AI Voice Agent for after-hours and overflow calls
Contractors captured ~30% more jobs and revenue, and Breezy became an AI-first company.
TL;DR Breezy used to sell field-service software that was slow to adopt. Our contractor customers' real pain was missed calls. I led an AI agent that answers, books and transfers their calls. Its pilot results convinced Breezy's founders to make AI agents the whole business.
Contractors lose jobs when nobody answers. Switching their whole software to ours was a big, slow decision.
An agent that plugs into the software they already use.
Keep asking them to switch software
Breezy ARR in the quarter after it refocused on AI agents
Thanks for calling Summit Heating & Air. How can I help tonight?
Our AC stopped cooling. It's 84 degrees inside.
I can get a technician out Tuesday 8 to 12 or 12 to 4. Which works?
Tuesday morning.
- Job
- AC repair
- Slot
- Tue 8–12
- Captured
- Name, address, email
- Asked every required field
- Confirmed name spelling
- No unneeded transfer
Impact
Shipped after launch, 2026
- FebLaunch with pilot customers~30% more jobs
- MayInstant Booking by job type~30% booked in-call
- JulEnglish + Spanish, voice catalogonly tested voices offered
- SepTransfer rules fixed with evalseval pass rate 43% to 100%
- OctBlind voice auditionswon 100% head-to-heads
Ask "for which jobs does this business trust automation?" Customers wanted repairs booked automatically, but a person on big installs.
Estimate Follow-up agent
54% of homeowners replied to quotes they had gone quiet on.
TL;DR Contractors send quotes and rarely follow up, so most die from silence, not a "no." I interviewed design partners, prototyped, and launched an SMS agent that follows up for them and hands interested homeowners to their sales team.
Design partners followed up by hand or not at all. Existing tools are built for large companies.
The AI writes each text within the owner's rules, and never closes or offers discounts.
Fixed templates · AI that closes the sale
of texted homeowners had a real conversation
First pilot, first two weeks
Owners don't want another tool to manage. They want a call only when someone is ready to buy.
Gamification for kids in a financial app
Kids' app usage grew 60% after I made the case with experiments.
TL;DR About 90% of Modak's product capacity went to parents. I believed the kids, who use the app every day, were underserved. I tested that without engineering first, then led the full build with engineering, design, marketing, compliance and the CEO.
Kids had few reasons to come back between purchases, and few found the savings tools.
Prove it with no-engineering tests before asking for capacity.
Argue for roadmap time on opinion
app usage after launch
Saving $5 a week adds up to $260 in a year.
Impact
Bring engineers in before the solution exists. They spot risks and simpler paths I would miss.
Simpler checkout for parents sending money to their kids
Removing one step lifted activation 10 points.
TL;DR A parent funding their kid's account was Modak's key activation moment. Support chats showed many parents funding their own wallet and thinking they were done. I A/B tested a direct transfer to the kid, with retention as the guardrail.
The flow followed the system's logic, not the parent's goal: "send money to my kid."
Make it feel like a direct transfer from the bank to the kid.
Keep the wallet step first
parent activation, retention unchanged
Parents saw "✓ added" and stopped.
One action that matches the parent's goal.
Impact
Activation was the real constraint. If the first funding step confuses a parent, we lose their trust before any habit forms.
How I work
Each habit links to a project that shows it.
I'm customer-obsessed
I combine quantitative and qualitative data. Funnels show me where something breaks. Calls, texts and customer interviews show me why. That's how I learned contractors trust automation per job type, not per company.
I'd rather test than debate
When opinions clash, I run something small and bring back evidence. At Modak, a few no-engineering tests did more for the kids' roadmap than any slide.
I make the tradeoff visible
I show the option we didn't pick and why. A clickable prototype usually settles a discussion faster than a long doc.
I jump in, and flag early
As the only PM at Breezy, I did whatever the product needed, including frontend with Claude Code. I bring engineers in early and raise blockers before they become surprises.
AI in my workflow