Where should a business start when implementing artificial intelligence?

Efficiency question :

To address the question “Where should a business start when implementing artificial intelligence”, start by defining the desired outcome and understanding the current process. The best solution is not necessarily the one that uses the most technology; it is the one that reduces manual handling, clarifies responsibilities and produces a reliable result with the least possible friction.

An effective approach is to:

• Define the desired business outcome before choosing a tool.
• Start with two or three frequent, measurable and low-risk use cases.
• Classify which data may or may not be used and establish clear rules.
• Test use cases with a small group, measure the gains and adjust practices.
• Roll out gradually with training, examples and appropriate governance.

Key issues:

The main issue is achieving real gains without creating new risks related to confidentiality, response quality or inconsistent employee use. A useful implementation must connect AI to concrete tasks and rules of use that everyone understands.

Challenges:

The main challenges are the quality of data provided to the tool, protection of confidential information, validation of responses and adoption of good practices. It is also important to avoid multiplying tools before priority use cases have been defined.

Technology:

Depending on the context, tools to consider may include Microsoft Copilot, ChatGPT, AI agents and, where appropriate, Power Automate or connectors to internal data. The choice depends on factors such as data volume, number of users, access rights, update frequency, existing systems and the desired level of automation. Technology should be selected after the process and expected results have been clarified, not the other way around.

Developer's advice:

Start with a frequent, verifiable use case. If you cannot explain how you will measure the gain or validate the result, the use case is probably not yet defined well enough.