Prioritising AI use cases: the EVORI scoring framework

This framework answers one question: which process should come first? It scores each candidate on four dimensions, from 1 to 5 points each. The precision of the number matters less than using the same reference points for every candidate, ideally with two people scoring independently.

The four dimensions

Value (1–5)

How much recurring effort or loss of quality does the process involve today?

• 1 — Occasional, a few minutes, one person.
• 3 — Noticeable every week, several people involved, occasional errors that create follow-up work.
• 5 — Daily, across teams, a bottleneck with visible consequences for customers or closing.

Feasibility (1–5)

How easily can a workflow access the data and systems?

• 1 — Paper, isolated systems, no interfaces.
• 3 — Data is available digitally, access can be resolved, systems are partly open.
• 5 — Structured data, accessible interfaces, a recurring pattern.

Data sensitivity (1–5, inverted)

A higher score means the data is less sensitive.

• 1 — Particularly sensitive data (health data, employee data in disputes); sharing externally is sensitive. A GDPR review is essential before any test.
• 3 — Personal data is involved, but it is ordinary business data; data processing arrangements can be clarified.
• 5 — Internal factual and product data without personal information.

Error tolerance (1–5)

What happens if a result is wrong, and will anyone notice?

• 1 — Errors have an external effect or create a legal commitment before anyone sees them.
• 3 — Errors can be corrected internally and are spotted during normal review.
• 5 — Results are suggestions that a person will review and use anyway.

Interpreting the scores

Add up the points (4–20) and rank the candidates, subject to two veto rules:

1. No candidate with data sensitivity = 1 starts until the data protection question is resolved, regardless of the total score.
2. No candidate with error tolerance = 1 starts without a defined checkpoint. See Human-in-the-loop in practice.

The total helps you compare candidates; it does not authorise a project to start. High feasibility does not make up for a lack of economic value. For example, a value score of 1 plus three scores of 5 gives 16 points, but saving a few minutes still does not justify custom development.

An additional requirement before implementation: weigh the expected reduction in workload and improvement in quality against one-off and ongoing costs, as well as internal effort. Human review also counts as effort. If the numbers do not add up or key assumptions remain unresolved, the decision is to hold off or investigate further.

Candidates with high value but unresolved data access can be earmarked for a later review. A score alone is not a basis for commissioning them.

The template

The assessment fits into a simple table:

Scroll sideways to see the full table.

(Example scores to illustrate the method, not project results.)

You can download the template as a text file and complete it with your team: Download the scoring framework. Use the time calculator to estimate the effort a candidate process currently requires.

The limits of this framework

The framework does not replace a business case or legal review. It creates a sound order of priority and helps avoid the two most costly mistakes: starting with the most sensitive process or the one with the least accessible data and systems.

If you would like support with the assessment, that is exactly what the AI Opportunity Check covers.