At Avogado Legal s.r.o., CEO Jakub Tomoszek and Head of Operations Dana, along with the back-office team, handle a huge volume of legal inquiries and client service. The work was honest and well done, but it ruthlessly consumed routine time – processing a single inquiry took an average of 25 minutes.
During their initial attempts to deploy AI, they ran into a classic problem of today's companies: informal attempts with freely available tools failed ("I tried it, it doesn't work, it generates nonsense"). Without context, the right model, and data, AI produced unusable outputs. Furthermore, in complex AI projects, C-level management often faces people's fear of making mistakes, losing control, and losing efficiency.
Key Results
- 75% time saving: The average time to process one client request dropped from 25 minutes to 6 minutes.
- 80% success rate without edits: 4 out of 5 email responses prepared by the AI agent require no human correction (only a quick visual check and sending).
- Delivery speed of 2 hours: A functional version of the agent (v0.2) connected to Google Drive and templates was created in just 2 hours of net time.
- Culture and attitude change: A transition from rejecting AI ("It doesn't work") to people from operations actively proposing new features themselves.
Instead of a massive IT project with huge assumptions, purchasing expensive software, or a year of development, we chose a completely different path: letting people grow around the technology.
The goal was not to naively replace people or introduce complete automation right off the bat. We chose an agile approach based on rapid prototyping and a Human-in-the-loop architecture (the human in the role of controller and quality guarantor). Thanks to this, users see how AI works step by step and lose their fear of it.
The transformation took place in the form of extremely fast iterations in close cooperation with operations:
- Version 0.1 (approx. 1 hour of work): We created a simple prompt without deeper context. The AI initially made mistakes (4 out of 5 answers required correction), but the Head of Operations, Dana, saw real potential for the first time. She immediately suggested concrete improvements: "This can really help us! Could it also automatically save attachments and rewrite data into a template?" The key mental shift had arrived.
- Version 0.2 (another 1 hour of work): We let the AI analyze a year's history of client communication and created a "rookie manual" for the agent from it. We connected the agent to Google Drive so that it saves conversations, downloads and analyzes documents from clients, and automatically inserts the extracted data into a template.
Results and change
A key factor in the transformation's success was a fundamental change in the behavior of managers and senior architects:
- Stepping back from execution: Seniors/leaders consciously stepped back from the daily decision-making of the teams. They stopped functioning as arbiters of the correctness of technical solutions and micromanagers of tasks.
- Role of mentors and coaches: Their main activity shifted to defining the technological strategy, removing systemic obstacles, and supporting the competency development of more junior colleagues.
- Acceptance of controlled risk: Leaders managed to overcome the protective complex. They let teams conduct their own analyses and bear responsibility for mistakes, which was a necessary step to build real autonomy.
The impact was immediate and measurable. The routine operations that previously wore the team down shrank to a quarter.
- Dramatic leap in efficiency: The average time per request dropped from 25 to 6 minutes.
- High accuracy: For 80% of emails, the team only needs to visually check and approve the response proposed by the AI.
- Cultural transformation: Because Dana saw the agent growing under her hands in small steps, the technology stopped being incomprehensible and scary. Today, the operations people themselves are churning out ideas for the agent's further development.
