How Smartly is improving customer service and automating technical support with generative AI

In today’s highly competitive marketing environment, digital advertising platform Smartly harnesses the power of AI to scale ads across channels. Smartly uses generative AI agents to automate ticketing, generate customer communications, and draft technically detailed resolution messages—boosting both customer satisfaction and support staff morale. To stay ahead in this fast-paced industry, delivering exceptional customer service and fast, effective technical support is crucial.

As the platform grew, Smartly’s technical support teams wanted to automate operations so they could maintain excellent customer service as support request volume ramped up. In order to grow with their business, they needed a robust generative AI solution that could integrate with existing infrastructure while keeping costs low and protecting data privacy.

“Support agents are a primary face of our brand, and as the Smartly platform grew, they were getting overwhelmed during busy shifts,” says Tatiana Kuumola, a support solutions engineer at Smartly. “Our support agents feel empowered now using AI tools, which has led to a more motivated and confident team.”

Smartly needed AI agents that could collect information from multiple platforms, review and summarize issues, and then create clear and detailed tickets for engineering teams. Once a customer’s issues were resolved, the AI agents would need to write professional customer resolution messages that maintained a consistent, on-brand tone.

To ensure security and customer privacy, Smartly needed the solution to run on its private hosting environment within a Kubernetes-based orchestration platform. Shipping data to a cloud service was not an option. This meant finding an LLM with broad natural language capabilities in a compact footprint that could run locally, integrate seamlessly and use minimal resources.

Before Llama, the support team was burdened with repetitive tasks, such as attaching ticket notes to conversations and duplicating tickets in JIRA, which led to inefficiencies. To make sure customers understood their technical issues, the team created descriptive titles and summaries for each ticket. When formulating resolution messages, the team carefully examined ticket details and crafted responses across various communication platforms that ensured that even complex technical issues were clearly communicated to the customer. Deploying Llama streamlined the process and freed up valuable time for team members to focus on other tasks.

“Llama’s superior text understanding, ease of integration, and compact size made it ideal for securely handling customer data, automating tasks, and maintaining high-quality output,” says Erik Karsten, a DevOps engineer at Smartly.

Applying Llama for ticket automation saved teams about 80% of their time by eliminating duplicative work while delivering consistent, professional customer messages. Llama also helped standardize ticket resolution messages, leading to clearer interactions with customers and reducing response drafting time by about half.

Llama 3 required no training or fine-tuning. The Smartly team used basic prompt engineering and few-shot learning to inject clear instructions with examples of formatting, such as ticket titles, descriptions, subject matter context, and writing style.

Because Llama 3 is open source, Smartly was able to deploy it on its internal infrastructure, maintain full control over its data and protect customer information. The model and prompt template integrated seamlessly with the company’s existing Kubernetes platform so it could be easily accessed by other services and applications, leveraging GPU units to enhance performance. This setup ensured Llama was integrated effectively and also optimized for efficiency in handling specific workloads.

“Initially, our AI model experienced slow performance on standard CPU workers within Kubernetes,” Karsten says. “To enhance speed and efficiency, we transitioned to GPU nodes, which were new to our team and required some development effort.”

As the project moved from prototype to production, the Llama community and knowledge base helped the Smartly team work through challenges, optimize the model’s performance on CPUs and GPUs, and tune infrastructure to improve throughput and scale efficiently.

Looking ahead, Smartly plans to expand its use of Llama for developing new internal tools such as ticket categorizations, product support analytics, and writing suggestions for error messages.

“With Llama automating ticket creation and streamlining resolution messages, our support team can now focus on higher-value tasks,” says Kuumola. “This allows them to address complex technical issues, interact with and update customers faster, and reduce duplicated work, improving overall support quality.”

Learn more about Smartly.