Drafting a new agreement based on a template and specific arrangements with a counterparty, all the way to a first draft: that is the task at hand here. Not the negotiating, not the signing, but the assembling of the text. The answer is partly. AI can do a substantial part of the writing work today, but the core of the task remains dependent on human judgment.
Three axes determine the picture: judgment, cost of error, and compliance. Drafting a contract is not filling in a form. The template provides a structure, but which clause goes where, how a deviating arrangement with the counterparty is translated into legal language, and whether a provision is needed in this context or not: that requires judgment that cannot be fully captured in rules. On top of that, a mistake in a contract is not small. An incorrectly formulated liability clause or a missing notice period often only becomes visible when it is already too late. And contracts touch on compliance: sector rules, standard terms that are legally required, internal approval lines. Together, these three mean that a text proposal from AI must always be reviewed by a human before it becomes a draft that may be sent to the counterparty.
The remaining axes do point in the direction of automation. Structuredness is low, but that is precisely what language models are good at: converting a moderately structured source into a fixed form. Physically there is no obstacle at all, the volume is large enough for most companies to justify a standardized approach, and client contact is limited to the moment the draft is shared, not to the drafting itself.
A contract manager receives an email: new supplier, standard purchasing terms, but with a deviating payment term and an additional confidentiality provision. Today this often means: finding the right template, manually processing the deviations, and checking whether the rest of the text still matches the new arrangements. AI can generate the first draft here based on the template and the email as input, including the adjusted clauses in the right place. What remains is the check: does the payment term match what was actually agreed, is the confidentiality provision not too broad or too narrow, and is there a reason to deviate from the standard text that needs to be recorded. That last part is human work, with AI as the initial drafter instead of the contract manager starting with a blank document.
At a company with a small number of widely varying contracts, where almost every agreement is a unique negotiation, the judgment required is higher and the gain from AI is lower: there is simply little repetition to practice on. At a company with many similar contracts, such as a SaaS company with hundreds of customer agreements based on a fixed template, this is different. There the pattern is predictable, the deviations are limited and can be categorized, and an AI-drafted concept structurally saves time. The precondition then is that the template library is up to date: an outdated standard clause that AI keeps repeating is a risk that only surfaces once there is a dispute.
The shift is not in replacing the contract manager, but in relocating time. Where that time used to go into assembling the first text, it now goes into checking and adjusting it. That is a different skill than writing: it is assessing whether an AI proposal is correct, and knowing when a deviation must escalate to someone with more authority. That same shift plays out across the broader contract process. Once a contract exists, it must end up in the right place and be findable, renewals and notice periods must be flagged on time, and sometimes it must first be checked whether the counterparty is not on a sanctions list before signing. These are each tasks with a different profile than the drafting itself, with their own balance between what AI can handle and what requires oversight.
The text. Assembling a draft based on a template and the supplied arrangements is achievable for AI today, provided there is a reliable and up-to-date library of templates and standard clauses underlying it. What AI does not do here: decide that a draft is ready to be sent. That final check, with a clear rule about when a deviating clause must first go to a lawyer, remains with a human. That is not an intermediate step that disappears once the technology improves; it is the reason why this task scores low on judgment, cost of error, and compliance, and that does not change with a better model.
Whether this means that hours can already be freed up in your own contract process now depends on how many of your contracts rely on a fixed template and how strict the approval lines are. The free quickscan gives an indication of that: twelve questions, no account needed, resulting in an indication of what portion of the hours in this profile can be taken over by AI today. The full work scan, which maps the work of an entire company including the fte capacity that frees up, is still under construction.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.