A 10-minute routine to find and prioritise AI automations

Executive overview

Most business owners stall on AI not because it can't help, but because it can help with too many things. This routine cuts through that paralysis in 10 minutes. It gives you a live list, a prioritisation matrix, and a small-experiment method to quickly test whether any task is actually automatable today.

Start with quick wins to build confidence, then move up to high-impact automations — never start with complex, high-risk tasks.

Table stakes before you begin

• You must know what "good" looks like for any process you want to automate — a quality threshold to measure against.
• The process needs documentation, or you must be willing to create it (AI can help with this).

Building your automation list: the rule of three

• Add a task to your list if it happens three or more times — daily, weekly, monthly, or quarterly.
• One-off tasks are ad hoc; twice is a fluke. Three times signals a repeatable pattern worth automating.
• The list isn't just for the business owner — team leads and frontline workers should maintain their own lists.
• Aggregate and cross-reference these lists; keep it simple (an Apple Note works fine).

Prioritising with the three-variable matrix

Each item on your list gets rated High / Medium / Low across three variables:

• Risk — high-value, low-volume tasks carry the most exposure if AI gets it wrong.
• Business impact — either revenue generated or time saved; quantify both.
• Ease of implementation — how quickly can it be running, and how easy to maintain?

Where to start on the matrix

• Quick wins first — low impact, low risk, easy to implement. Builds team confidence in AI.
• Priority zone next — high impact, low-to-medium risk, easy to implement. The sweet spot; spend most of your time here.
• Strategic zone last — high impact but hard to implement and/or medium-to-high risk. Only tackle these after you've banked wins.
• Avoid zone always — low impact, hard to implement, high risk. Many owners start here by accident, get burned, and write off AI entirely.

Understanding risk: the last-mile problem

• Not every task is off-limits — the issue is automating the final decision.
• Hiring and firing: AI can filter candidates and aggregate performance data, but should not make the final call.
• Client closing: AI can surface negotiation angles, but the relationship owner must close.
• AI's reliability and contextual depth aren't yet sufficient for high-stakes final decisions.
• Always measure ROI before and after any automation — know the baseline cost and the expected gain.

Running small experiments

• Don't automate the whole process first. Identify the core segment where AI does most of the work, and test that segment alone.
• Create a ChatGPT, Claude, or Gemini project with a focused system prompt tuned to the task.
• Run the AI on that segment and measure the output against your quality threshold.
• Run 5–6 iterations across varied inputs to assess consistency.
• If it meets expectations, proceed to full automation. If not, move on — don't invest further.

Example — PDF processing: Give AI a system prompt to extract structured data from PDFs; test on multiple varying PDFs; check consistency of output.

Example — Proposal writing: Load existing proposals as a knowledge base; ask the AI to draft a new proposal from a contract brief; compare against your database.

The 10-minute breakdown

1. 2 minutes — add new items to your list using the rule of three.
2. 2–3 minutes — rate each item H/M/L on business impact, ease of implementation, and risk.
3. 5–6 minutes — run one small experiment on the top-priority item; test only the AI-heavy segment.

One experiment at a time. Start small. Iterate.