AI in industry and organisations
Where artificial intelligence and data analysis genuinely improve production, operations and organisational decisions, and where they do not yet.
The question this programme answers
A plant manager, an operations director or an IT lead in a mid-sized Bulgarian company is asked a version of the same question every quarter: should we be doing something with AI? The material available to answer it is written almost entirely by companies that sell implementations. That material is not worthless — it is often technically accurate — but it is selective by construction. It rarely describes the conditions under which the project fails, because those conditions are the reader's problem and not the vendor's.
This programme exists to supply the missing half. We publish the preconditions, the failure modes, the realistic effort, and the cases where the correct decision is to fix something simpler first.
What we cover
Production and operations
Predictive maintenance, quality inspection, process optimisation, demand and production planning, scheduling. For each, the useful question is not whether a model can be built — it almost always can — but whether the surrounding system can act on its output fast enough and often enough to repay the effort.
Data foundations
Most pilots that fail, fail here rather than in modelling. Sensor coverage, sampling rate, timestamp integrity, unit consistency, historian retention, and whether anyone has ever looked at the data end to end. We treat data readiness as the first deliverable of any AI programme, not a prerequisite to be waved through.
Adoption, governance and procurement
Build versus buy. How to read a vendor claim critically. What a proof of concept should be required to demonstrate before it is allowed to become a purchase. How to structure a pilot so that a negative result is a useful, cheap outcome rather than a political failure.
Responsible AI and the EU AI Act
Obligations under Regulation (EU) 2024/1689 are phasing in, and most industrial organisations have not yet worked out which of them apply, or in which role. We publish plain-language mapping of obligations to situations, and the documentation worth preparing before anyone asks for it.
Our standing position
- If you cannot name the person who will act on a model's output, the use case is not ready, however good the model is.
- A pilot that cannot produce a clear negative result is not an experiment; it is a procurement process with extra steps.
- Data readiness is discovered, not assumed. Budget for finding out.
- Accuracy is not the deliverable. A changed decision is the deliverable.
Who this is for
- Operations and plant managers weighing whether a use case is worth investigating.
- IT and data leads evaluating a vendor proposal or scoping a pilot.
- Compliance and legal staff working out which EU AI Act obligations reach their organisation.
- Consultants and advisors who want reference material they did not write themselves.