AI for Swiss trades: quotes, dunning, WhatsApp triage – where ROI is fastest

What Swiss trades are about – and why AI pays back here

The Swiss Trade Association sgv-usam represents 230 branches with around 600,000 SMEs in commerce and services; an estimated 80,000 of them are classic trades: electrical, plumbing-heating-cooling, carpentry, painting, gardening, flooring, roofing, sheet-metal, automotive workshops and construction firms under 50 staff (sgv-usam 2024). Most firms have 3 to 25 employees, the owner works on site and does administration in the evenings.

This constellation is more AI-suitable than almost any other. The work is physical – material, tools, customer on site – not AI-substitutable. Administration, on the other hand, is highly formulaic: requests come by phone, WhatsApp or mail, quotes follow a recurring structure, cost estimates vary around standard parts, dunning follows a 3-stage scheme, service dispatch is calendar plus availability.

The software landscape in Swiss trades: Sage 50 Handwerk, Cresus Faktura, BauApp, Bexio, MEPA. These systems have open interfaces or at least CSV/REST APIs. AI as a layer on top needs no software migration – just workflows in n8n or similar tools, mediating between WhatsApp/email, ERP and dunning.

The key difference to fiduciary or medical: data sensitivity is lower (no highly protected personal data), formal compliance is easier, ROI is more visible. An 8-person trade firm that writes 60 quotes per month, of which 25 are executed, realistically saves 30 to 40 owner hours per month with quote AI. At an effective hourly rate of 120 to 180 francs, the setup amortises in three to six months.

Why it matters now

Three pressure points in 2025/2026 force Swiss trades into efficiency investment.

First: labour cost. Swiss trade hourly rates rose 2.5 to 3.5 percent per year in 2023 and 2024 (sgv-usam wage commentary 2024), with end-customer prices flat to slightly rising. Margins are thinner, every administrative hour that cannot be billed has a measurable cost.

Second: skills shortage. Many firms run over capacity, jobs are declined or delayed. Whoever processes inquiries faster and converts unproductive admin time into productive site time gains revenue without new staff.

Third: payment morale. Outstanding receivables in Swiss trades sit around 8 percent of annual revenue in 2024 (Creditreform Switzerland 2024). Consistent, automated dunning reduces that by 30 to 50 percent according to data. This is not "AI in the strict sense" – it is workflow automation with an AI text module for the dunning letter – but the effect is real.

A counter-argument to know: many owners are sceptical ("I know my customers, I do not need AI"). True for customer contact – the owner remains the face. AI does not replace the on-site advice; it replaces typing Excel at 10pm.

Three workflows with the clearest return

In trades, three applications deliver more than 70 percent of the value. All three are literally office work, not site intervention.

Quote and cost estimate generation. A request comes in by email, phone (transcribed via Whisper) or WhatsApp. The language model extracts key data (what, where, when, scope), matches against the internal material and rate catalogue and produces a quote draft in the office template. The owner reviews, adjusts prices, signs, sends. Time per quote: 8 to 15 minutes instead of 45 to 90.

Dunning automation. Open items are pulled from the ERP (Bexio, Sage, Cresus). A pipeline (see ai-mahnwesen-automation) generates dunning stages 1, 2 and 3 in customer-friendly language plus a fact block (invoice number, amount, due date). Sending via email or hybrid post (PostMail, klara). The owner only steps in for stage-3 cases or special situations.

WhatsApp/mail triage. Many tradespeople receive a large share of requests by WhatsApp. A WhatsApp bot (whatsapp-telegram-bot) reads the message, classifies it (emergency, new customer, existing customer, follow-up, appointment), replies in the firm name with a first acknowledgement and a clarification question, and hands the case structured into the office. For emergencies (burst pipe, power failure) the on-call duty is notified immediately by push. Reduces response time from hours to minutes.

Further reasonable use-cases with less leverage: automatic service-dispatch suggestions from location and calendar, photo-to-parts-list (photos of damaged component plus AI suggestion for spare parts), phone voice bot for appointment booking outside office hours (voice-agent-telefon). These usually come in phase 2 – after a successful quote and dunning pilot.

6 steps to a productive quote and dunning AI setup

1. 01Audit: read the ERP (Bexio/Sage/Cresus), collect templates, prepare 20 typical quotes and 10 typical dunning letters as training material.
2. 02Consolidate the catalogue: materials, hourly rates, standard items in a flat table (CSV or database view) so the model can access it.
3. 03Build the quote generator: prompt template with office tone, material/hours lookup, variant logic (e.g. standard, premium, budget).
4. 04Release workflow: every quote goes to the owner first for review. Sending happens only after a click. In the UI clearly mark "AI draft".
5. 05Wire up dunning: ERP cron pulls overdue invoices, language model drafts dunning text in 3 stages, sending via email (Brevo) or hybrid mail. Only stage 3 requires owner release.
6. 06Measure KPIs: time per quote, acceptance rate, days to payment, outstanding receivables. If no measurable improvement after 8 weeks, rework or stop.

When entry pays off

Entry here is shorter and more concrete than in regulated industries.

The audit (ai-readiness-audit) usually takes 1 to 2 days in trades instead of 3. Captured: ERP (Bexio/Sage/Cresus), mail/WhatsApp inbox, existing quote templates, material and rate list. Output: a pilot recommendation with a clear ROI model.

The pilot almost always focuses on quote generation, because the owner feels the effect immediately. Duration: 4 to 6 weeks, owner plus one office staff, 20 to 40 real quotes. Metric: time per quote and acceptance rate.

AI fits when (a) the firm writes at least 30 quotes per month; (b) jobs are not exclusively bespoke one-offs; (c) the owner does the administration and that is measurably too much. Below 30 quotes/month, setup rarely pays off – a better template and an FAQ bot are enough.

Managed phase (managed-service-monitoring) ensures the workflow keeps running, model updates ship, and the dunning workflow does not break on ERP updates.

Where AI for trades does not make sense

Three cases.

First: highly individual bespoke work. If every quote is a custom design (made-to-measure kitchens, restoration work on heritage buildings, special sheet-metal), the AI suggestion is often too generic. A RAG setup over your own completed projects beats a catalogue-based quote generator.

Second: solo self-employed under 30 quotes per month. Setup time (4 to 8 weeks pilot, 3 to 5 days audit) plus monthly cost (CHF 280 to 450) only amortises with volume admin. Solo carpenters writing 8 quotes per month do not need AI – they need a better template and 30 minutes of Bexio training.

Third: safety-relevant phone advice. A voice bot that says "I will have the technician call you back" on a gas emergency is dangerous. For emergencies (gas, power, flood), the bot must forward immediately or give the 144/112 hint – no own recommendation.

And a fourth pragmatic point: the WhatsApp Business API costs a small fee per conversation (Meta pricing 2026, in the CHF 0.02 to 0.10 range depending on country/mode). At very high volumes a dedicated VoIP/mail route beats a WhatsApp-only strategy.

Trade-offs

FAQ

Related topics

Sources

1. Schweizerischer Gewerbeverband sgv-usam – Branchenstatistik 2024 · 2024-11
2. Creditreform Schweiz – Zahlungsmoral KMU Schweiz 2024 · 2024-12
3. Bexio AG – REST API Dokumentation (Stand 2026-04) · 2026-04
4. Meta – WhatsApp Business API Pricing & Conversation-based Pricing 2026 · 2026-01
5. BFS – KMU-Strukturstatistik Schweiz 2024 · 2024-10