What your committee will ask

No. Your team stops chasing cases and decides the exceptions, the cases no rule covers. The agent moves the case forward and stops where a person is needed.

Your team. It sets the goal, the rules and how far the agent can decide, and it's the only one that changes the rules: the agent doesn't change one on its own. When a case comes up that no rule covers, the agent stops and hands it to someone on your team with what it tried and what's missing. Your team decides the case and, when reviewing the log, writes the rule that was missing.

It depends on what you tried. If it waited for someone to ask it something, the case still depended on whoever pushed it. Here the agent takes the case as soon as it comes in, moves it forward in your systems and stops only where a person is needed. If what you tried already did that, tell us about it on the call.

It can happen. Every step goes into the case log (system, action, channel and result) and your team reviews it. The rules your team defined always run the same way (deterministic execution): if a rule was wrong, the error repeats until your team fixes it. The language model interprets messages and drafts replies; it doesn't decide your rules. If a piece of data is missing or a system doesn't respond, it stops and hands the case to a person; it doesn't make up data that isn't in the system.

On Neuraan's infrastructure, which runs on AWS in the United States on enterprise cloud providers, with its own isolation for each agent. If your contract requires it, we set up a dedicated environment. Neuraan doesn't use your data to train models. Your team issues the access, and the agent signs in with its own user, not a person's. Details on providers, encryption and the language model are on the security page.

No. The agent works through the systems you already use: it signs in with its own user and with the access your team issues. Depending on the process, it connects to your systems, to file folders or to portals. Before we start, we put in writing which systems it touches.

Four things, and the first one usually takes a while. Access (users, permissions or credentials) issued by whoever administers your systems: request it early. Real cases in volume, because with few cases, few exceptions show up. Someone from operations to validate, not full time. A team that knows exceptions nobody wrote down will come up: finding them is the point.

It doesn't solve your entire department. It solves one well-scoped process and leaves you its model (rules, exceptions, who decides), plus a report with the results of your cases and a record of every exception and how it was resolved. It also doesn't decide what no rule covers. If your process isn't a fit, we'll tell you on the call.

We plan for four to six weeks, counted from when we get your access, for you to see the agent working with your own data, and the timeline can shift. Your team issues the access, and that weighs the most; real cases and having someone to validate also count. Every week you get a written update on progress and exceptions, and every timeline change is put in writing.

Yes, the pilot has a cost: it works with your systems and your real cases, not a demo. We don't publish the amount: we discuss it on the call, with the scope in front of us.