A few months ago, our team at Perkville—a loyalty and referral platform—was facing a growing challenge: an overwhelming volume of support tickets. Our support staff was stretched thin, and we needed a solution that would help us triage and resolve tickets faster, without increasing costs or complexity.
We naturally looked to AI to help us scale our customer service efforts. We signed up for an AI-powered live chat tool, hoping it would ease the burden. Unfortunately, we were unable to get accurate responses for customers and the inaccuracy led to a negative customer experience because the chat experience created the expectation that it would be as accurate as a real human. When we reached out to our support platform’s (CSS) provider, Salesforce Service Cloud, the cost of adding an AI agent to respond to email tickets was too expensive for our business model
We believed there had to be a cheaper and more reliable solution.
The Middle Way: Bringing AI into Our Workflow
Instead of letting AI run wild in free-form chats or email responses, we took a different approach. Using Middle, we connected OpenAI, Perkville, and our CSS to build a more controlled, intelligent solution.
Here’s what we did:
- Enriched case data with user data: Middle helped us unify disparate data sources, giving us the full picture of each case.
- Used AI to define case reasons: By feeding AI the case subject, body, and contextual data, we enabled it to accurately set the case reason. This alone saved staff time and gave us high-level insights into incoming cases.
- Automated smarter responses: With rich context and automatically set case reasons, we were able to generate smart auto-responses using CSS workflows, functionality that we were already paying for. This significantly reduced the manual workload on our team.
- Tackled spam head-on: Spam had been a persistent issue. Using structured AI to identify spam automatically, we developed intelligent auto-close rules to eliminate noise from our queues.

Results That Speak for Themselves
The best part? This entire system costs us just 60 cents per year in OpenAI usage fees to handle roughly one thousand support cases per month.
By orchestrating AI through Middle, we avoided expensive vendor add-ons and sidestepped the risk of handing over full control to AI. Instead, we applied AI within a structured framework that we could monitor, improve, and trust.