An AI agent in production: Opposio case study
Contesting a road fine: five risky steps
Contesting a fine means five risky steps, from picking the grounds to sending a registered letter on time. Receiving a fine notice is the start of an administrative process that many people would rather avoid: you have to understand the alleged offense, work out whether there's an admissible ground for appeal, find the right legal references, draft a formal letter without errors, and send it by registered mail within the deadline. Every step has its traps: a missed deadline, a poorly chosen ground, an awkward wording, and the whole effort falls apart.
Some drivers pay without contesting: not because the fine is justified, but because contesting costs more time and energy than the expected gain. A well-designed system turns this intimidating procedure into a few simple steps.
An honest caveat: contesting a fine never guarantees a win. The outcome depends on the case, the grounds, and the administration. A serious case study describes the system, not a promised result.
The Opposio system: three specialized agents, a control point at every stage
Opposio chains three specialized AI agents (reading, analysis, drafting), with a control point at each handoff. It's not a form that spits out a generic letter template: it's a chain of specialized AI agents, the wording taken straight from the site, where each link has a precise responsibility and hands off to the next.
The public journey breaks down into three steps:
- Read the received document. The user photographs their fine notice. The system extracts the useful information from it, with no manual entry.
- Analyze the grounds. An agent studies the case to identify the legal grounds for appeal that are worth considering and to estimate whether the effort is worthwhile.
- Draft the formal letter. A final agent produces a structured PDF letter, citing the applicable legal texts, ready to be sent by registered mail.
The key point is that there is a control point at every stage, and that the path taken varies depending on the document. A speeding offense, a parking ticket, a red light, or a phone behind the wheel don't call for the same analysis or the same letter. That's exactly what a frozen automation, a single script with a single template, would be unable to do. To understand what sets an agent apart from a plain script, see the definition of an AI agent.
The design choices that matter
Designing this kind of system is mostly a series of trade-offs. Here are the ones that shape Opposio.
A chain of agents rather than a single prompt. You could imagine one big prompt that does everything in one go. Bad idea in production: impossible to control, hard to fix when one part goes off the rails, opaque when the result is poor. Splitting it into specialized agents (reading, analysis, drafting) makes each step observable, testable, and replaceable independently.
Control points and validation. Between the steps, you check that what comes out of one agent is usable by the next. If reading the document is incomplete, the analysis doesn't start from false premises. These guardrails are what separate a demo that works once from a system that runs every day.
Reliability before magic. The goal isn't to impress with a chatty AI, but to produce a correct, usable deliverable, in a repeatable way. In the field, a boring but reliable system beats a brilliant but unpredictable one.
Public and testable. Opposio is live at opposio.com, accessible to anyone: a paid service you can try, not a screenshot in a presentation. The same principles of decomposition and validation apply to any business agent, as detailed in the guide build an AI agent with n8n.
Opposio, an AI agent in production, public and verifiable
Opposio is a real, public, testable AI agent: many talk about AI agents, few show one that actually runs, in public access, that you can test yourself. Opposio is that proof: an AI agent in production, public, and verifiable.
A few factual elements displayed by the service, to be taken for what they are. The advertised price is €19 to €29 depending on the complexity of the case, as a one-time payment, with no subscription. The site applies a billing policy: no payment if the appeal has less than a 10% chance. It should be read as a billing condition, not as a guarantee of winning. The other figures highlighted (a turnaround announced in a few minutes, a scope of 77 grounds for appeal) are claims made by the site, to be treated as such.
For me, the interest lies elsewhere: I run my own systems, not just those of clients. Designing, putting online, and operating an agent that produces a real legal deliverable forces you to confront everything a demo avoids: the awkward cases, the unreadable documents, the errors to catch, reliability over time.
What I take away from it for your projects
Opposio handles fines, but the principles have nothing specific to road traffic law. Splitting a process into specialized agents, placing a control point between the steps, aiming for repeatable reliability before the demo effect: these choices apply to any business process, whether it's processing incoming documents, qualifying requests, drafting framed replies, or orchestrating several tools.
Going for Growth is me, Fabien Cavanna, designing and operating these systems, from the first prototype to production. If you want to see what this work looks like applied to your business, take a look at the AI automation offering. And if you'd rather judge for yourself first, test opposio.com directly: it's the system described here, for real.