How an e-commerce retailer replaced a tired IVR and cut a 6-person call team to 1
Phase 2 of an e-commerce engagement. The old IVR was losing callers in menus. We replaced it with a voice agent that searches a 4,000-page product catalog live during the call and texts the buyer a link to the right product. A team of six dropped to one. Roughly $225,000 in annual savings. Still running in production 18 months later.
- Annual savings
- ~$225,000
- Call-handling team
- 6 → 1
- Successful transfers (5 mo)
- ~13,000
- Time in production
- 18 months
Phase 2 of an engagement with the same retailer covered in the FAQ voice agent case study. Different lessons, different scope. Most teams stop after Phase 1, so we tell this one separately.
The headline
Six people to one. Around $225,000 a year out of the cost base. 18 months in production with no rebuild. That's the version of this case the buyer remembers.
The IVR they wanted gone
The old menu system was tired. Callers got stuck in branches, hung up, and never came back. Six people across multiple departments were holding the line together. Worse, the team had started wanting the phone to sell things, not just route them. The IVR couldn't do that and never would.
So we replaced it.
What the agent does now
The agent handles HQ calls: orders, customer care, the general inbound bucket. One person on the team sits behind it for the cases that need a human. Two changes did most of the heavy lifting:
- Store-level routing. If a caller wants something from a specific store, or just gives a postcode, the agent routes to the right location's sales team. The first version routed a few postcodes near a regional border to the wrong store; we caught it inside a day.
- Live product search. For product enquiries, the agent searches the ~4,000-page product catalog in real time and texts the caller a direct link to the product they asked about. Most buyers tap the link before the call disconnects.
Extensions are now reachable by natural language. No "press 4 for sales", no menu tree to memorize. Disambiguation runs on the first turn of the call.
The Phase 1 knowledge-base sheet is still in place. The client edits FAQ answers themselves in three clicks. We aren't a bottleneck on that anymore.
Five months in, what we saw
Roughly 13,000 successful transfers handled in the first 5-month tracking window. Call-handling headcount dropped from six to one. Annual savings landed near $225,000 once you factor in the headcount, IVR licensing, and the calls that used to die in the menu.
Eighteen months later it's still running. Same routing logic across customer care, parts, delivery, and store-specific sales.
Why this phase is the one that matters
Phase 1 was the "we built a cool voice agent" build. Phase 2 is the one where the cost base actually moves. Headcount. Vendor licensing. Lost-call revenue. The work shows up on the P&L.
If you've already got a voice agent in production and you can feel that the next phase is sitting there, the Voice AI Roadmap is built to map where the next $200K is hiding.