A tool takes shape through a method, then improves through real work.
When we discuss artificial intelligence with clients, we also draw on our own work. At Progymédia, Philippe and Sébastien used AI to build a new internal management system, then continued improving it with the team.
The starting point was concrete: an older system built on FileMaker, surrounded by connected software, and administrative tasks that took up time. We wanted to rethink that setup around our operations, with the people who know them.
Start with operations before choosing tools.
An internal system eventually tells the story of a business. It contains useful habits, shortcuts, rules known to a few people and connections added as needs arose. Replacing screens without understanding that history can simply move the difficulties elsewhere.
Our monthly close, which previously took two to three days, was one of the situations to examine. That duration described our starting point. It prompted us to look at the steps, the information needed and the handoffs between tools, rather than simply looking for a more modern interface.
Build with AI and a clear direction.
Philippe and Sébastien completed an initial build in roughly three months with AI assistance. That timeframe belongs to our project, with our knowledge of our own needs and our development experience. It is not a standard schedule for rebuilding another company’s management system.
AI can help explore approaches and produce components that we can then examine. It does not automatically know every reason behind an internal rule. The team must explain the need, assess the suggestions and decide what deserves a place in the solution.
Involve the people who will do the work.
Our team then contributed to the tool’s continued development. That step matters: the people using a system encounter situations its designer may not have anticipated. An unclear field, information that is hard to find or a recurring exception becomes a design issue.
Useful feedback describes a task and its context. What was the person trying to do? Which information was missing? At what point did they need help? These questions turn a general impression into an improvement that can be discussed, built and checked.
Keep the working loop visible
01Understand the taskIdentify the information, rules and person responsible.
02Build and reviewUse appropriate tools, then examine the solution in context.
03Try and adjustGather feedback from the team and choose the next improvements.
AI can support this work. Each step retains a purpose and human review.
Keep human judgment where errors would have consequences.
Building with AI does not mean giving it every decision in the finished system. Development assistance and operational automation are separate subjects. For an action involving a record, an amount or a team decision, we need to define what can be suggested, carried out or approved.
Before a similar project, we also recommend clarifying which data may be used with AI tools, what access is needed and which checks should take place. Those choices depend on the environment and the sensitivity of the information, and should involve the people responsible.
Assess the result in everyday work.
The speed of code production tells us something about development. On its own, it does not tell us whether a tool is useful. To assess what comes next, the questions should concern actual use: is a task clearer, is information easier to reach, is a correction easier to make?
To compare a situation before and after, we need to observe the same task under clearly understood conditions. The lesson we take from this project is the value of moving between design and use. A first version opens that work; it does not settle every question.
An internal project whose lessons inform our work.
Our internal management system is a separate project from Boutik, designed for our own company’s needs.
What we can share with you is an approach: understand operations, choose an initial scope, build with suitable tools and learn with users. If an administrative task is slowing your team down, let’s start with what actually happens before deciding what role AI could play.
Before starting your own project
✓ Choose a recurring task and describe how it works today.
✓ Identify sensitive information and access rules that need clarification.
✓ Invite the people using the system to review early trials.
✓ Define which decisions require human review.
✓ Agree on an observable result to compare before and after.
