AI & Automation
Automation that works in production
I have been automating business processes in industry for years, most recently an order pipeline with over 4,000 records per day. AI is one tool among several, not an end in itself.
Automation first, AI where it helps
Many workflows can be solved more reliably and cheaply with classic automation than with AI: imports, validation, reconciliation, reports. I use AI where unstructured content is involved, such as free-text enquiries or documents.
AI agents
An agent answers questions based on your own content, pre-qualifies enquiries or prepares cases. It works within fixed boundaries: defined data sources, clear handover to humans and no decisions with legal effect.
Integration
Agents and workflows connect to existing systems via APIs, for example CRM, ERP, ticketing or your website. Failures are logged and reported like with any other interface.
Data handling & privacy
Which data is sent to an AI provider is defined and documented per project. Options include providers with EU data processing, data minimisation before sending, and solutions without external AI. Data protection requirements are assessed per project.
Costs & operations
Ongoing AI costs are usage-based and run through your own provider account, transparently and with limits. Operations include monitoring, rate limits against abuse and regular review of answer quality.
Reference: Dlo
Dlo was this website's assistant: an agent with a fixed system prompt and rate limits against abuse. It is currently switched off on purpose, because enquiries here are better handled through the contact form. That is a lesson too: not every process needs AI.