A warranty claim opened with context
Hear how the agent understands context, connects to approved systems and leaves a clear next step.
Ecommerce, refurbished electronics
Approximately 72% of conversations reach a clear resolution
A three-agent deployment covering support, post-purchase surveys and branch-line support. Across current production workloads, approximately 72% of conversations reached a clear outcome, while AI handled 53 to 55 percent of volume for three months straight.
Dring AI workflow demo
A warranty claim opened with context
An outbound follow-up checks the order and battery-health record, then opens the appropriate claim without calling the pending case resolved.
72%of current production conversations reach a clear resolution
17-18Kcalls a month, current run-rate
43,000calls handled across 2.5 months
24/7availability, no queue
Context
A marketplace built on volume
Turkey's largest refurbished electronics marketplace runs a high-transaction storefront: order status, warranty checks on graded devices, and returns, all at scale.
Before Dring, inbound volume regularly outpaced the support team, with wait times growing around promotions and restocks. There was also no consistent way to check in with buyers after delivery, or to staff a dedicated line for in-person branch questions.
The agent
Three agents, three moments
An inbound support agent answers the main line, reads the order record, and resolves warranty and return questions on the call, refurbished-specific cases like a battery-health dispute included.
An outbound agent calls buyers after delivery for a short satisfaction check, and a separate agent staffs a store and branch support line for questions that come up in person.
Hi, I bought a refurbished phone last week, order 48213, and the battery health shows lower than listed.
I have your order here. Refurbished units are graded on battery health at intake, yours is listed at 89%. If it's reading lower than that, I can open a warranty claim right now.
Yes please, it's showing 81%.
Claim opened. A replacement or refund option will be in your account within the hour.
ResolutionClosed on call
Handle time~40s avg
How it was delivered
From call reasons to a live line
- Discover
Map the call reasons
The highest-volume reasons for calling, order status, warranty, returns, were mapped from the existing support line before any prompt was written.
- Build
Three agents, one order system
Inbound support, outbound survey and branch-line agents were built against the marketplace's order and CRM records.
- Test
Refurbished-specific edge cases
Simulated conversations covered grading disputes and mismatched battery-health readings before go-live.
- Launch
Gradual rollout
The agent took a share of inbound volume first, then moved to the full line once resolution held steady.
- Evolve
Tuning after live transfer
Once live transfer went live, average handle time fell to about 40 seconds. The team then tuned transfer rules so the agent stopped handing off calls it could resolve itself.
Results
What the logs show
| Metric | Result | Note |
|---|---|---|
| Production conversation resolution | ~72% | Clear outcome on the call, with contextual human handover when needed |
| Operating cost advantage | ~81% lower | Measured against comparable human-only coverage for the same workflow |
| AI share of inbound volume | 53-55% | Stable across three months, with human support available for exceptions |
| Monthly run-rate | 17-18K calls | Current steady state |
| Volume processed | ~43,000 calls | Across 2.5 months |
| Availability | 24/7, no queue | No wait regardless of call time |
| Average handle time | ~40 seconds | After live-transfer rollout |
What we learned
Steady, not spiky
The resolution rate held around 72% across the measured production period, with a clear path to a human when the request needed judgement or extra care.
On this selected deployment, operating cost was up to 81% lower than the agreed human-only baseline, while human judgement stayed available for exceptions.
The 53-55% AI share held for three straight months, not a launch-week peak that faded. That stability was the actual signal to watch, more than any single week's number.
Live transfer cut handle time to about 40 seconds, but it also revealed a new problem: the agent was transferring calls it could have closed on its own. Over-transferring turned out to be its own quality issue, worth tuning for separately from resolution rate.
The inbound line gets the attention, but the survey and branch-line agents cover moments the inbound line never sees: satisfaction after delivery, and questions that come up in person rather than by phone.
Move your inbound volume to AI
See what a phased rollout looks like for your own support line.