All use cases
Phones & Booking

A Milwaukee auto recycler trained her AI phone agent like a new hire, and her answered calls went from 61% to over 90%

Natasha Broxton's voice AI, "Lisa," takes 2,000 to 3,000 calls a month at Select Auto Parts & Sales. She says it was horrible at first, and it still makes things up, like telling callers the yard doesn't buy batteries. She listens to the calls and fixes it.

Real storyDone with Voice AI phone agent (product not named), plus a custom vehicle-intake app
TradesAll trades

What they did

Natasha Broxton, CEO of Select Auto Parts & Sales, a 125,000-square-foot indoor auto recycling yard in Milwaukee, wrote a first-person piece for Auto Recycling World (November 9, 2026) about three years of using AI in the business. The problem, in her words: "We missed calls, customers waited too long for answers, and too many decisions depended on me." Her first move was a voice AI agent: "I trained her to be an automotive recycling specialist, not a generic phone bot. Her name is Lisa. She asks the right questions, captures what we need, takes calls when we are busy and after hours, and gives my sales representative complete information." Later she built a custom app for vehicle intake: "A vehicle comes in, my inventory specialist enters the information, my office manager completes the compliance portion, a social media post is prepared, and a dismantle list is created instantly instead of waiting on me." Next she wants Lisa connected to Checkmate, the yard's inventory system, so she can look up a part while the customer is still on the phone, and if they don't have it, take down the request so the team can get the part from another recycler.

How it went

Her numbers: "We went from answering 61% of incoming calls to more than 90%, handling 2,000 to 3,000 calls a month without adding headcount." She is blunt about the start: "Lisa was horrible in the beginning. I listened to the calls and trained her like a new team member." And it still goes wrong: "Recently, she started telling customers we do not buy batteries. We do. I never told her either way, so she filled the gap herself." Her fix was to catch it and correct it: "It works for me. I do not work for it." The intake app "took two to three weeks to build, and plenty of things didn't work." She doesn't name the phone product or say what it costs. The figures are hers, from her own article.

Worth knowing

The battery story is the most useful line here. A phone AI won't leave a blank where your answer should be. If you never told it something, it will make up an answer and sound sure about it. So before you go live, write down the questions callers actually ask, including the yes/no ones you think are obvious (Do you buy X? Do you pull parts? Do you deliver?). Then do what she did: listen to real calls every week and treat each wrong answer as a training note, not a reason to quit. Her advice for anyone who doesn't know where to start: "Start where the money is leaking. Pick one problem." For her that was missed calls, not a big custom build.

Try it yourself

You are helping the owner of a small Wisconsin auto parts, salvage, body or repair shop train an AI phone agent the way you'd train a new counter hire. You find gaps and wrong answers, and you draft fixes. The owner approves every change before it goes into the agent. TASK: (1) Build a "what callers ask" answer sheet so the agent never has to guess. (2) Review a batch of call transcripts and list every answer that was wrong, made up, or should have gone to a person. CONTEXT (fill in the [BRACKETS]): - [SHOP]: Example: Rock River Auto Salvage (fictional), Watertown, WI, 6 employees - [WHAT WE DO]: Example: used parts, we pull or you-pull, buy junk cars, buy scrap batteries and catalytic converters, local delivery Tue/Thu - [WHAT WE DON'T DO]: Example: no installs, no shipping outside Wisconsin, no titles over the phone - [HOURS AND HANDOFF]: Example: counter 8-5 Mon-Fri, 8-12 Sat; anything about price or a specific part goes to the sales rep with a callback number - [TRANSCRIPTS]: paste 10-20 recent call transcripts or call summaries from the agent STEPS: 1. From [WHAT WE DO] and [WHAT WE DON'T DO], write 25 yes/no questions callers are likely to ask (for example "Do you buy batteries?") with the exact short answer. Mark any you can't answer from my info as NEEDED - ASK OWNER. 2. Read every transcript in [TRANSCRIPTS]. For each call, note: what the caller wanted, what the agent said, and whether it was right. 3. Flag every answer the agent gave that isn't backed by my info as MADE UP, even if it sounds reasonable. 4. Flag every call that should have gone to a person per [HOURS AND HANDOFF] but didn't. 5. For each flag, draft the one-line rule or answer to add to the agent. RULES: - Never invent prices, part availability, or policies. If it's not in my info, it's NEEDED. - Quote the agent's exact words for every flag. - Keep fixes short: one rule or answer per line. OUTPUT FORMAT: Part A: 25-question answer sheet (question | answer | source or NEEDED). Part B: call review table (call # | caller wanted | agent said | right? | flag). Part C: draft fixes for my approval. Part D: the 3 gaps that would cost me the most sales. FOLLOW-UPS TO TRY NEXT: - "Write the script for when we don't have the part: take name, number, vehicle and part, and promise only a callback." - "Make a 10-minute Friday call-review checklist I can do every week." - "Turn Part C into the exact wording to paste into my AI agent's instructions." FILLED-IN EXAMPLE (fictional Wisconsin business): - [SHOP] = Rock River Auto Salvage (fictional), Watertown, WI - [WHAT WE DO] = used parts, buy junk cars, buy scrap batteries - [WHAT WE DON'T DO] = no installs, no shipping - [HOURS AND HANDOFF] = 8-5 weekdays; prices go to the sales rep - [TRANSCRIPTS] = last week's 15 after-hours call summaries

Want help getting this working? Rise Systems in Oconomowoc sets these up.

Want this running in your business without the DIY? Rise Systems in Oconomowoc sets these up.

Free 15-minute AI check-up
Read the original

Source: Auto Recycling World · November 9, 2026

Related