A leading IT integration company specializing in tender procurements and installation services was facing significant inefficiencies in its workflow. The manual process of analyzing complex technical specifications, matching them with compliant products, and generating commercial proposals was time-consuming, prone to errors, and a major bottleneck. The company sought to leverage Artificial Intelligence to automate this process, aiming to create a system that could act as a technical expert.
The Challenge: Overcoming Data Silos and AI Limitations
The core challenge was to develop an AI-driven solution that could accurately and reliably generate tender-compliant commercial proposals. The process involved several complex requirements:
- Prioritized Sourcing: The system first needed to check for suitable products in the official registry of the Ministry of Industry and Trade to ensure regulatory compliance.
- Internal Inventory Check: If no compliant product was found in the government registry, the system then had to search the company’s own product inventory.
- External Search: As a final step, if no product was found in the internal or official databases, the system had to search the broader internet for suitable alternatives.
- Accuracy: Early experiments with general-purpose AI models proved unsuccessful. The models often “hallucinated,” providing incorrect or non-compliant product recommendations, and struggled to process large, domain-specific files like the 50 MB Excel sheet from the Ministry’s registry. They failed to correctly interpret the hierarchy of data sources, often ignoring the strict requirement to prioritize the official registry.
It became clear that a simple, off-the-shelf solution was not viable.
First of all, I explained to the company’s representative that we should split this task into a few steps in order to make the system.
That’s the system I prefer to use when I work with my clients:
It became clear that a simple, off-the-shelf solution was not viable. A more robust, structured approach was needed, one that could guide the AI through a logical sequence of checks while leveraging a specialized knowledge base—a concept known as Retrieval-Augmented Generation (RAG).
The Solution: A Sequential AI Agent in n8n
Here is a breakdown of the workflow:
Initial Query – Ministry Registry: When a new tender’s technical specification is received, the n8n workflow triggers the AI agent. Its first action is to query a database containing the data from the Ministry of Industry and Trade. This step ensures that any proposals generated are fully compliant with government procurement regulations from the outset.
For this, a dedicated vector database was created in Pinecone. This database was populated with the company’s entire product inventory, turning unstructured product data into a highly searchable knowledge base. The agent queries this vector database to find a match from the company’s available stock.
This tiered approach ensures that the AI’s search is both efficient and logical, always prioritizing the most reliable and compliant data sources before widening its search.
The Outcome
- Increased Efficiency: The time required to generate a commercial proposal was drastically reduced from hours to minutes.
- Improved Accuracy: By structuring the search process and using a dedicated knowledge base, the instances of AI “hallucinations” were eliminated, resulting in highly accurate and compliant proposals.
- Enhanced Scalability: The automated system allowed the team to handle a significantly higher volume of tenders without a corresponding increase in manpower.
This case study demonstrates that while general AI models have limitations, they can become powerful business tools when integrated into a structured workflow with specialized data sources like a Pinecone vector database.
This project serves as a successful example of how to overcome the common challenges of AI implementation and build a practical, effective solution for a complex business problem.