The vendor call nobody picked up

Logistics · AI inside

The vendor call nobody picked up

A voice agent for vehicle hiring, and what happened after it worked

  • 15–25% → 70–80%vendor pick-up in the proof of concept
  • ~10xmore vendors reachable
  • Parkedafter Nov 2025, when nobody owned it
The vendor call nobody picked upscreenshots and video coming

The problem

Every sales order needed a vehicle. Vehicle hiring managers collected rates and bids by calling vendors one at a time, on top of a WhatsApp bidding system that already existed. Most of a hiring manager’s day went to the first call, not to the quality of the placement. Vendor engagement on outreach sat at 15 to 25 percent.

What I decided

The bottleneck was engagement, not supply. That reframed the project from “automate calls” to “raise the response pool”, which is a different product.

The process I designed

  1. Contact governance first. Who may be called, from which list, with what script, how many retries, in which language, and who owns the outcome. No vendor demo covered this part.
  2. Outreach strategy and deployment workflow. The agent plugs into the existing WhatsApp bidding flow, makes the first call, collects rate and interest, and hands off to a human only on escalation.
  3. Human ownership by design. Hiring managers stay accountable for placement quality. The agent removes the initial call, not the judgement.
  4. Coordination. Business team, external AI partners, company leadership, the internal IT team. I worked directly with the President’s office on objectives and requirements.

What it changed

In the proof of concept, vendor pick-up rose from 15 to 25 percent to 70 to 80 percent and the number of vendors reachable grew about tenfold. The first call was removed from the hiring manager’s day entirely.

What happened next

It was parked. After November 2025, when I left the company, there was no owner, no visibility and no path to scale. The learning stayed with the people who built it.

I tell this half because it is the most useful thing I know about AI in operations: a proof of concept can hit every number and still die of ownership. It is why every AI roadmap I write now has a named owner and a gate before spend.

Decisions I would defend

  • The hard part was never speech. It was scripts, retry budgets, call windows and who answers for the result.
  • Measure the pool, not the calls. A thousand calls that nobody answers is not outreach.