Repetition
How often does the task recur?
Volume
How much time and how many handovers does it require?
Foutgevoeligheid
What errors occur, and what do they cost?
Uitzonderingen
How many cases fall outside the normal process?
Data
Is the required information usable and lawfully available?
Human oversight
Where must an employee review or decide?
Start with the process, not the tool
A tool may offer impressive capabilities, but process selection starts with work that demonstrably causes friction today. Observe the inputs, steps, exceptions, handovers and outcome. Only then decide whether fixed rules, conventional workflow automation or AI is appropriate.
Define the process narrowly enough to measure it clearly. “Automate administration” is too broad. “Check the data in one type of supplier invoice and prepare it for posting” is better defined.
Six criteria for a good first automation
Repetition and volume determine the potential reach. Stability and available data determine how much variation the system must handle. Frequent errors can make automation attractive, but also increase the need for validation.
Exceptions deserve particular attention. A process with many implicit decisions or changing rules may appear technically automatable while remaining unpredictable in practice. Map ordinary cases and exceptions separately.
- Are the inputs recognisable and sufficiently standardised?
- Can a process owner assess the desired outcome?
- Can errors be detected and corrected quickly?
- Can the system safely stop or hand a case over when uncertain?
- Is it clear which data are used and retained?
Sources for this section: NIST · Belgian Data Protection Authority
Fixed rule, AI assistant or agent?
Use fixed rules when the logic is stable and can be described completely. An AI assistant fits when interpretation is needed but a staff member reviews the output and takes the action. An agent has more scope to act, so it requires tighter access controls, logging, boundaries and a fallback.
Choose the least complex approach that solves the problem. Greater autonomy is not an end in itself. As impact and uncertainty increase, oversight should increase proportionately.
Sources for this section: European Union · NIST
Illustrative example: invoice processing
A fictional company wants to process incoming invoices. It assumes that invoices arrive through one channel, suppliers are known and staff currently spend considerable time copying fields.
The first phase covers only data extraction and validation. Unusual suppliers, unreadable documents and missing purchase order references go to a staff member. Posting and payment remain out of scope until data quality, error handling and access rights have been validated.
When should you hold off on automation?
Hold off if the process keeps changing, staff describe the intended way of working differently or essential data are incomplete and unreliable. Automation usually amplifies problems in an ill-defined process.
Start by observing the process and establishing a baseline. Then choose one path with sufficient volume, limited exceptions and a clear owner. Define in advance what success, uncertainty and a stop condition mean.
Frequently asked questions
How quickly can you see results?
That depends on process maturity, integrations and risk. Agree a short validation period first and measure against the baseline, not a general expectation.
Do you need to clean up all data first?
Not necessarily. You do need to know which data are critical for the chosen scope and whether they are accurate, complete and accessible enough.
What is the difference between an assistant and an agent?
An assistant supports an employee with substantive work. An agent can carry out steps itself within agreed limits. That additional scope to act requires stronger controls.
How do you ensure compliance and safety?
Define the purpose, data, access rights, logging, exceptions, retention periods and human oversight from the outset, then review them periodically.
Sources
The sources below support the indicated factual and regulatory passages. The practical decision frameworks are professional recommendations from DSC Solution.
- NIST — Artificial Intelligence Risk Management Framework (AI RMF 1.0)26 January 2023Used for: risk management, governance, and the ongoing assessment and management of AI risks.
- NIST — AI RMF Playbookaccessed 29 August 2026Used for: practical actions for Govern, Map, Measure and Manage.
- European Union — Regulation (EU) 2024/1689 — AI Act13 June 2024Used for: risk-based obligations, documentation, oversight and AI literacy.
- Belgian Data Protection Authority — Information brochure on artificial intelligence systems and the GDPRDecember 2024Used for: Belgian considerations for AI and personal data.
