AI automation: a lot of noise and few concrete answers
Over the past two years, AI automation has gone from being a topic for tech conferences to showing up in every business conversation. The problem is that most of those conversations mix real use cases with overblown promises, and companies that try to act on that information have no idea where to start.
This article is not about what AI can do in theory. It is about what makes sense to implement in a company of 20 to 200 people over the next 6 to 12 months, and in what order.
Before you automate, you need your data in order
AI automation depends on data. Not on huge volumes, but on clean, consistent and accessible data. A company that is not clear on where its data comes from, what each of its indicators means, or how they get updated is not ready to automate processes with AI. It is ready to automate chaos.
The right order is always: first define the metrics that matter, then centralize and clean the data, then automate the reporting processes, and only then consider predictive models or AI-based automation. Skipping steps is expensive.
The three use cases where automation delivers real value in SMEs
Reporting automation. This is the first step and the most impactful one for most companies. Getting rid of the manual spreadsheets that someone updates every week, automating the consolidation of data from different sources, and having reports ready with no human intervention. It is not glamorous, but it frees up hours of work from skilled people who should be doing other things, and it removes manual consolidation errors.
Anomaly detection. With centralized data and well-defined metrics, you can set up automatic alerts when something falls outside normal ranges. You do not need a sophisticated machine learning model for this. Simple rules on the data itself, with thresholds defined by the team, are enough to catch problems before they escalate. A drop in average ticket, a rise in service time, a fall in conversion rate: any of these can be detected automatically if the data is in the right place.
Basic demand forecasting. For companies with at least 12 months of data history, demand forecasting models are technically accessible and have a direct impact on inventory management and resource planning. They do not require a team of data analysts. They require clean historical data and someone who knows how to build the right model for the specific business.
AI automation does not replace human judgment. It amplifies it. But only if the data it runs on is reliable.
What does not make sense to do yet for most SMEs
Large language models (LLMs) like GPT or Claude have interesting applications in companies, mainly in content generation, customer support and analysis of unstructured text. But putting them into critical business processes without a solid layer of structured data underneath just adds unnecessary complexity.
If you are not clear on your real margin by product line, an AI chatbot answering questions about the catalog is not going to solve that problem. It is going to add another layer of technology on top of a problem that still is not solved.
The order matters more than the technology you pick
The question that should guide any automation initiative is not "what technology do we use?". It is "what decision do we want to be able to make that we cannot make now, or what process do we want to run on its own?". Starting from that question, the technology picks itself. And usually, the answer is simpler than it looks.