AI use case prioritisation is the problem hiding behind every AI budget meeting. Most organisations have more AI ideas than they can fund: every function has a list, and every idea comes with a sponsor who is sure it will pay off. Budget, clean data and engineering time run out long before the list does, and a year later the business is looking at a handful of stalled pilots and asking where the return went.
The latest data shows how common this is. McKinsey’s State of AI 2026 survey (1,719 respondents across 97 countries, fielded May to June 2026) found that nearly nine in ten organisations now use AI regularly, and 44% say it is scaling across the enterprise, up from 38% a year earlier.
The share reporting any EBIT impact from AI has barely moved, though. It sits at 37%, essentially unchanged from 2025, and only about 6% of organisations qualify as AI high performers. That gap between adoption and impact is what disciplined AI use case prioritisation is meant to close.
Gartner’s forecasts point to why. It expects more than 40% of agentic AI projects to be cancelled by the end of 2027, mostly because of rising costs, unclear business value or weak risk controls. Each of those problems is visible before a project starts, if someone looks for it. Good AI use case prioritisation is simply the habit of looking.
This guide gives you a repeatable AI use case prioritisation framework. It sets out a weighted scoring model with six criteria and clear 1-to-5 anchors, four knock-out gates that park risky ideas early, a value-feasibility matrix and a worked example you can adapt for your own portfolio. If you are still framing the fundamentals, start with what enterprise AI really means.
[CHART, update values: The gap between AI adoption and AI value. Regular AI use: nearly 9 in 10 · Scaling AI across the enterprise: 44% · Any EBIT impact from AI: 37% · AI high performers: ~6%. Source: McKinsey, The State of AI in 2026, published August 25, 2026.]
Key takeaways
- AI use case prioritisation replaces pitch-driven decisions with a weighted score, knock-out gates and a value-feasibility matrix.
- Wider AI use hasn’t lifted financial returns yet: 37% of organisations report any EBIT impact from AI, unchanged from 2025 (McKinsey, 2026).
- Score each use case on six weighted criteria: business value, feasibility, risk, cost, time to value and strategic fit.
- Run four knock-out gates (data, regulation, ownership and measurability) before any idea gets a score.
- Plot the results on a value-feasibility matrix to separate quick wins from big bets.
- Re-score the portfolio every quarter and agree kill criteria at approval, so stalled pilots stop drawing budget.
Step 1: Build a structured use case long list
AI use case prioritisation only works when every idea arrives in the same shape. If one idea shows up as a polished vendor demo and another as a two-page business case, you end up scoring the quality of the pitch.
Pull ideas from business units, operations data (look for where time, errors and rework pile up), customer feedback and any existing automation backlog. If the long list looks thin, our roundup of 10 enterprise AI use cases in 2026 is a useful prompt.
Then write each one up on the same one-page template:
| Field | What to capture |
|---|---|
| Problem statement | The business problem, in the business owner’s words |
| Process and users | Which workflow changes and who uses the output |
| Target outcome | The metric that should move (cost per case, cycle time, revenue, error rate) |
| Current baseline | How that metric performs today |
| Data required | Source systems, volume, quality and access constraints |
| AI pattern | Prediction, document processing, generative assistant, agent, computer vision |
| Business owner | A named person accountable for adoption and results |
If an idea can’t fill in the problem, outcome and owner fields, it isn’t ready to score. Send it back to the sponsor instead of scoring a guess. A clean long list is the foundation of AI use case prioritisation that leadership will trust.
Step 2: Score each use case on six weighted criteria
The scoring model is the core of AI use case prioritisation. Each use case gets a score from 1 to 5 on six criteria, and each criterion carries a weight. Risk and cost are inverted, so a higher number is always better (5 means low risk or low cost). That keeps the arithmetic simple.
| Criterion | Weight | Score 1 weak | Score 3 moderate | Score 5 strong |
|---|---|---|---|---|
| Business value | 30% | Minor efficiency, hard to measure | Clear savings or revenue in one function | Material, measurable impact on a top business KPI |
| Feasibility data and technology | 25% | Data missing or inaccessible; unproven technique | Data exists but needs cleaning or integration | Clean, accessible data; proven pattern |
| Risk inverted: 5 = low risk | 15% | High regulatory, safety or reputational exposure; autonomous decisions | Moderate exposure with human review | Low exposure; internal, reversible outputs |
| Cost and effort inverted: 5 = low cost | 10% | Large build, new platform, high inference cost | Moderate build on existing platforms | Configuration of existing tools |
| Time to value | 10% | More than 12 months to a measurable result | 4 to 12 months | Under 3 months |
| Strategic fit | 10% | Isolated experiment | Supports a departmental goal | Directly advances an enterprise priority or reusable capability |
| Total weight | 100% | Higher is always better. Risk and cost are inverted so a 5 means low risk or low cost. | ||
To get the score, multiply each criterion score by its weight, add them up and multiply the total by 20. That gives you a number out of 100:
The formula
Priority score = 20 × (0.30V + 0.25F + 0.15R + 0.10C + 0.10T + 0.10S)
V business value · F feasibility · R risk (inverted) · C cost and effort (inverted) · T time to value · S strategic fit. Each is scored 1 to 5, so the result lands on a 0 to 100 scale.
Set the weights to match your strategy, then lock them. An operations team focused on cost might give more weight to cost and time to value, while a growth-focused business might push value and strategic fit higher. Agree the weights with leadership before anyone scores, so nobody can adjust the model later to favour a pet project.
Locked weights are what keep AI use case prioritisation objective as the list grows.
Be realistic about cost. In McKinsey’s 2026 survey, one in five respondents said AI operating costs, including token costs, were already limiting how much their organisation uses AI. Score cost on what it takes to run the system for the long haul (inference, monitoring, retraining and support) as well as the initial build. A shared AI platform and LLMOps foundation makes those run costs much easier to see and control.
Score as a group. Ask a business owner, a data or IT lead and someone from risk or compliance to score each use case on their own, then talk through any criterion where their scores are two or more points apart.
Those gaps usually expose the data problem or compliance issue that would otherwise surface halfway through the build. Scoring as a group turns AI use case prioritisation into a shared decision rather than one team’s opinion.
Step 3: Apply knock-out gates before ranking
A weighted score can hide a deal-breaker. An idea with huge potential value can outscore safer options even when there’s no usable data behind it. Knock-out gates catch that. Any use case that fails a gate is parked until the issue is fixed, and it doesn’t get ranked in the meantime.
Gates are the safety net in any AI use case prioritisation process. If regulatory exposure is the blocker, our responsible AI governance and assurance work is usually the fastest way to clear it.
- Data gate. The required data exists, can be accessed legally and is good enough to build and test on.
- Regulatory and ethics gate. The use case is allowed under the regulations and internal policies that apply, and any high-risk decision has a defined human oversight model.
- Ownership gate. A named business owner has committed to adoption, process change and measuring results.
- Measurability gate. A baseline exists for the target metric, so you can prove whether the project worked.
The data gate is the one most ideas fail. Gartner predicts that through 2026, organisations will abandon 60% of AI projects that aren’t supported by AI-ready data.
The measurability gate matters just as much: McKinsey found that AI high performers are twice as likely as other organisations to have defined processes for measuring the impact of their AI initiatives.
Our guide to enterprise RAG architecture, costs and failure modes is a useful reality check when you score feasibility.
Frost & Sullivan Best Practices 2026
Global Company of the Year for Digital CX
11 pages of independent analysis. Free PDF, one short form.
Download the report →Step 4: Plot the value-feasibility matrix
The ranked list gives you an order. The matrix shows you what kind of portfolio you’re building. Plot each use case with a value composite (the average of business value and strategic fit) on one axis and a feasibility composite (the average of feasibility, cost and time to value) on the other. Split both axes at the midpoint of 3. Together, the ranked list and the matrix turn AI use case prioritisation into a portfolio view rather than a queue.
| Quadrant | Profile | What to do |
|---|---|---|
| Quick wins | High value, high feasibility | Fund now and use them to build credibility and reusable capability |
| Big bets | High value, low feasibility | Invest in data and platform foundations first and release funding in stages |
| Fill-ins | Low value, high feasibility | Deliver only if they reuse existing assets at low cost |
| Deprioritise | Low value, low feasibility | Archive and revisit if conditions change |
Most healthy portfolios combine several quick wins that pay back within months with one or two big bets that build longer-term advantage. That balance is the real goal of AI use case prioritisation.
Worked example: scoring four AI use cases
This example is illustrative. It shows how AI use case prioritisation separates four common enterprise ideas using the default weights.
| Use case | Value 30% | Feasibility 25% | Risk (inv.) 15% | Cost (inv.) 10% | Time 10% | Fit 10% | Score /100 | Gates |
|---|---|---|---|---|---|---|---|---|
| A. Invoice and document processing | 4 | 5 | 4 | 4 | 5 | 3 | 85 | Pass |
| B. Customer support agent-assist copilot | 4 | 4 | 3 | 3 | 4 | 4 | 75 | Pass |
| C. AI demand forecasting | 5 | 2 | 4 | 2 | 2 | 5 | 70 | Pass data remediation needed |
| D. Fully autonomous claims decisioning agent | 5 | 2 | 1 | 2 | 1 | 4 | 57 | Fails regulatory gate |
[CHART, keep as is: Illustrative value-feasibility matrix for the four use cases above. A: document processing; B: support agent-assist; C: demand forecasting; D: autonomous claims agent.]
Here’s how to read the result:
- A (document processing) is the clearest quick win. The pattern is proven (see our AI document intelligence case study), the data is available and returns can show up within a quarter, so start here.
- B (agent-assist copilot) is a second quick win. A person reviews every suggestion, which keeps the risk manageable. That human sign-off is exactly what separates a copilot from an agent.
- C (demand forecasting) has the highest value but weak data foundations. Treat it as a big bet: fund a data remediation phase first, then score it again.
- D (autonomous claims agent) would rank near the top on value alone, but it fails the regulatory gate. That lines up with Gartner’s 2026 view that most agent deployments are still narrowly scoped and fully autonomous agents aren’t ready for the majority of enterprise use cases. Redesigned as a human-in-the-loop recommendation tool, it carries far less risk and would likely move into the big bet quadrant.
Not sure which AI use cases to fund first?
Compunnel’s Applied AI practice scores your long list, clears the knock-out gates and turns the top use cases into a governed roadmap, then builds them for production.
Explore Applied AI →Step 5: Govern the whole portfolio
A single AI use case prioritisation workshop gives you a snapshot. Data improves, regulation changes, costs fall and business priorities shift, so the scores go stale within months. A few habits keep the portfolio current:
- Re-score every quarter, updating feasibility, cost and risk as platforms, data and regulation change.
- Set kill criteria when you approve a project. Agree the metric, the target and the deadline that will trigger a decision to stop, scale or pivot.
- Release funding in stages (discovery, pilot, production), with each stage tied to evidence from the one before.
- Favour use cases that build shared data pipelines, governance or platforms that later projects can reuse.
- Give extra credit for workflow redesign. Nearly three-quarters of AI high performers say they have fundamentally redesigned workflows because of AI, up from 55% a year earlier, compared with about a quarter of other organisations (McKinsey, 2026). Score strategic fit higher when a use case changes an end-to-end process.
Common AI use case prioritisation mistakes to avoid
- Starting with the technology. Asking “where can we use agents?” tends to produce demos. Asking “where do cost, delays or errors pile up?” tends to produce value.
- Scoring value without a baseline. If you can’t measure how the process performs today, you can’t prove it improved.
- Forgetting run costs. Inference, monitoring, retraining and support keep costing money long after launch. Put them in the cost score.
- Letting the loudest sponsor win. Locked weights and group scoring keep the decision objective.
- Treating every pilot as permanent. Without kill criteria, stalled pilots can keep drawing budget for years.
How Compunnel Digital helps
Compunnel Digital runs AI use case prioritisation workshops that take you from a long list of ideas to a scored, governed roadmap. Our Applied AI Engineering team then builds and scales the priority use cases, from document intelligence and data platforms to AI copilots and autonomous operations, for clients in healthcare, financial services, the public sector, manufacturing, retail and education.
Have more AI ideas than budget? Get in touch to run an AI use case prioritisation workshop with our team.
Frequently asked questions
What is AI use case prioritisation?
AI use case prioritisation is a structured way to rank potential AI projects so the organisation funds the ones with the best mix of business value, feasibility, acceptable risk, cost and time to value first. It replaces decisions driven by whoever argues hardest with consistent scoring.
What criteria should be used to prioritise AI use cases?
Effective AI use case prioritisation uses business value, data and technical feasibility, risk, cost and effort, time to value and strategic fit. Add knock-out gates for data availability, regulatory exposure, business ownership and measurability, so a high-scoring idea with a deal-breaker doesn’t get approved.
How many AI use cases should we pursue at once?
It depends on capacity, but most organisations do better with a small portfolio: a few quick wins that prove value within months, plus one or two big bets that build long-term capability. Fund only what your data, engineering and change management teams can actually take to production.
How often should AI use case scores be updated?
Re-score the portfolio at least once a quarter, and straight away after a major change such as a new data source, a regulatory update, a platform decision or a shift in business priorities. AI use case prioritisation is a standing governance activity, not a one-off workshop.
Should generative AI and agentic AI use cases be scored differently?
Score agentic AI use cases on the same criteria, but look harder at risk and cost. Gartner’s 2026 CIO and Technology Executive Survey found that only 17% of organisations have deployed AI agents so far, while more than 60% expect to within two years. Agents that act on their own need stronger security controls, and their running costs are harder to predict, so score those two criteria conservatively.
Conclusion
The organisations getting real returns from AI tend to fund fewer projects and choose them more carefully. To put this AI use case prioritisation framework to work:
- Standardise inputs with a one-page use case template.
- Score every idea on six weighted criteria with clear anchors.
- Park anything that fails the data, regulatory, ownership or measurability gate.
- Use the value-feasibility matrix to balance quick wins against big bets, and re-score every quarter.
Take your top three ideas, score them this week and let the results decide what gets funded. Want a second opinion on the scores? Book a prioritisation workshop with our team.
Sources
- McKinsey & Company, The state of AI in 2026: On the road to ROI, August 25, 2026. Online survey of 1,719 participants in 97 countries, May 4 to June 8, 2026.
- Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, June 25, 2025.
- Gartner, Lack of AI-Ready Data Puts AI Projects at Risk, February 26, 2025.
- Gartner, What the 2026 Hype Cycle for Agentic AI Reveals, April 15, 2026 (cites the 2026 Gartner CIO and Technology Executive Survey).
