At 8:28 on a Saturday morning, a Gig Harbor homeowner opened a chat on a leak detection company’s website. His water bill had doubled compared to the same time last year. No visible leak, no wet spots in the yard — just a number on a bill that didn’t make sense and a growing suspicion that something underground was quietly costing him money.
Nobody at the company was watching the website that Saturday morning. It didn’t matter. The company’s AI website agent asked where the leak might be, confirmed the service address, explained how meter-to-house detection works, and collected his email and phone number. By the time the team looked up from their weekend, the job was waiting for them.
That conversation is one of 520 we analyzed from a single UpChat deployment: a water leak detection company serving Washington’s Puget Sound region, from February 2025 through July 2026. It’s a very different business from the Florida fence company we profiled earlier — lower volume, higher urgency, and a customer who usually doesn’t know they have a problem until the bill arrives. That difference is exactly what makes the numbers worth looking at.
The headline number: 127 of 128
Every conversation in the export is classified by what the visitor was trying to do — schedule a service, ask a sales question, get help as an existing customer, or just look around.
Here’s how the serious inquiries break down:
| Conversation type | Conversations | Ended with contact details |
|---|---|---|
| Schedule a service | 53 | 53 |
| Customer service | 51 | 50 |
| Sales question | 15 | 15 |
| Information request | 9 | 9 |
| Total serious inquiries | 128 | 127 — 99.2% |
Of the 128 conversations where a visitor had a clear service request or account question, 127 ended with a name and a working email or phone number. One got away.
A fair caveat, because we’d rather undersell than overclaim: classification and capture aren’t fully independent — a chat that ends with a booked appointment naturally gets classified “Schedule.” The cleaner way to say it is this: when a visitor engaged seriously with the agent, the conversation almost never ended without contact details. No form abandonment, no “we’ll call you back” dead ends, no voicemail. For a company whose average job value runs well into the hundreds, one captured after-hours lead can pay for the software for a year.
Half the conversations happened when the office was closed
Across all 520 conversations, 52% started outside Monday-to-Friday business hours — nights, early mornings, and weekends. More than a third of all captured leads (49 of 134) arrived in those windows.
That’s an even higher after-hours share than the 47% we measured at the Florida fence deployment, and the reason is intuitive once you see the conversations. A fence is a project you plan; a leak is a problem you discover. And you discover it at odd hours: the bill lands in the evening inbox, the water meter check happens after dinner, the wet spot in the garage shows up on a Sunday.
The data backs that up:
- “Leak” appears in 149 conversation summaries — the single most common topic.
- 40 conversations mention the water meter, usually some version of “everything’s off and the dial is still spinning.”
- 24 conversations describe a high, doubled, or tripled water bill — the classic trigger. One Everett homeowner opened a chat at 6:50 PM because her bill had tripled; she left her address, email, and phone number, and flagged that the line had failed before.
- Roughly 70 conversations asked about pricing, quotes, or estimates — urgent customers still comparison-shop.
One more detail we didn’t expect: the agent fielded inquiries from other trade professionals, including a plumbing business owner asking about underground detection rates for a customer’s property. When your website answers competently at 7:31 AM, even the pros treat it like a colleague.
A different business, the same pattern
When we compared this deployment with the Florida fence account, the contrast was the point.
The fence company is a volume business: 8,652 conversations, more than 3,000 captured leads, roughly 180 a month. The leak detection company is the opposite: about 30 conversations and eight captured leads in a typical month, with sharp seasonal spikes — July and August 2025 nearly tripled the average, right when summer irrigation season stresses aging pipes.
Different volumes, different products, different states. Same pattern:
- A large share of demand shows up after hours. 47% in Florida, 52% in Washington. If your website goes quiet at 5 PM, roughly half your pipeline talks to a competitor — or a contact form they’ll never finish.
- Serious inquiries convert at very high rates when someone — or something — answers immediately. 89% sales-intent capture in Florida; 99% serious-inquiry capture here.
- Most visitors are still just looking. About 64% of the leak-detection chats were incomplete inquiries: anonymous browsers who asked a question or two and left. That’s normal, and it’s fine — we’ve written before about why tire-kickers are good business. Tonight’s anonymous question about crawl-space moisture is next spring’s detection job.
For a home-services owner, the takeaway isn’t that chat is magic. It’s that the leads were always there — arriving at 8:28 AM on a Saturday with a doubled water bill in hand — and the only question is whether anything on your website is equipped to catch them. That’s the case we lay out in detail on our website chatbot for home services page, with the booking flow to go with it.
What the agent actually had to know
Water leak detection isn’t a business a generic chatbot can fake. Reading the transcripts, the agent routinely had to:
- Distinguish residential from commercial jobs, and meter-to-house service lines from irrigation systems and interior slab leaks.
- Confirm service-area coverage across Puget Sound — Seattle, Bellevue, Renton, Everett, Kirkland, Gig Harbor, even Vashon Island — and gracefully hand off out-of-area requests, like a 10:58 PM inquiry from the Olympic Peninsula about a 430-foot service line, with a referral instead of a dead end.
- Explain non-invasive detection to homeowners worried about fresh sod and finished floors.
- Set honest expectations on pricing, which is quoted case-by-case, and route the details to the owner for review.
That’s trade-specific competence: the agent trains on the company’s website, services, and service areas, then handles the conversation the way a knowledgeable office manager would — except the office manager doesn’t work at 10:58 PM.
Methodology
The numbers in this post come from a single UpChat deployment’s full conversation export: 520 website conversations from February 2025 through July 2026, timestamps in Pacific time. “Captured” means the conversation ended with a visitor-provided email or phone number. “Serious inquiries” are conversations classified as Schedule, Customer Service, Sales, Sales Lead, or Information; the remaining 392 — incomplete inquiries and miscellaneous chats — were mostly anonymous browsers, and we’ve excluded them from the capture-rate claim rather than let them flatter it. After-hours means outside Monday–Friday, 8 AM to 6 PM Pacific. The client asked to remain anonymous; homeowner details have been anonymized.
Someone’s reading a doubled water bill right now
The next lead won’t schedule their emergency around your office hours. Whether you run leak detection in Washington or fencing in Florida, the pattern holds: answer instantly, answer competently, and the serious inquiries rarely get away.
Start your free 14-day trial — your agent trains on your website and goes live in about five minutes.
Put your best employee to work
Deploy an AI agent that replies in zero seconds, captures leads, and books demos — 24/7. Live in under five minutes.
Never miss a playbook
One useful email a month on AI, support & growth. No fluff.




