Copilot Studio agent for comparing payments-provider contracts

A fintech serving e-commerce in 4 EU countries worked with 7 payment providers (Stripe, Adyen, PayU, Mollie, plus local providers). Each contract ran 60–90 pages with different fee models: interchange++, blended, tiered pricing, fixed fees per transaction, fees per MCC, chargeback fees, FX margin, refund fees. Every provider wrote it their own way.

Every quarter the finance team (CFO + 1 controller) renegotiated terms with half the supplier base — either at the provider's initiative (price-list change) or their own (push for a lower rate). Analyzing a new proposal against the current contract took 8 hours per contract. At 3–4 negotiations per quarter that's over 30 hours — almost a full week of controller time on comparisons alone, before the actual negotiation started.

Another wrinkle: transaction data from the company's internal payments-orchestration platform wasn't wired into contract analysis. The controller had the contract in PDF and aggregated data in Excel, but manually checking the firm's real cost under the new terms against the last 3 months of volume was tedious and error-prone.

Weeks 1–2: workshops with the CFO and controller, building a structural fee dictionary (28 fee types, 12 pricing models, mapping every clause used by the 7 providers). Without that dictionary even the best language model gets lost in fintech specifics — terms like „interchange++” and „blended pricing” sound similar but are economically very different.

Build took 4 weeks. Microsoft Copilot Studio as an agent in Teams: a controller drops in the PDF of a new contract, and the agent returns a structural representation in 2 minutes (pricing model, fees per category, special clauses, risks). Azure OpenAI parses the contract behind the scenes and compares it to the structural version of the current contract from SharePoint. Integration with the internal payments-orchestration platform: the agent calls the platform's API, pulls the company's real transaction volume from the last 3 months broken down by country, MCC and payment type, then simulates the firm's cost under the proposed terms vs. the current ones.

The last 1.5 weeks were a pilot on 3 providers (Stripe, Adyen, PayU) with controller verification. The agent generates a delta table: for each transaction category it shows current cost, new cost, EUR/month delta, % delta and flags the biggest changes. The controller accepts or corrects. After 3 tuning cycles the agent hit 96% accuracy on identifying key changes — the remaining 4% are specific clauses requiring legal commentary.

After the rollout, the CFO and controller complete the same quarterly work in half a day instead of a week. The recovered time moves into the actual work: provider conversations, legal review of atypical clauses, scenario modeling for the next year. The controller notes that recommendation quality to the CFO has gone up sharply — there's time for a second and third iteration of the analysis instead of stopping at the first.

Real financial impact in the first year post-rollout: a five-figure EUR saving from better-negotiated terms. The agent surfaced details that had been slipping through: one provider had been quietly raising its FX margin a small percentage each quarter, compounding into a meaningful 12-month increase. Another had a „minimum monthly volume fee” clause the company wasn't aware of — the agent pulled it from a 73-page contract and put it on the delta table.

An unexpected effect: providers started treating the company as a higher-tier negotiator. Showing up to a call with a concrete scenario table built on the company's own transaction data shifted the dynamic noticeably. Two providers proactively offered better terms before the company even pushed for renegotiation.