The brief, when it came in from a regional trading group late last year, seemed simple enough: build an AI system that reviews supplier contracts and flags anything unusual before someone signs. We've done document extraction and clause analysis before. We said yes and moved on to scoping.
What we actually built ended up being something considerably more interesting — and considerably more useful — than a contract review tool. Here's what happened and why it matters if you're thinking about this category.
The obvious part: extraction and risk flagging
The first piece is what most people imagine when they hear 'contract AI.' The agent reads the contract — in Arabic or English, or both, since supplier contracts in the UAE often have both versions and they don't always match — extracts the key commercial terms, and flags anything outside the company's standard parameters. Payment terms over 90 days. Liability caps below a certain threshold. Exclusivity clauses the team hadn't noticed. Auto-renewal provisions buried in schedule 7.
That part took about six weeks and worked well. But during the scoping conversation, the CFO asked a question that sent the whole project in a different direction: 'Once you know what's in the contract, can the system track whether we're actually complying with the obligations in it?'
The less obvious part: obligation tracking
This is where it gets genuinely useful. Most contract management problems aren't in the signing — they're in the forgetting. A contract signed eighteen months ago has a price review clause that triggers next month. A supplier agreement has a notification requirement that needs to be exercised sixty days before renewal. A distributor contract has volume commitments that, if you check today, you're already behind on.
We built a second agent layer that reads the extracted obligations and maps them to calendar triggers, volume thresholds, and real-time data from the company's ERP. It monitors continuously and surfaces alerts — not 'you have a contract expiring in 30 days' but 'your volume commitment to Supplier X requires 200 more units this quarter and your current run rate puts you short by around 40. Here are the three options and their cost implications.' That's a different product.
Practical agentic AI insights for UAE and GCC leaders — no spam, just what's actually working in production.
The part we didn't see coming: commercial analytics
Once you have a structured, queryable representation of your entire contract portfolio, a whole class of questions becomes answerable that previously required days of manual audit work. What's our total liability exposure under force majeure across active supplier contracts? Which of our distribution agreements have price escalation clauses tied to CPI, and what's the likely impact at current inflation rates? How many active contracts have dispute resolution clauses that require arbitration in a jurisdiction outside the UAE?
These are questions that a large enterprise's legal and commercial team would love to be able to answer quickly. Before this kind of system exists, they often can't — not because the information isn't there, but because it's locked inside individual document files that nobody has time to systematically read and cross-reference.
What the architecture actually looks like
Three agents, each with a distinct role. An extraction agent reads the document and produces a structured representation of every meaningful clause — commercial terms, obligations, conditions, notice requirements, termination provisions. A monitoring agent watches for triggers against live business data and the calendar, surfacing alerts when obligations need action. An analytics agent answers ad-hoc queries against the portfolio. They share a common data layer and a single audit trail, so any action taken or recommendation made is traceable back to the specific contract clause it came from.
Where this is heading in 2026
The next frontier we're seeing clients ask about is outbound — an agent that doesn't just read supplier contracts but drafts negotiation responses based on company policy, flags clauses the company has successfully removed in past negotiations, and tracks which terms have been accepted and rejected across the portfolio over time. That's a more sensitive workflow — you're closer to actual legal action — but the demand is real and the technology isn't the limiting factor. The limiting factor is governance design, which is a recurring theme in everything we build.