Commissioning a chatbot: how to choose the right agency
Direct answer: A good AI chatbot development agency starts with your support tickets and business processes, not with technology. It defines measurable goals (first-contact resolution, response time, qualified leads), builds the bot with access to your real data, tests it multilingually and hands it over to operations with clear KPIs. Updated: July 2026.
The Swiss market is full of chatbot promises: no-code builders, generic GPT wrappers, offshore agencies with standard packages. Meanwhile, customer expectations are real: answers in seconds, around the clock, in German, French, Italian and English. Any company commissioning a chatbot therefore faces two questions: which bot architecture fits the use case, and which agency can build it so it does real work instead of just rerouting enquiries?
This guide answers both from practice: bot types compared, honest cost drivers, an agency selection checklist and the typical project flow from discovery to live operations.
Why Swiss companies are commissioning chatbots now
Three developments have changed the math.
First, language model quality. Modern AI chatbots understand context, typos and mixed languages. They can search knowledge bases, look up order status, book appointments and push structured data into CRM systems. The difference from the rule-based bots of 2020 is fundamental: back then the bot failed on any phrasing outside its script. Today projects almost never fail on language capability - they fail on missing connections to real data and processes.
Second, the labour market. Qualified support staff are scarce and expensive across Switzerland. A bot that fully resolves 40 to 70 percent of recurring enquiries visibly changes workforce planning: the team focuses on complex cases while the bot absorbs volume, night hours and weekends.
Third, customer expectations. Someone with an evening question about a quote, a booking or a product does not wait until Monday. Companies that answer instantly and competently win the deal. That holds for e-commerce as much as for fiduciary services, insurance, real estate or healthcare.
What matters is the right framing. A chatbot is not a marketing gadget - it is a piece of process automation with a conversational interface. Strategically it belongs in the same discussion as business process automation and an overarching AI integration strategy: which enquiries have volume, which follow clear rules, where is the business case?
Chatbot types compared
Not every use case needs the same bot. The three common architectures differ in effort, capability and risk.
| Type | Strengths | Limits | Typical use |
|---|---|---|---|
| Rule-based bot | Predictable, inexpensive, no hallucinations | Fails on free-form phrasing, high maintenance per scenario | Simple FAQ, menu guidance, form replacement |
| AI knowledge bot (RAG) | Answers free-form questions from your documents and knowledge bases, multilingual, verifiable answers | Information only, performs no actions | Customer service deflection, internal knowledge search, product advice |
| Agentic chatbot | Gets things done: books appointments, checks order status, qualifies leads, writes data to the CRM | Highest integration and testing effort, needs clean APIs and guardrails | First-level support, sales, self-service processes |
In practice, most successful Swiss projects combine an AI knowledge bot with targeted agentic capabilities for the two or three most valuable processes.
How to recognize a good chatbot development agency
These seven questions separate the builders from the demo sellers in selection meetings.
Does the agency ask for your support data first?
A serious agency wants to see your ticket categories, enquiry volumes and most frequent questions before quoting. Anyone naming a fixed price without a data review is selling a template, not a business outcome.
Are there hard KPIs from day one?
First-contact resolution, deflection rate, response time, qualified leads per month, escalation rate. A chatbot project without measurable goals can neither be defended nor improved after six months.
How does the bot handle not knowing?
The single biggest quality difference: a good bot says what it does not know and hands over cleanly to a human, including the conversation context. Ask about hallucination guardrails, source citations and escalation logic.
Is multilingualism a core topic or a checkbox?
Decisive for Switzerland: German, French, Italian and English, plus tolerance for Swiss German input. Ask to see real multilingual conversations, not just a translation feature.
Which integrations are included in the price?
The value lives in the connections: CRM, ERP, calendar, shop system, ticketing. Clarify which interfaces the quote covers and what counts as extra effort.
Where do model and data run - and is that FADP-compliant?
Ask about hosting location, data processing agreements, data minimization and how personal data is treated in prompts and logs. For many Swiss industries, Swiss or EU hosting is a hard requirement.
What happens after go-live?
A chatbot is half finished at launch. Monthly conversation reviews, new intents, knowledge maintenance and KPI reporting belong in a clear operating model with defined responsibilities.
What chatbot development costs in Switzerland
A serious price can only be named after a short analysis, because it depends on five drivers:
Scope of use cases. A knowledge bot answering questions from 50 documents is a different project from an agent that books appointments, checks orders and qualifies leads into the CRM. Every fully automated process brings analysis, integration and testing effort.
Integrations. Connecting CRM, ERP, shop or ticketing is usually the largest single item. Clean, documented APIs reduce the effort considerably; legacy systems increase it.
Languages and channels. Website widget, WhatsApp, Teams or telephony, one language or four: every additional channel and language expands testing and maintenance scope.
Data protection requirements. Industries with sensitive personal data (health, finance, fiduciary) need stricter architecture decisions on hosting, logging and access control.
Operating model. One-time handover to your team or continuous optimization by the agency - both are legitimate, but they change the cost structure.
Our recommendation: start small and measurable. A focused first rollout with one clear use case delivers a solid business case within weeks, on which further stages can be justified cleanly. That is exactly how we structure projects for AI chatbots and virtual assistants: a clear business purpose, live data access and hard success KPIs from day one.
How a chatbot project runs at AETHER Digital
Phase 1 - Discovery (1 to 2 weeks). We analyze your support tickets and enquiry channels, quantify volume per category and define which cases the bot should resolve fully and where it hands over to humans. Result: a prioritized use-case catalogue with the expected effect on handling time and cost.
Phase 2 - Design. Conversation design in your brand voice, escalation logic, guardrails against hallucinations, data protection architecture (hosting, logging, processing agreements under the revised FADP).
Phase 3 - Build. Connection to your knowledge sources and systems, multilingual testing with real phrasings from your tickets, pilot operation on a subset of enquiries.
Phase 4 - Deploy and iterate. Go-live with a KPI dashboard, monthly conversation reviews, continuous capability expansion along measured demand.
When the chatbot is part of a larger automation initiative, we connect the project to an AI roadmap so the bot, process automation and data strategy build on each other instead of running in parallel.
Frequently asked questions
What does an AI chatbot development agency actually do?
It translates a business problem (too many recurring enquiries, lost leads outside office hours, an overloaded support team) into a productive chatbot: requirements analysis, conversation design, connection to knowledge sources and systems such as CRM or ERP, multilingual testing, data protection setup, go-live and continuous optimization against KPIs.
How long does it take to have a chatbot developed?
A focused AI knowledge bot connected to existing content is typically productive within a few weeks. Agentic bots with multiple system integrations take longer because interfaces, test cases and escalation logic must be built properly. The key is a pilot with measurable KPIs instead of a months-long project before the first value.
What does an AI customer service chatbot cost in Switzerland?
The price depends on use cases, integrations, languages, channels and data protection requirements. A knowledge bot is significantly cheaper than an agent with CRM and ERP connections. After a short analysis of your enquiry types you receive a fixed-price offer with a clearly defined scope.
Can an AI chatbot understand Swiss German?
Modern language models handle Swiss German input remarkably well and reply in standard German. What matters is that the agency tests with real phrasings from your tickets, including dialect, typos and mixed-language enquiries.
Is a chatbot compatible with the Swiss data protection act (revised FADP)?
Yes, if the architecture is right: a clear legal basis, data minimization in prompts and logs, regulated data processing agreements, transparent user information and, depending on the industry, Swiss or EU hosting. That belongs in the design phase, not in a retroactive audit.
Does the chatbot replace our support team?
No, it changes their work. The bot absorbs volume, standard cases and off-hours; the team focuses on complex, high-value cases. The best metric is not the number of bot conversations but first-contact resolution and the hours freed up per week.
Does AETHER Digital build chatbots for Zurich and the rest of Switzerland?
Yes. We develop chatbots and voice agents for SMEs and enterprises in Zurich, Winterthur, Basel, Zug, Bern, St. Gallen and across Switzerland - multilingual, to Swiss quality standards, with personal support from the canton of Zurich.
Related case studies
Continue reading
How Swiss SMEs turn AI hype into a defensible roadmap: a practical guide to use-case prioritization, cost and ROI, FADP compliance and a 90-day implementation plan for companies in Zurich and across Switzerland.
A practical executive guide to turning search impressions into qualified pipeline: how a Swiss SEO audit should diagnose technical blockers, local SEO visibility, CTR leaks, content gaps and conversion friction in Zurich, Basel and national markets.
Why Swiss founders and CTOs are walking away from off-the-shelf SaaS in favour of bespoke platforms: multi-tenant architecture, revFADP compliance, Swiss-sovereign hosting, and the 2026 stack explained without hype.