Worksheet: The Agentic AI Opportunities Audit

A structured way to decide what to build first, what to build later, and what to decline. You score every candidate process on four questions, compute a priority score, and let the score place it in a quadrant. The discipline is the point: a number beats an opinion, and a shared number beats an argument.

The four questions (your four score columns)

Every process gets four scores, each 1 to 5. Each score answers one plain-language question.

1. Human vs. AI. "Should we use AI for this?" 1 means keep it fully human (judgment, relationships, accountability). 5 means it is a strong fit to shift to an agent (repetitive, rules-based, high volume). This is suitability, not permission.
2. Risk. "Could using AI negatively impact the business?" 1 means low risk if it goes wrong. 5 means high risk: money moves, data leaves, a promise is made, a regulator notices. High is bad here.
3. Business Impact. "Would using AI drive business results, or avoid a loss?" 1 means low impact. 5 means high: hours returned at scale, revenue protected, a penalty or an incident avoided. Count both upside and avoidance of loss.
4. Feasibility. "Can we actually use AI for this?" 1 means not feasible today (data missing, no tool access, unclear rules). 5 means very feasible: the data exists, an agent can reach it, the actions are scoped and clear.

Priority Score and Quadrant

Priority Score = Business Impact + Feasibility + (Human vs. AI) + (6 minus Risk).

Every term rewards a good candidate, and the (6 minus Risk) term turns low risk into points. The score runs 4 (avoid) to 20 (build now). Rank your list by it.

Quadrant comes from two derived axes:

• Value = Business Impact. High when 4 or 5.
• Feasibility with safety = Feasibility, reduced by 1 for every risk point above 3. (Effective feasibility = Feasibility minus the greater of 0 and Risk minus 3.) High when 3 or more.

A candidate that scores 5 on Business Impact and 5 on Risk is not a Quick Win with ambition. High risk drags its feasibility-with-safety down, and it lands in Strategic Bet, where it belongs. Pretending otherwise is how incidents get scheduled.

The audit table

Fill one row per process. Functional Area, Process Name, and Description are given for the Grand Inflation candidates; you supply Frequency, Avg Hours/Month, the four scores, the Priority Score, the Quadrant, an Owner, a Recommended Next Step, and Notes.

Two of these look like Quick Wins and are not. The AP invoice auto-processor and the official social auto-poster both hold outbound or financial authority, which pushes their risk up and their feasibility-with-safety down. The week will test that instinct. Watch what happens to a process that can pay a stranger or speak for the company without a human in the loop.

Now your organization

List eight processes from your own operation. Aim for spread: at least two that return hours, two that protect revenue, two that reduce risk. Then score and place them exactly as above.

The closing question

Which one process is your first build at home, what is its Priority Score, and what single guardrail makes it safe? One sentence. If the sentence has no number in it, it is not finished.