A method with checkpoints.
Six stages. At each one you know what we need from you and what you get back. A written brief is enough to begin.
Discovery
We learn the process. You give: description, examples. You get: a clear picture of the task.
Scoping
We define inputs, outputs and what "correct" means. You get: a written scope.
Prototype
A first version on your samples. You get: real output to react to.
Build
The full workflow with its controls. You get: a working agent in a test setup.
Evaluation
Tested against your cases. You get: a report of results and known gaps.
Deployment and monitoring
Installed in your environment, then we watch logs and quality over time. You get: the agent, its documentation and issues caught early.
What is built into every agent.
Source traceability
Each value keeps its document, page and original text.
Deterministic checks
Quote matches, totals and tie-outs run as code. In our agents the model does not mark its own work as verified.
Review queue
Uncertain or important items wait for a person.
Approvals
Actions with consequences need an explicit yes.
Logging
Every run is recorded so it can be audited later.
Checks confirm a quote exists. They do not prove a conclusion is right.
Agents make mistakes. We reduce them, make them visible and keep a person in charge of decisions. We do not guarantee accuracy or regulatory outcomes.
Data stays where we agree on it.
We work with the minimum data needed, keep it in agreed locations and can run on private infrastructure.
[ security and data handling → ]Questions we are often asked.
Do we need a call to start?
No. A written description of the process and, if possible, a sample document is enough. We reply with questions or a short written plan. A call is available if you prefer one.
How is a project priced?
Every project is scoped in writing before work starts. Most projects are priced by stage: a fixed price for the prototype, then a fixed price for the build once the scope is clear. Ongoing monitoring and changes can continue as a monthly arrangement. Third-party costs such as model API usage and hosting are listed separately.
Who owns the result?
Ownership and licence terms for the deliverables are set in the contract before work starts. Our own pre-existing tools stay ours and are licensed to you as part of the delivery. Open-source components keep their own licences, listed in the handover pack.
Is our data used to train models?
No, unless you agree in writing.
Where does the model run?
Either through a cloud model API under its provider’s terms, which we share with you, or on open-source models inside your own environment when data cannot leave it.
What happens to our data after the project?
We delete it after final delivery or on your request, and confirm deletion in writing if you ask.
Can you sign an NDA first?
Yes. We prefer to sign one before you share any documents.
What if the agent makes a mistake?
Agents make mistakes. We design them so errors are visible: every figure has a source, checks run in code, and anything that matters waits for a person to approve. We do not guarantee accuracy.
Start with a written brief.
Describe the process and, if you can, attach an example later.