Start with the problem, not the algorithm

The most important question is not which model to use. It is which decision we want to improve. AI applied to business management only makes sense when it enters a concrete process: classifying documents, predicting risk, suggesting priorities, finding anomalies, answering questions about data or automating repetitive tasks with human validation.

When the conversation starts with technology, the risk is creating an impressive but hard-to-maintain demonstration. When it starts with the process, the company is able to measure impact, define who is responsible and control risks.

Five use cases with practical application

INFOS can position AI as a layer applied to the existing ecosystem. This matters because ERP, HR, document management and BI already have data, permissions and rules. AI is not left adrift. It starts working on business context.

The first cases should be useful enough to generate adoption and controlled enough not to create unnecessary risk.

  • Document classification: identify type, supplier, critical fields and routing.
  • Management assistant: answer questions about order book, stocks, sales, margins or production.
  • Operational forecasting: anticipate stockouts, delays, absenteeism or consumption deviations.
  • Task prioritisation: suggest orders, customers or exceptions that deserve attention.
  • HR support: analyse trends in attendance, performance and development needs.

Internal data is the competitive advantage

Generic tools know how to write, summarise and converse. But they do not know how a company works. They do not know that a family of articles has its own rule, that a customer has a specific SLA or that a given supplier should only be used under particular conditions.

The advantage appears when AI is fed by governed enterprise data: ERP, documents, commercial history, production, stocks, HR and BI. Even so, access does not mean total openness. Each user should only see what they would already be authorised to consult.

Governance: the least flashy and most important part

Enterprise AI needs rules. Which data goes in? Where is it processed? Who sees the result? What is logged? How is an incorrect answer corrected? In which decisions is human oversight mandatory? These questions must be answered before putting AI into production.

With GDPR, the AI Act and internal rules, the company should document use cases, purpose, data involved, risks, mitigation measures and those responsible. This does not block innovation. On the contrary: it builds the confidence needed to scale.

Practical checklist

  • Define the purpose and expected value of the use case.
  • Map the data used and the permissions required.
  • Create logs of requests, responses and actions taken.
  • Identify decisions that require human validation.
  • Prepare a process for correction, feedback and continuous improvement.

How to design an AI pilot

A good pilot should be small, measurable and tied to a real routine. For example: automatically classifying supplier invoices for a limited set of entities; answering management questions about order book and margin; or predicting the risk of delay in manufacturing orders based on history.

The pilot should have a simple success metric. It could be time saved, error reduction, increased coverage, risk anticipation or service improvement. Without a metric, the discussion becomes subjective.

  • Choose a process with volume and clear pain.
  • Use sufficient historical data for testing.
  • Define a pilot group with real users.
  • Compare performance before and after.
  • Create a scalability plan only after proving value.

The role of MULTI Connect, pplPortal, Document Management and Qlik

Each product can receive AI in a different way. In MULTI Connect, AI can support consultation, integration, automation and analysis of operational data. In Document Management, it can classify, extract and suggest routing. In pplPortal, it can help identify trends, support talent development and reduce administrative tasks.

With Qlik Sense and MyBusiness-ITV, AI can bring business users closer to the data. Instead of waiting for a report, the team can explore questions, understand exceptions and reach the root cause faster. The key is to keep the data model governed.

Conclusion: useful AI is AI embedded in the process

AI applied to management should not be a decorative layer. It should be integrated into processes, respect data and permissions, explain results and always leave room for human oversight. Success is not about saying that the company uses AI. It is about reducing friction, anticipating problems and improving decisions.

The safest path is to start small, prove value and scale with governance. In this way, AI stops being a promise and becomes a working tool.