Field guide / 3. Prioritize
How to prioritize AI use cases and investments
How should organizations prioritize AI use cases?
First exclude or defer opportunities that lack an owner, evidence, feasible delivery, or acceptable controls. Score the remaining candidates with a consistent rubric covering value, effort, feasibility, and risk, while showing uncertainty. Then sequence them against strategy, dependencies, budget, and delivery capacity. Accountable leaders approve the next investment stage; a ranking informs that decision but does not make it.
The method in short
Qualify before ranking
Use the discovery and evaluation evidence: objective, process, people, problem, intervention fit, value, feasibility, and risk. Do not let a high score override a non-negotiable constraint.
Normalize and score consistently
Use documented dimensions, anchored scales, agreed weights, and a rubric version. Keep uncertain inputs visible; do not silently turn missing evidence into an average score.
Sequence a feasible set of commitments
Compare strategic contribution, prerequisites, shared resources, concentration risk, and available capacity. Separate learning experiments from delivery-ready work.
Approve, record, and revisit
Leaders authorize bounded investment with success and stop criteria. Record rationale and overrides in the portfolio; re-evaluate when evidence, strategy, cost, or risk changes.
The shared framework
Fourteen stages, from objective to portfolio.
Every guide uses this sequence to connect a business problem to an accountable investment. Revisit earlier stages when new evidence changes the assessment.
Stage 1: Business Objective
Name the outcome the organization is trying to move, so every later idea has something to be measured against.
Stage 2: Department
Locate the team that owns the objective and the work that feeds it.
Stage 3: Process
Describe how the work actually runs today: steps, handoffs, systems, and exceptions.
Stage 4: People
Capture who performs and depends on the work, and what they know that no document records.
Stage 5: Problem
State the friction in plain terms: delay, rework, error, cost, or risk, with evidence.
Stage 6: Opportunity
Frame a specific improvement that addresses the problem, not yet tied to any technology.
Stage 7: AI/Automation Fit
Compare AI, RPA, IDP, workflow automation, process redesign, and no technology change.
Stage 8: Value
Estimate benefit in the objective's own terms, with assumptions written down.
Stage 9: Feasibility
Check data, systems, skills, and change capacity needed to deliver.
Stage 10: Risk
Surface compliance, quality, security, and adoption risks and the controls they need.
Stage 11: Scoring
Rate each opportunity on the same dimensions so comparisons are consistent and explainable.
Stage 12: Prioritization
Sequence opportunities by score, dependencies, and capacity, with reasons people can challenge.
Stage 13: Investment
Match approved effort to the sequence; people with budget authority decide what gets funded.
Stage 14: Portfolio
Track opportunities over time as a living set: delivered, in flight, waiting, and retired.
Use eligibility gates before a weighted score
Weighted scoring is a comparison tool, not a way to average away unacceptable conditions. Define mandatory gates before reviewing the list. An opportunity with strong potential value may still need a hold if the process has no owner, required data cannot be used, or a consequential failure has no adequate control.
Evidence and ownership gate
Require a bounded problem, accountable owner, baseline evidence, and a measurable intended outcome. Return vague ideas to discovery with a specific question rather than rewarding confident language.
Feasibility and control gate
Require a credible delivery path and review by the relevant control owners. A failed gate means investigate, redesign, hold, or stop—not lower the risk weight until the candidate passes.
Intervention gate
Check that AI remains appropriate after comparing RPA, IDP, workflow automation, process redesign, and no technology change. Prioritize the business improvement, not an AI adoption target.
Build a transparent prioritization rubric
The following is a suggested editorial rubric for a human review process, not a specification of IntakeOS's scoring configuration. Organizations should define dimensions and anchors that match their objectives, then apply them consistently to comparable candidates.
If using a weighted total, first align scale direction so a higher rating always means a more desirable result. A normalized weighted average can combine ratings with agreed weights; retain the individual ratings and evidence beside the total. Do not combine raw currency, percentages, and subjective ratings without a defined conversion.
Agree weights and anchors before seeing favored candidates. Record the rubric version and calibration decisions. If two reviewers disagree, resolve the evidence or interpretation rather than quietly editing the total. Consistent inputs matter more than an elaborate formula.
Strategic contribution and value
Ask which agreed business outcome improves and by how much under conservative assumptions. Define anchors in terms of your own materiality thresholds. Distinguish measured benefit, estimated benefit, and benefits that cannot yet be quantified.
Feasibility and delivery effort
Assess data readiness, integration, process stability, supportability, adoption, and specialist effort. Anchor readiness to evidence such as confirmed access or a validated test, not subjective optimism.
Risk and reversibility
Consider residual exposure after controls, blast radius, recovery, and reversibility. Non-negotiable limits remain gates. Within acceptable bounds, lower exposure or easier recovery can affect sequencing.
Time to learning and dependencies
Ask how soon a bounded test could resolve uncertainty and whether prerequisite work benefits other candidates. A useful enabling investment may have little standalone return but unlock several assessed opportunities.
Handle uncertainty without false precision
Keep evidence confidence separate from attractiveness. A high-potential candidate with unverified data is different from a modest opportunity with validated inputs. Label measured, estimated, and missing inputs; show conservative and favorable scenarios when assumptions materially affect rank.
Test ranking sensitivity. Would reasonable changes to adoption, implementation cost, review burden, or weights reverse the order? If so, do not present the exact rank as settled. Either gather the decisive evidence or treat the candidates as a comparable group.
Use separate queues for learning and delivery. A small assessment can be justified because it resolves a consequential unknown, even when a large implementation is not yet justified. Its success criterion is learning that changes a decision, not an unsupported savings target.
A ranked list is not a delivery plan
Check shared capacity
Identify competing demands on data teams, integration engineers, reviewers, business owners, and change management. Select a feasible set of commitments rather than starting every high-scoring candidate.
Map prerequisites
Sequence access, data cleanup, process redesign, and platform changes before dependent solutions. Avoid counting the same enabling cost or business benefit independently in every candidate.
Balance concentration
Consider dependence on one supplier, one fragile data source, or one business team. A portfolio can look diversified by department while sharing the same operational failure point.
Keep mandatory work distinct
A regulatory or contractual obligation may not compete on discretionary ROI. Record its basis and scope separately rather than distorting the scoring rubric to force it to the top.
Illustrative decision: experiment, deliver, or fix the process
Imagine three candidates: an AI-assisted document review with uncertain correction effort; a workflow routing change with confirmed rules; and an automation request caused by unclear approval ownership. This is a hypothetical decision exercise, not a customer result or an ROI comparison.
The document review may earn a bounded experiment to measure quality and review burden. The routing change may be delivery-ready if integration and controls are confirmed. The approval request may return to process redesign. Ranking all three as implementation projects would hide their different readiness states.
If the same integration specialist is needed for the experiment and routing change, the sponsor must choose a feasible sequence. The rationale should name the outcome, evidence, dependency, and opportunity cost—not simply cite the highest total.
Make funding a separate, accountable decision
For each selected commitment, name the sponsor, approved scope, resource envelope, owner, expected outcome, controls, acceptance criteria, and stop or rollback condition. Approve the next appropriate stage rather than treating early discovery as permission for full rollout.
Retain the original assessment and explain any override of the ranked order. A strategic dependency or constrained resource can justify a different sequence, but the reason should be visible. Revisit the decision when evidence or conditions change, not only when a project fails.
Common mistakes include changing weights to favor a sponsor's idea, equating capacity with cash savings, assuming missing evidence means low risk, and mistaking deterministic scoring for objective truth. Repeatable rules improve consistency; they cannot correct unsupported inputs by themselves.
Continue the guide
Common questions.
How do you prioritize AI investments?
Qualify each opportunity, apply a consistent evidence-based scoring rubric, and sequence eligible candidates against strategy, dependencies, risk, budget, and delivery capacity. Leaders then approve a bounded investment stage with success and stop criteria. Ranking is an input to that decision, not approval.
Should the highest-scoring AI use case always be funded first?
No. A high-scoring candidate may depend on unavailable data, shared specialists, prerequisite work, or an unresolved control. Prioritization must consider those constraints and the feasible portfolio, with an explicit rationale when the funded sequence differs from rank.
What does deterministic scoring mean?
Deterministic scoring applies explicit rules to structured inputs so the same inputs and configuration produce the same result. It supports consistent comparisons and traceable reasoning, but does not prove the inputs are correct or replace human investment and risk decisions.
How should uncertain AI opportunities be ranked?
Show uncertainty alongside the score, distinguish measured from estimated inputs, and test whether reasonable assumption changes reverse the order. Put candidates needing evidence into a learning queue rather than presenting them as delivery-ready investments.
Related practice notes
For focused interview techniques, examples, and scoring detail alongside this field guide:
A note on the product
IntakeOS uses 8-dimension deterministic scoring to qualify structured intake data consistently, while VARA gathers the process context behind the assessment. This helps reviewers compare candidates and prepare documented implementation handoffs instead of rebuilding the rationale for every meeting. IntakeOS does not make investment decisions for you: leaders own priorities, budget approval, and risk acceptance, and must validate the underlying assumptions.
IntakeOS supports this work and does not make funding decisions; those stay with your leaders. Learn about VARA, how we handle data on the trust page, or the wider platform.
See a demo