Auchan Tech showed how AI saves up to 70% of a systems analyst’s time on routine tasks

Auchan Tech published a practical analysis of how generative models are already being integrated into the daily work of a system analyst. The author's main conclusion is simple: AI does not relieve the analyst of responsibility, but is capable of significantly accelerating the preparation of questions, diagrams, and requirement drafts.

Three Work Scenarios

The material breaks down three typical tasks that an analyst encounters in almost every product cycle: requirements gathering and clarification, process modeling, and Use Case preparation. The author compares what models produce with a simple request versus a structured prompt with role, context, constraints, and expected response format. The examples use ChatGPT, Qwen, and DeepSeek, with emphasis not on the model's "magic," but on what the analyst must provide as input.

• List of clarifying questions for the stakeholder on a personal account
• PlantUML diagram of user registration in an online store
• Use Case for sorting reviews on a product card
• Comparison of basic and detailed prompts by result quality

Where AI is Already Useful

The most vivid example concerns requirements gathering. For a query about a personal account with order viewing and contact data editing, a basic prompt produced a lot of noise: general questions, unnecessary topics, and weak structure. But when the author demanded focus on business logic, roles, constraints, alternative scenarios, and dependencies, the models began producing a much more usable list. According to the author's assessment, this method can yield approximately 80% of needed questions in five minutes instead of forty minutes of manual preparation.

A similar picture emerged with diagrams. If you simply ask for PlantUML code for user registration, the model draws too generic a skeleton. When the prompt added system participants, mandatory fields, alternative scenarios, HTTP status codes, errors, and formatting requirements, the result became noticeably closer to something presentable to an architect. The diagram framework was assembled in 10–15 minutes instead of approximately 30, but exceptions and business rules still had to be clarified manually.

The effect with Use Cases is the same: a template query produces a vague description, while a detailed one produces an almost technical draft with preconditions, postconditions, main scenario, alternatives, API parameters, and response codes. Particularly useful is that the model can immediately break down the scenario into client-side and server-side logic.

But here the risk of hallucinations becomes apparent fastest: AI easily adds fields, checks, and rules that don't actually exist in the system.

"AI will not replace the analyst, but can save them up to 70% of time

on routine work."

Where Risks Begin

The article proceeds fairly soberly through the limitations. First, the quality of the output directly depends on the quality of the input: a weak prompt almost certainly yields a weak result. Second, the model doesn't know the company's internal context, so it won't guess real integrations, regulatory constraints, and agreements between teams. Third, when working with cloud services, security becomes an issue: uploading sensitive data, internal schemas, and unreleased requirements to them is risky if the company has no clear rules and a protected perimeter.

Separately, the author reminds that final responsibility is not delegated to the model. Logic verification, requirement validation, API detail correctness, and TZ sign-off remain the human's task. Therefore, the best use case for AI here is not "write the document for me," but "make the first framework, highlight gaps, suggest alternatives, and speed up the routine."

The article concludes with a short prompt checklist: set a specific goal, assign a role to the model, describe the context, fix the response format, require accuracy, ask for clarifying questions, and always manually check the result.

What This Means

For system analysis, AI is transitioning from experiment mode to a working tool, but only in combination with a strong human at input and output. The winning teams are not those who simply "plugged in GPT," but those who learned to turn models into a fast draft for requirements, diagrams, and technical scenarios without losing quality control.