Should we start with the highest-value AI opportunity?

Usually not. Start with the highest-confidence opportunity that builds capability the larger items will reuse. The largest opportunity typically carries the weakest data and the heaviest governance load, which turns it into a slow first project.

How many AI projects should run at once?

For most mid-market companies, one in delivery and one in preparation. Parallelism fails on shared constraints — the same data owners, the same reviewers, the same integration backlog.

AI Opportunities

How to prioritize AI projects: a scoring model that survives review

A transparent scoring model for prioritizing AI opportunities across value, confidence, effort, risk and readiness — plus how to sequence the first twelve months.

Miguel Torres, Founder, Merjora · Published 22 August 2026 · Updated 3 September 2026

In short

Prioritize AI opportunities with an explicit score across five dimensions — annual value, estimate confidence, implementation effort, residual risk and organisational readiness — then sequence so that the first delivery is the highest-confidence item that also builds a capability the next items need. Ranking purely by estimated value reliably produces a first project that stalls, because the largest opportunities usually carry the weakest data and the heaviest governance burden.

Key takeaways

Score five dimensions, not one. Value alone is a misleading ranking key.
Sequence for capability build-up, not just for size.
Publish the score inputs; a score nobody can inspect will not be trusted.
Re-score quarterly — readiness changes faster than value.

The five dimensions

Dimension	What you are measuring	Typical evidence
Value	Annual gross benefit range	Volume × handling time × loaded cost, plus error/rework
Confidence	How much rests on assumption	Whether volumes and times are measured or estimated
Effort	Build and integration load	Systems touched, data preparation, change management
Risk	Consequence of a wrong output	Regulatory exposure, customer impact, reversibility
Readiness	Ability to absorb the change	Data access, process ownership, skills, governance

Turning scores into a sequence

Rank by value-weighted confidence first. That produces a shortlist you can actually defend. Then apply two sequencing rules: the first delivery should be reversible, and it should build reusable plumbing — document ingestion, evaluation harness, review UI — that later items inherit.

The second wave can then take on higher-value, higher-risk work with the benefit of a working measurement baseline and an organisation that has seen the pattern once.

The opportunity matrix, used properly

A value-versus-effort matrix is useful for communication, not for decisions. It hides risk and readiness, which are the two dimensions that most often kill delivery. If you present a matrix to a board, annotate each bubble with its confidence level and readiness constraint.

Re-scoring cadence

Quarterly: readiness and effort, which move as integrations and skills land
Twice a year: value baselines, using actual volumes rather than the original estimate
After every delivery: recalibrate confidence using what the last project taught you

Frequently asked questions

Should we start with the highest-value AI opportunity?: Usually not. Start with the highest-confidence opportunity that builds capability the larger items will reuse. The largest opportunity typically carries the weakest data and the heaviest governance load, which turns it into a slow first project.
How many AI projects should run at once?: For most mid-market companies, one in delivery and one in preparation. Parallelism fails on shared constraints — the same data owners, the same reviewers, the same integration backlog.

Where would this apply in your business?

Merjora maps your workflows, scores each opportunity and quantifies the likely value range before you spend budget.

Discover your AI opportunities