AI Sales Qualification for Windows & Doors
A German window and door dealer was burning sales-rep hours on unqualified WhatsApp leads. We built an AI qualification agent that handles initial conversations, gathers budget, timeline and region, and books on-site consultations directly into the rep's calendar. Roughly seventy percent of inbound leads are now handled without human involvement until the appointment itself.
Background
The dealer sells and installs windows and doors for private homeowners and small construction projects in three federal states. Most inbound enquiries arrive through their website's WhatsApp button, with a long tail from Google Ads, classifieds and referrals. The sales team — three field reps and the owner — used to spend the first ten minutes of each lead manually qualifying budget range, regional fit, project type (new build vs. retrofit), planned timeline and whether the customer already had a competitor quote. A large share of these conversations stalled before any qualification was reached: the lead asked one question and ghosted, the rep replied four hours later, the moment was gone. Out of every ten WhatsApp enquiries, only two or three turned into actual on-site appointments — but every one of them ate into the reps' selling time. The owner wanted to reclaim that time without losing the personal feel the WhatsApp channel gave them.
Solution
We built a conversational agent on the dealer's existing WhatsApp Business number. The agent picks up new conversations within seconds, introduces itself plainly as an AI assistant of the dealer, and walks the prospect through a short qualification: property type, number of windows or doors, desired material (PVC, aluminium, wood-aluminium), rough budget range, timeline, postal code. If the postal code falls outside their service region, the agent says so politely and ends the conversation. If the prospect qualifies, it offers two or three calendar slots in the corresponding rep's schedule. The agent is built on Claude, conversational state is held in Langflow flows, calendar and CRM writes go through n8n, and lead data sits in PostgreSQL on Hetzner Nuremberg. Every conversation is logged and a one-paragraph summary is written to the rep's inbox the moment the appointment is booked, so the rep walks into the consultation already briefed.
Outcome
Roughly seventy percent of inbound WhatsApp conversations are now handled end-to-end by the agent up to the point of a booked appointment or a polite decline. The conversion rate from inbound enquiry to on-site consultation has improved noticeably, mostly because the agent's instant response prevents the "ghosting" pattern that used to kill warm leads. Each rep gets back an estimated five to seven hours per week previously spent on first-contact messaging, which they now spend on actual on-site visits and quote follow-ups. The owner's stated rule — "never make the customer feel like they are talking to a script" — has held; the personal tone was tuned during the first three weeks and now matches the dealer's brand voice.
Lessons Learned
Two points worth recording. First, transparency about being an AI assistant did not hurt conversion the way the owner initially feared — in fact, customers seemed more patient when they knew. Our first prototype tried to be ambiguous, and the rare moments when the model produced a slightly off-tone reply read like an evasive human; once we made the AI status explicit in the opening message, those same moments read as a clear, predictable bot and lost their awkwardness. Second, the booking flow needed more guardrails than we expected. Early on, the agent occasionally offered slots that conflicted with travel time between two appointments because we had not modelled the regional driving distances. We added a small constraint layer that filters slot suggestions by postal-code proximity to the rep's previous and next appointment. That fix removed a category of complaints overnight and is now a default in our calendar-integration pattern.
A similar problem?
Drop us a short note and we will reply within one business day.
Request a project