From 52-minute first reply to 14 seconds — used-car dealership · Kyiv
Inbound from three channels — answered by humans on a desk between 09:00 and 19:00.
Auto Pivdenny ran six yards in greater Kyiv, with inbound from OLX, Instagram and the website all aggregating to a Telegram group three sales managers shared. Replies during business hours were merely slow (median 17 minutes). Outside business hours they were lost — silently — until somebody noticed a thread on Monday morning.
The owner, Pavlo, didn't want a chatbot on the website. He'd seen them. He wanted, in his words, "a worker who reads OLX at 11pm and books a test drive for tomorrow morning, who knows our 220 cars, and who writes the customer's number in the CRM."
We agreed on the contract number in the first call: median first reply ≤ 60 seconds, measured across all three channels, all hours. If we missed it, the pilot was free.
An inventory-aware lead hunter on every channel.
We connected the four channels to one agent core. The agent reads the live inventory feed (220 cars, updated nightly), greets the customer in their language, qualifies against eight rules co-written with Pavlo on Day 1, and books a test drive into the dealership's calendar — checking against staff capacity at the right yard.
The eight rules are unsexy and very specific to this dealer: budget & financing, trade-in or not, manual or automatic, mileage tolerance, body type, must-have features (panoramic roof was a real one), preferred yard, and crucially — whether the customer is "researching" or "ready Saturday".
Hot leads — defined as: ready in ≤14 days, budget within ±10% of a real car on the lot — go to the sales TG with a one-line summary and the booked slot. Cold leads stay in the CRM with a follow-up scheduled for day 7.
| Day | Stage | Output | Owner |
|---|---|---|---|
| Day 0 | Brief | Workflow scoped · 8 qualification rules · contract KPI co-signed | Vlad K. · Pavlo |
| Day 1–2 | Channel wiring | OLX · Instagram · website · sales TG connected; agent identity provisioned | Marko Y. |
| Day 3 | Inventory ingest | 220 cars indexed; nightly sync from KeyCRM established | Olha B. |
| Day 4 | Tone calibration | Per-language voice profile written with the salesfloor | Iryna S. |
| Day 5–9 | Shadow mode | Agent drafts; humans send. 312 conversations evaluated. | All |
| Day 10 | Calibration review | Tone tightened; 3 escalation rules added; refusal contract signed | Iryna S. · Pavlo |
| Day 11–13 | Live · ramp | 10% → 50% → 100% of inbound. Daily metrics call at 19:00. | Marko Y. |
| Day 14 | Hand-off | Contract KPI met. Move to ongoing service. Owner dashboard delivered. | Kateryna L. |
The contract called for ≤ 60 seconds. We hit 14.
Across all four channels, all hours, all languages — measured by the agent's own audit log and verified against KeyCRM timestamps. The largest single category of wins was the night shift, which used to be a graveyard.
- +38% test drives booked in Q1 vs. the same quarter pre-pilot
- 71% of leads auto-qualified end-to-end without a human
- +44% CRM completeness — fields populated per record
- 0 added headcount. The salesfloor stayed at 38.
- Two managers reassigned from inbound to outbound proactive follow-up
What didn't work the first time. The agent's first month over-qualified — it was rejecting some "researching" leads who would have converted in 30 days. We added a soft-warm bucket that goes into the CRM with a 14-day check-in instead of a hot handoff. Conversion on that bucket runs around 18%.
The contract said the worker would reply in under a minute, day or night. The first night, my night manager messaged me at 02:14 in a panic — he thought one of the salesfloor was awake on a Saturday. It was the agent. We sold the car a week later.
Your inbound looks like Pavlo's? Let's draw your version.
A 30-minute call, your eight qualification rules, the number that goes in your contract.