Key highlights
- 40,000+ monthly tickets handled autonomously by an AI customer support agent deployed across 25+ product lines spanning hearing aids, health and wellness, and supplements.
- 87% reduction in support costs, with monthly e-commerce support spend dropping from $50,000 to roughly $6,500.
- $200K/month in recovered revenue from refund requests converted into retained sales across the full product catalog through a multi-layer retention strategy.
- 25% final-layer save rate — one in four customers who had already committed to a refund changed their mind at the final pushback layer (measured during the first 24 hours of testing).
- 4-point drop in company-wide refund rate, verified across a 45-day evaluation window using a three-step independent verification process on every ticket.
- CSAT improved from 80% to 90%, while response times moved from hours to instant.
A high-growth DTC brand running e-commerce customer support at a volume humans couldn't consistently handle.
4AM Media operates a direct-to-consumer portfolio across hearing aids, health and wellness, and supplements. That's more than 25 product lines sold directly to consumers, and each one generates ongoing customer support load for setup, troubleshooting, and returns. Like a lot of scaled DTC brands, their support function grew alongside order volume — and so did the cost and complexity of managing it.
The company had built support around 15+ full-time internal agents plus a sizeable outsourced offshore team, handling over 40,000 tickets a month at a cost north of $50,000. The operation worked. The issue was what it did with that volume. Roughly 80% of tickets were refund-related, and the standard procedure was to process those refunds with almost no pushback. No conversation, no attempt to understand the customer's actual issue, no attempt to save the sale. Support had basically become a refund processing factory, with revenue going out at the same speed it came in.
The founder had already proven the refund recovery methodology worked. Humans couldn't execute it at scale.
The core issue wasn't the volume of refund requests. It was what happened when those requests came in.
The founder had designed and manually tested a layered refund recovery methodology over six months with a small, specialized team. It's a multi-step protocol that engages the customer, identifies the underlying issue, offers resolution alternatives, and only processes the refund after the customer has moved through every layer. The result was a 70% sales save rate. Seven out of ten customers who came in asking for a refund kept the product.
So the methodology worked. The constraint was execution. With 40,000 tickets a month, and each one requiring four or five decision points under time pressure, human consistency topped out around 60 to 70%. The rest of the tickets leaked through as instant refunds because someone skipped a step or misapplied the logic. Training, process documentation, incentive structures: all of it had been tried. At that volume and complexity, human consistency has a ceiling, and the ceiling was well below what the business needed.
Compounding all of this, response times were slow. Customers waited hours, sometimes overnight, for replies, especially outside U.S. business hours. Slow responses increased frustration, and frustration increased refund intent. The outsourced support team was the firm's primary line of defense against all of it, and the math didn't work.
A year ago, we would have burned a village down to get the refund rate down one percent.
An AI customer support agent that executes the proven methodology on every ticket.
CustomAI Studio built and deployed a custom AI customer support agent directly into 4AM Media's existing helpdesk infrastructure. The agent was trained on the company's product knowledge base across 25+ product lines, historical ticket data, and the founder's refund recovery SOPs, then integrated via API with the company's CRM and order management system. Deployment took a few weeks, with no disruption to the existing workflow and no new tools for customers or the team to learn.
The agent handles the full customer support lifecycle. It parses multi-intent messages, classifies intent (cancel, refund, product question, or some combination of those), retrieves real-time order and account data, runs structured troubleshooting, and applies the founder's four-layer retention strategy on every refund request before any conversation gets escalated to a human.
That retention strategy mirrors how a skilled support agent should operate. Layer one is understanding: the agent opens with an empathetic question about the customer's experience. A meaningful share of customers who initiate refund requests don't actually want a refund; they want to be heard, and that step alone resolves a measurable percentage of tickets. Layer two is troubleshooting, where the agent identifies the specific issue (fit, sound quality, setup difficulty) and delivers tailored guidance pulled from patterns across the 25+ product lines. Layer three is alternatives: an upgrade, a partial credit, an extended trial, or a personalized consultation. Layer four is escalation. If the customer still wants a refund after going through every layer, a human agent completes the process with full context already captured.
Throughout, the agent maintains a brand-consistent tone, references the customer's specific device model, purchase history, and prior interactions, and logs every conversation for performance tracking and continuous improvement. The methodology the founder proved works at 70% with a specialized team now runs on every single ticket, at full volume, without the consistency falling apart under load.
The goal is not to interfere with refunds, but to make sure customers aren't returning a product due to something that could have been easily resolved.
From $50K/month cost center to $200K/month revenue engine, verified across a 45-day evaluation window.
Every number below was measured over a 45-day evaluation window using a three-step independent verification process on each ticket. These aren't projections.
The company-wide refund rate dropped four percentage points across the full ticket volume. On the highest-volume, highest-value product line alone, the AI customer support agent generates roughly $50,000 a month in recovered revenue, meaning sales that would have been refunded under the old process. Across all 25+ product lines, the total runs around $200,000 a month. That's revenue that had been walking out the door every month, for years, because humans couldn't consistently execute logic the founder had already proven worked.
On the final pushback layer alone (the step where the agent reframes the refund and creates a genuine moment of pause), 25% of customers who had already committed to a refund changed their mind. That figure is from the first 24 hours of testing, and the founder believes continued optimization can push it to 50 or 60%.
The cost savings were immediate. Monthly support costs dropped from approximately $50,000 to roughly $6,500, an 87% reduction. The team went from 15+ full-time agents to 1 or 2 supervisors handling the escalations that come through the fourth layer. Several of the agents who'd previously been processing refund tickets were redeployed to higher-value work like outbound customer engagement, marketing support, and the more complex cases that genuinely need human judgment.
Response times moved from hours, sometimes overnight, to instant. CSAT scores improved from roughly 80% to 90%. Most customers didn't realize they were talking to an AI, because the system operates within the same helpdesk channels they were already using.
E-commerce customer support went from a cost center to an active contributor to the bottom line. A department that cost $50,000 a month became a function that generates $200,000 a month in recovered revenue, at scale, with better customer satisfaction than the human team ever achieved.