Client: Ostrowski Online Trade
Case study
The AI assistant has reduced the time needed to prepare responses to quote requests by 50% – it now helps salespeople handle approximately 3000 requests per month.
Range
Implementation of the AI Assistant
Is your sales team drowning in requests for proposals and technical questions? Instead of hiring more people for the Sales Department, DobreGniazdka.pl, LuxMarket.pl, and OtoLampy.pl decided to implement an AI Assistant that understands customer inquiries, technical nuances, checks inventory, and prepares offer drafts in Gmail. The result? Freeing up half the sales team's time.
Challenge
graduation e-commerce B2B/B2C
without drastically increasing employment
In every growing B2B and B2C e-commerce business , there comes a point when the customer service team can't keep up with the volume of inquiries. In the case of Otolampy.pl , LuxMarket.pl, and Dobregniazdka.pl , the customer service office dealt with approximately 3000 emails per month.
Every message—from simple order status inquiries to complex quote requests—required manual analysis, catalog searches, and manual responses. Our goal at Cognize was to relieve the sales team of repetitive tasks and accelerate quote processing.
A game changer in the B2B bidding process
Automatically create carts and offers right in Gmail
This implementation goes beyond standard chatbots. We've created an intelligent AI sales assistant using a human-in-the-loop model. How does it work in practice?
01. Intention Analysis: AI recognizes that the customer is asking about a set of products in the email (often writing in everyday language).
02. Live Verification: The system connects via API and MySQL, checking inventory and availability in real time.
03. Action in the store: The assistant physically creates a ready-made basket in the store system.
04. Draft in Gmail: The merchant receives a ready-made draft response that includes a professional product list in the form of tiles and a direct link to pay for the finished cart.
Effect? The salesperson doesn't search for products in a catalog. They check the AI's note, add any discounts, and send the offer in seconds.
Case Berker Q: How did we teach AI to distinguish technical nuances in customer inquiries?
The biggest technical challenge was the specific nature of the electrical equipment industry. While lamp names in lighting are unique, socket series (e.g., Berker Q.1, Q.3, Q.7) differ in minimal technical details. Traditional AI models initially reported incorrect products, unable to distinguish between such similar parameters.
Agentic Workflow Solution from Cognize
Instead of a single query to the AI, we broke the process down into multi-step logic.
Extraction and normalization
AI extracts only the features from the email (series, color) and translates everyday language into store nomenclature.
Multi-variant Search
The system repeatedly searches the Pinecone vector database, taking a large sample of data.
Verification
The final AI stage compares the results to the customer's strict specifications, eliminating selection errors.
Thanks to this approach, the precision of product selection has increased to 90%.
Modern technology stack
What should it look like to automate customer offers?
Security
When implementing the AI Agent, we focused on an architecture that provides business security – no dependence on a single supplier and maximum precision.
01. Process Orchestration: n8n (the heart of automation)
Instead of hard-coded code, we chose a low-code platform. This is the system's "nervous system," allowing for rapid changes to business logic without costly programming effort. This guarantees high flexibility and easy scalability of processes in the future.
02. Artificial Intelligence: Interchangeable LLM Models (Resilience to Market Changes)
An agnostic architecture that eliminates the risk of vendor lock-in. The system allows for seamless switching between models (e.g., OpenAI or Gemini) – guaranteeing stability regardless of changes in the AI market.
03. Knowledge Base: Pinecone (Vector Database) for instant product matching
Semantic search that "understands" context, not just keywords. This allows AI to accurately distinguish technical nuances (e.g., Berker Q.1 vs Q.3), eliminating errors in product selection that are common in standard search engines.
04. Frontend: Dynamic HTML/CSS Email Templates Transferring the UX experience from the store directly to email. Instead of simple links, we've created aesthetically pleasing product pages with photos and an "Add to Cart" button. We're shortening the purchase path, which directly translates to higher conversions.
Results of implementing the AI Assistant in the electrical engineering industry
Automation that brings real savings
The six-month process (3 months of prototype building and 3 months of fine-tuning based on team feedback) brought tangible business benefits:
• Time: Reducing the response time to emails 50 %.
• Quality: Standardized, professional and comprehensive answers that build brand authority.
• Comfort: Eliminating the most tedious element of a salesperson's work – browsing through product catalogs.
• Security: Thanks to internal notes from AI, employees are informed about low stock levels before they send an offer.
Partnerships
Are you planning a similar implementation?
An AI assistant to automate sales in your company?
At Cognize, we specialize in designing systems that transfer complex sales processes to the online channel. Our AI assistant implementations for Otolampy.pl , LuxMarket.pl , and Dobregniazdka.pl demonstrate how AI technology can truly support sales teams and save them time.
Do you have an idea, a ready specification or a business need?
Write to us. We will contact you within 24 hours.