By Jim Parke |
A contract system can find an expiring insurance certificate, flag an overdue notice and compare a proposed clause with an approved form in seconds. That is useful work. It does not decide whether a supplier's proposed change is acceptable for a programme that must stay operational for twenty years, or whether enforcing a right today will make an already strained relationship harder to repair tomorrow.
As automation moves through the contract lifecycle, the commercial role can become more consequential. The manager who once spent hours tracking dates may have more time to examine the deal, the changing context and the people who must make it work. That opportunity is real only if the organisation knows which decisions require human judgment and supplies the evidence to make them.
Let systems handle the repeatable work
Drafting from an approved template, locating inconsistent terms, tracking obligations and alerting a team to a renewal date are natural candidates for software support. A searchable record of agreements can also help a sponsor see where several contracts depend on the same supplier, technology or approval.
Those tasks are not trivial. An overlooked notice can close off an option; a missed renewal can interrupt a service. Better records and timely prompts give the programme a firmer basis for action. In a report sponsored by contract software vendor Agiloft, World Commerce & Contracting's Tim Cummins proposes assessing AI tasks by complexity, frequency, interconnectivity and cost of failure. The report suggests AI can handle frequent, lower-risk tasks with limited oversight, while negotiation, risk assessment and dispute avoidance need human interpretation and validation. That is a useful framework, provided the organisation tests what a tool can actually do in its own workflow rather than assuming a general performance claim applies to every contract.
The manager still needs to know what the system is looking at. Are amendments, side letters and negotiated departures included? Is the contract information complete, current and classified correctly? If a search returns no exception, does that mean none exists or that a document was never captured? Automation can make a weak record faster to use without making it reliable.
Decide which risk the organisation is willing to take
Software may identify that a liability cap differs from the standard clause. Whether that difference is acceptable depends on the exposure, the supplier's capacity, available insurance and the value of the proposed service. It depends too on the sponsor's alternatives and on which risk the programme is able to manage better than the supplier.
The commercial decision cannot be made by comparing text alone. A clause giving the sponsor broad termination rights may look protective, yet exercise of the right could leave an essential service without a replacement. A supplier's proposed flexibility might create useful room for innovation or undermine a critical performance obligation. The manager must ask what the words will cause each party to do when conditions change.
This is why contract review should include delivery, finance, operations and legal expertise at the right points. Their contributions will differ, but the final recommendation should state the trade-off clearly: what is being accepted, why, who owns the remaining risk and what event would trigger a different decision.
Keep negotiation connected to the programme's purpose
Automated comparison encourages a focus on departures from a preferred form. Some departures matter greatly; others are bargaining positions that distract from the central result. A capable contract manager can distinguish between them and spend negotiating effort accordingly.
The starting question is what the programme must be able to do. If several suppliers must test a system together, access and coordination may matter more than an apparently favourable individual milestone. If the programme expects rapid change, the method for pricing and approving variations may matter more than perfecting a list of distant contingencies.
The manager also has to interpret what the other party is trying to protect. A demand for a broad exclusion may reveal that the supplier cannot price an uncertain interface. A request to retain intellectual property may reflect a legitimate business model. Understanding that interest can lead to a narrower, workable allocation instead of a prolonged exchange of standard positions.
Check the tool as well as its answer
Using AI to analyse contract material introduces another set of decisions. What information is being entered, and who may access it? Can the tool show the source text behind a finding? How are errors detected? When should a recommendation be reviewed by a person with the relevant legal or technical expertise?
Australia's National AI Centre sets out six essential practices in its current Guidance for AI Adoption, including deciding who is accountable, testing and monitoring systems, and maintaining human control. The US National Institute of Standards and Technology's voluntary AI Risk Management Framework, version 1.0 and currently under revision, likewise calls for defined human oversight and management of risks from third-party AI software and data. For contract teams, that means a clear owner for each tool, tested use cases, a way to correct errors, a plan for tool failure and a record of who accepted a material recommendation. The Australian Government also has a separate, mandatory AI policy for non-corporate Commonwealth entities; it should not be confused with the National AI Centre's general guidance.
One particularly important test is whether staff retain the ability and authority to disagree with the system. A flagged clause may be harmless in context. A clause marked as ordinary may conceal a programme-specific risk. Cummins' report also warns of fragmentation costs when an AI output needs significant human reworking before it can be used. If review becomes a ritual of accepting the automated answer, the apparent oversight will add little protection.
Maintain the relationship after signature
The executed contract is the beginning of a long operating relationship. A manager must ensure notices reach the right people, changes are recorded, performance data is understood by both sides and emerging problems are raised while options remain. They need to distinguish a disagreement over price from a technical problem that the parties can solve before arguing over who will pay.
Digital tools can show trends in variations and performance. They cannot create candour between organisations or decide when the sponsor should use a contractual right, negotiate a solution or change course. Those judgments require an understanding of the programme's purpose and of the consequences for the people delivering and operating it. Our article What should a major project contract decide before things go wrong? considers how the agreement can support those decisions from the start.
The most useful question for a sponsor is therefore not how many contract-management tasks can be automated. It is what its people will do with the time and information automation gives them. A system can tell a manager that a decision is due. The manager must still know what is at stake, who should be heard and whether the proposed answer will help the programme deliver what it promised.
Further reading
- Tim Cummins, World Commerce & Contracting, Humans and AI: Together, Transforming Contract Management (sponsored by Agiloft, 2025).
- National AI Centre, Guidance for AI adoption: implementation guidance.
- US National Institute of Standards and Technology, AI Risk Management Framework: Core (version 1.0, 2023; revision in progress).
- Digital Transformation Agency, Policy for the responsible use of AI in government (version 2.0, effective 2025).