Building an AI assistant on AWS to boost sales with PınarOnline

Customer inquiries had become one of the biggest operational pressures at PınarOnline. The company’s 4-year-old ecommerce operation was expanding quickly, but support relied on three disconnected channels: a staffed call center, email, and a dedicated WhatsApp line. When agents were unavailable outside of business hours, customers, particularly those looking to place large bulk orders, often left without completing their purchases. Even routine issues, like login errors or missing discounts, required manual intervention, limiting the team’s ability to respond at scale.

The challenge extended beyond volume. Because interactions were fragmented, PınarOnline lacked visibility into what customers were asking and where problems were occurring. Common issues, like confusion about pricing or delivery, were repeated without being addressed. The sales team also had no reliable way to identify or follow up with high-value customers interested in large purchases. PınarOnline needed a way to respond at any hour, reduce operational load, and turn customer interactions into actionable insights.

PınarOnline selected LimonCloud because of the company’s early experience delivering generative AI solutions on AWS in Türkiye and its ability to deliver production-grade generative AI solutions at speed without compromising enterprise-grade reliability, security, and compliance. After an initial technical briefing, the two teams brought PoBo into production within 7–8 weeks. Before launching PoBo, PınarOnline conducted extensive testing with approximately 1,000 employees across technical, network, and manufacturing teams, validating responses against real-world scenarios. Given the regulatory requirements associated with food products, ensuring accuracy and compliance in every response was essential.

By combining deep AWS expertise with a strong understanding of PınarOnline’s ecommerce and operational workflows, LimonCloud tailored PoBo to the company’s business model rather than deploying a generic AI assistant. The solution architecture was centered on Amazon Bedrock, a platform for building generative AI applications and agents. Amazon Bedrock Guardrails implemented safeguards customized to the solution requirements and enforced responsible AI policies. Amazon Bedrock also provided access to Anthropic’s Claude Haiku and Sonnet models to balance response speed with more advanced reasoning, helping PoBo handle high volumes of customer queries while delivering accurate, context-aware, near real-time customer support.

Amazon Elastic Container Service (Amazon ECS), a fully managed container orchestration service, hosts microservices that support near real-time communication, logging, and environment separation. Amazon Relational Database Service (Amazon RDS), an easy-to-manage relational database service, stores conversation data, sentiment signals, and error logs for ongoing analysis. Additionally, AWS Lambda, a serverless, event-driven compute service, handles asynchronous workflows, while Amazon Simple Notification Service (Amazon SNS)—a fully managed Pub/Sub service for A2A and A2P messaging—sends email alerts to the sales team when customers express interest in bulk purchases.

The solution includes a product discovery tool that highlights discounted items, a structured lead-capture flow that routes inquiries directly to sales teams, and a reporting interface that shows recurring customer questions in near real time. “PoBo is not a chatbot at this point. It’s also providing some insights, help, and products that potential customers might be interested in,” says Emre Ersoy, AI team lead at LimonCloud. “Without customization, AI assistants aren’t very useful for customers. Each customer has different problems that require tailored solutions.” Beyond the initial development, LimonCloud continues to support PınarOnline as a managed service provider. The team proactively monitors performance, responds to potential issues in real time, and delivers enhancements aligned with PınarOnline’s changing requirements.

PoBo now supports approximately 7,000 unique users per month, generating 18,000–21,000 interactions, with customers asking an average of two to three questions per session. Call center volume has decreased by 50–60 percent, and customer satisfaction has increased by 20–25 percent. In its first month of operation, PoBo helped PınarOnline to obtain 25 bulk purchases by capturing inquiries outside of business hours, interactions that previously might have been missed.

Beyond reducing support load, PoBo has given PınarOnline continuous visibility into customer behavior. “Using AWS services, we’re now able to track our reporting much more closely. We can see how many customers are reaching out each day and better understand their feedback,” says Emre Ünlü, product and business growth group manager at PınarOnline. Insights from customer interactions have led directly to improvements. For example, delivery tracking features were added in response to frequent order status questions, such as “Where is my package?” By consolidating support, sales signals, and feedback into a single system, PınarOnline can respond faster, improve the customer experience, and identify revenue opportunities in near real time.