The 3A Philosophy
Before you build an agent, you need to know what it should do. Every meaningful use case falls into one of three categories, and that choice drives complexity, cost and impact.
Prepare decisions, summarise documents, draft replies. The human stays in the loop and makes the final call. Low barrier to entry, high acceptance, low risk.
Take over rule‑based processes end to end. Invoice validation, data extraction, standard documents, system updates. The agent decides within clear boundaries, the human steps in only for exceptions.
Run continuously in the background, watch workflows, spot risks and escalate intelligently before problems surface. The agent doesn't make your team replaceable, it makes them more far‑sighted.
Services
Six product formats under one roof, each belongs in one of the three 3A categories. You get what your task really calls for, not what the latest trend conference recommends.
Intelligent assistants with real contextual understanding that serve customers around the clock and cut support costs. Integrated into your website, ticketing system or app. Not the FAQ parrots of 2019, assistants that actually understand what customers are after.
Retrieval‑Augmented Generation for precise answers drawn straight from your company data. No hallucinations, no invented facts, only what's verifiably in your documents.
Automation of back‑office processes: data capture, scheduling, reporting, document processing. We don't build script wastelands held together by five different tools, but cleanly orchestrated agent chains that debug and explain themselves.
Seamless connection to existing CRM, ERP and ITSM systems. The agent reads, writes and syncs across your systems so data is no longer copied around but orchestrated. Includes error handling and an audit trail.
Specialised agents for your industry, trained on your processes, your data, your terminology. Instead of a generalist that does everything half well, a specialist that's better in your domain than any off‑the‑shelf industry generalist.
Multiple specialised agents working in orchestrated harmony, handling complex workflows end‑to‑end: one researches, the second checks, the third decides, the fourth executes. The discipline of disciplines, and the biggest lever when used right.
Investment
The three pricing tiers map to the three agent types. One‑off build; ongoing operations optional. All prices are indicative and finalised in conversation.
AI chatbot or assistant with real contextual understanding, integrated into your website or application and connected to your knowledge base. The fastest path into productive AI use with measurable ROI.
Automation of one defined business process with AI agents, API integrations, error handling and monitoring. The sweet spot for SMEs who want a clear, recurring process fully handed over.
A complex system with multiple orchestrated AI agents, RAG hookup, decision logic and end‑to‑end automation. For companies that want to run several processes as a coherent agent team.
All prices are indicative. The exact scope is defined in a personal conversation. The 90‑minute potential analysis (CHF 490) is fully credited on engagement. Ongoing operations and monitoring optional as a monthly retainer.
Book a potential analysis · CHF 490Frequent Questions
The questions that come up before every agent project. Answered briefly and honestly.
A chatbot answers questions. An agent executes tasks. The chatbot tells you how you can cancel an invoice; the agent actually cancels it, including the refund in the accounting system and the customer notification.
Technically: agents have access to "tools", to APIs, databases, email, calendar, documents. They plan work steps, execute them, check the result and adjust their approach. A chatbot is a conversation surface; an agent is a digital colleague.
Well suited: recurring processes with clear rules, plenty of text or document processing, connections between several systems, 24/7 availability requirements. Concretely: invoice checking, ticket triage, quote generation, scheduling, report generation, data sync between CRM and ERP.
Less suited: one‑off strategic decisions, highly emotional customer conversations, processes without clear rules or with fewer than a handful of cases per month. The effort and risk simply don't pay off there.
Yes, if you take the setup seriously. We work with the enterprise endpoints of the major providers, where your data is not used for training. On request we run the system in Swiss data centres and use exclusively European or locally hosted models.
For particularly sensitive areas (health, finance, public sector) agents can also run on your own infrastructure, we clarify the requirements in the potential analysis and build the setup so that compliance requirements are met.
That depends on the task. For nuanced text and code work, typically Claude. For broad knowledge questions and multimodality, often GPT or Gemini. For agents with browser or computer access, Perplexity or Anthropic‑based orchestrations. For small, fast tasks, locally hosted open‑source models.
We build model‑agnostic, you're never tied to a single provider, and the system can be switched to a newer or cheaper model when needed, without rebuilding the logic.
No agent is perfect, no more than any employee. That's why every agent we build has clear escalation paths: if confidence drops below a threshold, or certain combinations appear, a human is brought in automatically before the action is executed.
On top of that, every agent logs its decisions and the reasoning behind them, that makes audits possible and gives you control back, even where the call was autonomous. For critical processes we deliberately start at an "Assist" level and only let the agent act autonomously after a proving phase.
In 90 minutes, we identify three processes in your company that are right for Assist, Automate or Amplify agents, with a clear effort estimate and ROI projection. The fee is fully credited on engagement.