The Prioritisation Problem
Every enterprise AI engagement starts the same way: a long list of potential use cases, each one championed by someone with genuine conviction. Customer service automation. Demand forecasting. Document processing. Predictive maintenance. Fraud detection. The list is never short.
The problem is not a lack of ideas. It’s the absence of a shared framework for making hard trade-offs. Without one, prioritisation becomes political — whoever argues loudest or has the most seniority wins.
The Three-Axis Scoring Framework
After working through this problem with dozens of clients, we’ve converged on a simple three-axis scoring model. Each use case gets scored on:
1. Business Impact (1–5) The potential value if successful — revenue generated, cost reduced, risk mitigated. This should be quantified where possible. “High impact” is not a number.
2. Technical Feasibility (1–5) How achievable is this given current data availability, team capability, and tooling? A use case that requires data you don’t have, or capabilities that are 18 months away, scores lower regardless of its business value.
3. Time to Value (1–5) How quickly can you demonstrate results? Use cases that produce measurable outcomes in 90 days score higher than those requiring a year of infrastructure work before any business value appears.
Multiply the three scores. Use cases with a total above 50 (out of 125) are strong candidates. Below 25, reconsider.
The Forcing Functions
Scoring alone isn’t enough. Two forcing functions prevent the list from inflating indefinitely:
Constraint 1: The executive sponsor test. Every use case on the shortlist needs an executive who will be held accountable for its success and is willing to dedicate team time to it. No sponsor, no priority.
Constraint 2: The dependency audit. Some use cases look high-value but require infrastructure or data that is also a dependency for several others. Sequencing matters. Build the foundation before the house.
Starting Small on Purpose
The clients who make the most progress with AI are consistently the ones who resist the urge to run five pilots simultaneously. One well-resourced initiative with a clear owner and a defined success metric produces more organisational learning — and more business value — than five underfunded experiments.
Pick one. Make it work. Build from there.