The suitability rubric

Not every task an agent could do is worth automating. The way to decide is to score each opportunity on two axes: impact, and feasibility with safety. Impact asks how much volume and cost the task carries and whether it sits on the constraint. Feasibility and safety ask how structured and rule-bounded it is, how reversible, how good the data is, and how much it would hurt to get wrong. Plot the two, and the order of work becomes clear.

How do we rank opportunities?

Each candidate task gets a score on both axes, built from plain sub-questions rather than a black box. On impact: the monthly volume, the fully-loaded cost, and whether relieving it unlocks the constraint. On feasibility and safety: how well-structured the inputs are, how rule-bounded the decisions, how reversible an error is, how complete the data, and how exposed the work is to customers or regulation. The scores are estimates, shown with their confidence, not false precision.

What do the four quadrants mean?

High impact and high feasibility is the low-hanging fruit: do it first. High impact but low feasibility is strategic: stage it, and fix the blockers before automating. Low impact but high feasibility is a fill-in: cheap, or skip. Low on both: leave it. The grid stops a business chasing a hard, glamorous project when an easy, valuable one is sitting next to it.

What makes an ideal first pilot?

The first pilot is chosen not for size but for what it proves. The best candidate is high-volume, well-structured, internal rather than customer-facing, reversible, measurable, and low in regulatory stakes. A win there de-risks the whole model on your own data, earns the trust of sceptical owners and anxious staff, and produces the before-and-after that justifies everything after it.

Why measure before deploying?

A saving that was never baselined cannot be claimed. The rubric only works if the cost and the volume it scores are real, captured before any agent goes live, so the result can be proven afterwards. That discipline is what separates a credible estimate from a hopeful one.