they matter."— James Whitfield, Senior Commercial Litigator
Your expertise deserves a better vehicle.
If your practice contains knowledge that has never been systematically encoded, Praxa can build the product that makes it operational.
Work with Praxa