# Prioritising AI Use Cases: Methods and Matrix

> Agree on a small set of criteria (value, feasibility, risk, strategic fit), fix their weights before anyone scores, then rate every candidate on the same scale. Treat the ranking as a proposal: the real decision is a portfolio of one or two quick starts plus one larger bet you prepare for.

Canonical URL: https://www.ruebstudio.de/en/guides/evaluating-ai-use-cases/prioritize-ai-use-cases  
Author: Dr.-Ing. Marcus Rüb, CEO  
Published: 2026-10-30

After a first round of ideation, most companies do not have too few AI use cases. They have too many. A list of twenty ideas is of little use until someone decides which two or three get budget now, which ones wait, and why. This guide covers the decision logic: choosing criteria, weighting them, applying a scoring method and turning the result into a portfolio you can defend in a board meeting.

## What does it mean to prioritise AI use cases?

Prioritising AI use cases means scoring an existing list of candidates against criteria agreed in advance, and turning the scores into a decision. The output is not just a ranking. It is a short list of what starts now, what gets prepared, and what is deliberately parked.

That makes it a different job from generating ideas. If you do not have a candidate list yet, start with [Finding AI Use Cases: Methods That Work in Mid-Sized Companies](https://www.ruebstudio.de/en/guides/finding-ai-use-cases/how-to-find-ai-use-cases), which covers process interviews and document analysis. This guide picks up where that one ends: the list exists and now it has to get shorter.

A good prioritisation has three properties:

- **Traceable.** Anyone involved can see why a use case ended up at the top or the bottom.
- **Comparable.** Every candidate is rated on the same scale against the same criteria.
- **Revisable.** New evidence, for example from a pilot, can be fed back in without starting over.

## Which criteria should you score AI use cases on?

Four criteria cover what matters in most organisations: business value, feasibility, risk if the system gets it wrong, and strategic fit. Add a fifth, urgency, if deadlines or market pressure play a real role. Keep the set small and make sure the criteria do not overlap.

**Business value.** How often does the task happen, how much working time does it absorb, and what does an error or delay cost today? A quoting process that handles many enquiries a week carries more weight than an annual report. Score value against the process as it runs today, not against a hoped-for future state.

**Feasibility.** Is the required data digital and accessible? Are there interfaces to the systems involved, such as ERP, CRM or the document store? Is the process defined clearly enough to check whether an AI agent did it correctly? Feasibility is the criterion most often overrated in ideation sessions.

**Risk.** What happens when the system is wrong? A misrouted support ticket gets corrected by a person. A flawed ranking of job applicants can have legal consequences. Regulation belongs here as well. The [European Commission](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai) lists CV-sorting software and credit scoring among high-risk uses with strict obligations, while spam filters fall into the minimal-risk group with no specific rules. That is no substitute for a legal review, but it flags early where the extra effort sits.

**Strategic fit.** Does the use case support a goal leadership already cares about, such as faster quotes, less dependence on a few experts, or better service? Use cases without that link tend to lose their sponsor once the pilot ends.

**Urgency (optional).** Is there a reason to act now, such as a key expert retiring or a legacy system being switched off? This maps to what WSJF calls time criticality.

Avoid criteria that measure the same thing twice. Separate columns for "data quality" and "feasibility" mean one problem gets counted twice.

## How should you weight the criteria?

Weights define how much each criterion contributes to the total. They have to be fixed before scoring starts, and senior management should own them. If weights are adjusted after the scores are in, any desired outcome can be engineered, and the exercise loses credibility.

A reasonable starting point gives value and feasibility similar weight and the other two a little less. For example:

- Business value: 35%
- Feasibility: 30%
- Risk (scored inversely, a high score means low risk): 20%
- Strategic fit: 15%

This split is an assumption, not a standard. A company new to AI that mainly needs to build trust will weight feasibility and risk more heavily. A company with a clear growth agenda will give strategic fit more room.

Two rules help:

1. **Keep knock-out criteria separate.** Some things do not belong in the weighting at all. A use case that would amount to a practice banned under the EU AI Act is not scored low; it is not scored.
2. **Run a sensitivity check.** Shift the weights by a few points. If the top three swap places straight away, they are effectively tied and the decision needs other arguments.

## Which prioritisation methods work for AI use cases?

Four methods are common: the value–feasibility matrix, weighted scoring, RICE and WSJF. They differ in how many criteria they capture and how strongly they account for effort and timing. None is right in itself. What matters is that everyone involved uses the same one.

| Method | Criteria | Strength | Weakness |
|---|---|---|---|
| Value–feasibility matrix | 2 axes | fast, good for workshops | ignores risk and strategy |
| Weighted scoring | 4–6, your choice | fits your organisation | decimals suggest false precision |
| RICE | Reach, Impact, Confidence, Effort | makes uncertainty explicit | built for product features |
| WSJF | Cost of delay divided by size | captures urgency | relative estimates need practice |

**Value–feasibility matrix.** Two axes, four quadrants: high value and high feasibility first, low value and low feasibility dropped. It works well for a first sort in a workshop. A separate guide in this section goes deeper into defining the axes and the usual placement mistakes; here it is one tool among several.

**Weighted scoring.** Each use case gets a score per criterion, for instance on a 1 to 5 scale. Scores are multiplied by the weights and summed. The benefit is that you capture exactly the criteria that matter to you.

**RICE.** Developed at [Intercom](https://www.intercom.com/blog/rice-simple-prioritization-for-product-managers/) for product planning, RICE multiplies reach, impact and confidence and divides by effort in person-months. Impact uses a fixed scale from 0.25 (minimal) to 3 (massive); confidence uses steps of 100%, 80% and 50%. For AI work the confidence factor is the useful part, because it forces teams to state how solid their value estimate really is. Reach is harder to translate when the "users" are internal processes rather than customers.

**WSJF.** Weighted Shortest Job First comes from the [Scaled Agile Framework](https://framework.scaledagile.com/wsjf) and divides relative cost of delay by relative job duration. Cost of delay combines user and business value, time criticality, and risk reduction or opportunity enablement. It suits organisations that already plan in an agile way and keep AI initiatives in the same backlog as other projects.

## What does a weighted scoring matrix look like in practice?

A worked example with five typical candidates shows the logic. Scores are illustrative, on a 1 to 5 scale, with risk scored inversely (5 = low risk). Weights as above: value 35, feasibility 30, risk 20, strategic fit 15.

| Use case | Value / Feasibility / Risk / Fit | Score | Decision |
|---|---|---|---|
| Draft quotes from customer enquiries | 4 / 4 / 4 / 4 | 4.00 | quick-start candidate |
| Triage of support tickets | 3 / 5 / 5 / 3 | 4.00 | quick-start candidate |
| Demand forecasting | 5 / 2 / 4 / 5 | 3.90 | larger bet, fix data first |
| Search across technical manuals | 3 / 3 / 4 / 4 | 3.35 | prepare |
| Pre-screening job applications | 3 / 3 / 1 / 2 | 2.45 | park, high-risk area |

Three lessons stand out:

- **Ties are normal.** Quote drafting and ticket triage score the same for different reasons. One offers more value, the other carries less risk. The tie is broken by asking what your first project needs to prove, not by adding decimals.
- **A small gap is no gap.** Demand forecasting sits just behind, yet its profile is completely different: high value, low feasibility. It does not belong at the bottom of the list; it belongs in a different category.
- **Risk can dominate.** Applicant pre-screening falls back because of its risk score. In cases like this, treat risk as a threshold rather than a weight.

For more detail on the two quick-start candidates, see the overview pages on AI agents for [quotation](https://www.ruebstudio.de/en/ai-agents/quotation) and [technical service](https://www.ruebstudio.de/en/ai-agents/technical-service).

## What do you need before you can prioritise reliably?

You need three things: a candidate list where every use case is described in the same way, the right people in the room, and at least a rough view of where the data lives. Without these, the group ends up scoring assumptions instead of projects.

**Comparable one-pagers.** Describe each candidate on a single page: the process, the roles involved, frequency, the data and systems used, the expected output, and what happens when it goes wrong. Without this, every participant scores a different mental picture of the same idea.

**The right people.** The business owner knows the process, IT knows the data and interfaces, privacy and legal know the limits. Senior management sets weights and budget. When IT is missing, feasibility is almost always scored too optimistically.

**A quick data check.** For the top candidates, an hour per use case is often enough: where is the data, in what format, and who is allowed to access it? This short check tends to reshuffle the ranking more than any debate about weights. For a fuller review of data, systems and error tolerance, see [assessing AI use case feasibility](https://www.ruebstudio.de/en/guides/evaluating-ai-use-cases/ai-feasibility-assessment).

## How do you run the prioritisation step by step?

The process has six steps and usually fits into two or three working sessions. The order matters: criteria and weights first, scoring second, decision last. Reverse that order and you end up justifying a decision that was already made.

1. **Clean the list.** Merge duplicates and break oversized ideas into individual process steps. "AI in sales" is not a use case; "drafting the first reply to emailed enquiries" is.
2. **Apply knock-out criteria.** Anything legally off-limits or against company policy goes before scoring starts.
3. **Fix criteria and weights.** Four to six criteria, weights set by leadership and written down.
4. **Score independently.** Each participant scores alone first, then the scores are compared. Large gaps are useful: they show where people understand the use case differently.
5. **Test sensitivity.** Nudge the weights and see whether the top of the list holds.
6. **Decide the portfolio.** Turn the ranking into a decision about what starts, what is prepared and what waits (see the next section).

Short anchor descriptions for each scale point make scoring far more consistent. "Value 5" might mean the task happens daily and ties up several people; "Value 1" might mean it is rare and takes little time. Without anchors, everyone reads the scale differently.

## How do you turn a ranking into a portfolio?

A ranking does not tell you what to do next. That takes a portfolio decision: one or two quick starts with high feasibility, one larger initiative with high value whose prerequisites you start building now, and an explicit list of what waits. This mix builds experience without losing sight of the bigger prize.

Three buckets work well:

- **Start now.** High feasibility, manageable risk, noticeable value. These projects build skills and trust inside the company. In the example, quote drafting or ticket triage.
- **Prepare.** High value but missing prerequisites, usually data. The work here is not the AI itself but the groundwork, such as a clean data store for demand forecasting.
- **Park.** Low value, high risk or an unclear process. Parking is not rejecting: better models, clearer regulation or new data can change the score later.

The portfolio view prevents two common failures: running only small projects that never get past the pilot, and fixating on the biggest initiative before the organisation has any hands-on experience. If you want support building and steering such a roadmap without hiring a full-time role, see the page on a [fractional head of AI](https://www.ruebstudio.de/en/fractional-head-of-ai).

## What are the limits and risks of prioritisation?

Every scoring method produces a number that looks more precise than it is. The inputs are estimates, often made by people with limited AI experience. The biggest risk is not choosing the wrong method; it is trusting the result blindly and forgetting that scores age quickly.

**False precision.** A score of 3.9 versus 4.0 is not a difference. Round to one decimal place and treat close candidates as equal.

**Biased scoring.** Whoever proposed a use case tends to rate it higher. Independent first scoring, plus one person whose job is to argue the other side, helps.

**Outdated assumptions.** Something that was technically hard at scoring time may be feasible a few months later with newer models. Regulation moves too: according to the [European Commission](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai), the AI Omnibus pushed the high-risk deadlines back, to 2 December 2027 for Annex III use cases and 2 August 2028 for AI embedded in regulated products. A later date does not make those use cases any cheaper to get right.

**No substitute for risk management.** Prioritisation treats risk coarsely. For a structured approach, the [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework), published in January 2023, organises the work into four functions: Govern, Map, Measure and Manage. For specific legal questions, consult a qualified adviser; this guide is not legal advice.

**When a formal method is overkill.** With only two or three candidates, a structured conversation is enough. Scoring pays off when the list is long or several departments compete for the same budget.

## How do you get started?

Start small. Take your existing idea list, write one-pagers for the ten most promising candidates, and score them in a single session with the business owners, IT and leadership. The goal of the first round is a decision on one or two starting projects, not a perfect ranking.

In our own operations, re-scoring the remaining candidates after the first pilot has proved worthwhile. What the pilot teaches you, for example how much work the data integration really took, makes the estimates for the rest of the list far more realistic.

If you do not have a candidate list yet, go one step back to [finding AI use cases](https://www.ruebstudio.de/en/guides/finding-ai-use-cases/how-to-find-ai-use-cases). If you want to build criteria, weights and one-pagers together with several departments in one day, a facilitated [workshop](https://www.ruebstudio.de/en/workshop) is a practical format. For the starting projects, [planning an AI pilot](https://www.ruebstudio.de/en/guides/evaluating-ai-use-cases/ai-pilot-project) covers scope and success criteria, and [calculating the ROI of AI projects](https://www.ruebstudio.de/en/guides/evaluating-ai-use-cases/how-to-calculate-ai-roi) helps firm up the value estimate. More guides are in the [guides section](https://www.ruebstudio.de/en/guides).

## Frequently asked questions

### How many criteria should an AI use case prioritization matrix have?

Four to six is usually enough. Adding more rarely changes the top of the list, but it makes scoring slower and gives people room to rescue a pet project with a tailor-made criterion.

### Should we use RICE or WSJF instead of our own scoring model?

Both come from product and portfolio planning. If your teams already use one of them, reuse it. If not, a weighted scoring model with your own criteria is easier to explain to a management board the first time round.

### Who should take part in scoring?

The business owner of the process, IT for data and systems, and someone covering privacy or legal. Senior management sets the weights before scoring starts, not afterwards.

### How often should we repeat the prioritisation?

At the latest after each completed pilot, and whenever models or regulation change significantly. A quarterly review is a practical starting point, not a fixed rule.

## Sources

- [Intercom: RICE, simple prioritization for product managers](https://www.intercom.com/blog/rice-simple-prioritization-for-product-managers/)
- [Scaled Agile Framework: Weighted Shortest Job First (WSJF)](https://framework.scaledagile.com/wsjf)
- [European Commission: AI Act regulatory framework and timeline (checked October 2026)](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai)
- [NIST: AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework)
