Score It Before You Fund It
Most organisations have more AI ideas than capacity, and choose between them based on who advocated most persuasively. Scoring candidates against the same criteria turns that into a defensible decision, and frequently changes the answer.
Run this once per candidate and record the scores. Comparing three scored use cases is far more useful than debating one.
Score This Use Case
One candidate at a time. Run it again for each.
A structured way to compare candidates rather than a definitive verdict. Score several and compare: the relative ranking is more reliable than any single absolute score.
How to Use the Result
The score measures deliverability alongside value. A high-value use case that scores poorly is not a bad idea. It is a later one, and knowing that is the point.
Your first should be visibly useful and boring
The best first project delivers something people can see working within weeks. It buys the credibility and the organisational learning needed for anything ambitious. Starting with the most transformational idea is the most reliable way to stall.
Sponsorship outranks value
Between two use cases with similar scores, take the one with an engaged sponsor who wants it. Adoption is the variable that decides outcomes, and no business case survives an owner who is indifferent to the result.
Time to value compounds
A modest result delivered in eight weeks builds momentum, learning and appetite. An ambitious result delivered in eight months arrives after the sponsor has changed, the priorities have moved, and enthusiasm has cooled.
The Five Criteria
Fifteen questions across five dimensions. A candidate needs to be adequate across all of them, not exceptional in one.
Value
The size and certainty of the benefit, and whether anyone will be able to prove it afterwards.
Feasibility
Whether the problem is well understood and whether the approach is proven rather than speculative.
Data readiness
Whether the data required exists, is accessible, and is of usable quality today.
Risk
What a failure would cost, and whether human oversight can be built in proportionately.
Sponsorship
Whether a named owner wants it, will resource it, and will be there when it lands.
The Use Cases That Score Well Almost Everywhere
Across Australian organisations, the same categories consistently score highest. If you are choosing a first project, start here.
Finding information people already struggle to find
Retrieval over an existing document corpus, policies, procedures, technical documentation, prior advice. The value is immediate and obvious, the data already exists, and failure means a poor answer rather than a business consequence.
- Policy and procedure lookup for staff across a large document set
- Technical documentation search for support or field teams
- Prior work and precedent retrieval in professional services
- Onboarding support that reduces demand on experienced staff
Structured extraction from documents
Pulling defined fields out of invoices, forms, contracts and applications. Well-understood, accuracy is measurable, and it replaces work nobody enjoys. The benefit is easy to quantify because the manual baseline is known.
- Invoice and purchase order data capture
- Application and intake form processing
- Contract clause extraction and comparison
- Compliance document checking against a defined rule set
Triage and routing of inbound work
Classifying and directing enquiries, tickets, applications or correspondence. Low risk when the system routes rather than resolves, since being wrong means something goes to the wrong queue and gets moved.
- Shared inbox and correspondence classification
- Support ticket prioritisation and assignment
- Application routing to the right assessment team
- Flagging items that breach a threshold for human attention
What to avoid as a first project
Anything customer-facing without a human in the loop, anything making consequential decisions autonomously, and anything depending on data you have not inspected. Each is achievable later; none is a sensible place to start.
- Autonomous customer-facing decisions with no human review
- Anything with regulatory consequence if it gets one wrong
- Use cases requiring data you have not yet examined
- Projects whose sponsor is enthusiastic but has no capacity
Next Steps
AI Business Case Calculator
Take your highest-scoring candidate and build a risk-adjusted case.
Build the case →AI Maturity Scorecard
Check whether the organisation can deliver what you are about to fund.
Score maturity →AI Project Cost Estimator
Estimate what the winning candidate will actually cost to build.
Estimate the cost →Frequently Asked Questions
How many use cases should we score?
Three to six is the practical range. Fewer than three gives no basis for comparison and the exercise becomes a justification rather than a decision. Beyond six it turns into a project of its own and stalls. Score them, record the numbers, and keep the list. The second and third candidates become your natural follow-on projects once the first is delivered, and you will already have done the analysis while the context is fresh.
Should the highest-value use case always win?
No, and this is the most common prioritisation error. Value is only one of five dimensions, and the highest-value candidate is frequently the one with the messiest data, the widest scope and the most stakeholders, which is exactly why it has not been done already. Delivering a mid-value use case successfully builds the capability, credibility and organisational learning needed to attempt the big one. Organisations that lead with their most ambitious idea usually spend a year on it and deliver nothing.
What score means a use case is ready to fund?
Above seventy-five per cent means it should proceed and will likely deliver close to what the business case predicts. Between fifty and seventy-five is workable but expect complications, most often in data quality or scope, so build in contingency and keep the scope tight. Below fifty per cent, funding it now is likely to produce an expensive lesson. That is not a permanent verdict. It usually means the data needs work or the problem needs defining more precisely before the idea is ready.
How do we handle a use case with high value but poor data readiness?
Split it. Make the data preparation an explicit, separately funded first phase with its own deliverable, rather than burying it inside an AI project where it will consume the budget and be invisible. That is more honest, easier to approve, and produces value regardless of whether the AI initiative proceeds, clean, accessible data serves reporting, compliance and every future project. It also means the AI phase, when it comes, can be scoped accurately rather than optimistically.
What if the sponsor is enthusiastic but has no time?
Treat that as a red flag rather than a detail to work around. Sponsorship means allocated time, not expressed interest, and a sponsor without capacity produces the same outcome as no sponsor, decisions wait, subject-matter input does not arrive, and the project drifts. Either secure a genuine time commitment before starting, find a different sponsor who has capacity, or choose a different use case. Proceeding on enthusiasm alone is one of the most reliable predictors of a stalled project.
Can we use this for comparing vendor proposals too?
It is designed for comparing use cases rather than proposals, but the data readiness and risk dimensions transfer well to assessing whether a vendor has understood your situation. A proposal that assumes clean data when yours is not, or that omits human oversight for a high-consequence process, is telling you something about how carefully it was scoped. For vendor assessment specifically, the due diligence checklist covers the commercial and technical questions more directly.
Send Us Your Shortlist
Scored three or four candidates and unsure which to back? Send them through with the scores and we will tell you which one we would fund first, and why.