The AI Automation Prioritization Worksheet

Step 1: your North Star (the one number)

Every business needs one number that tells it whether it is winning. For most of you, that is monthly profit, not revenue, not vanity numbers. Profit is what buys you your time back.

The question that sets it: how much money would you need to make per month, indefinitely, to actually feel free? Not rich. Free.

The filter for every AI tool you will ever be sold: will this move my North Star? Yes, do it. No, do not.

Step 2: your admin leak, in dollars

Reference points from the talk: owners average 16.4 hrs/week on admin (Time etc/Censuswide, 2023). 57% of US work hours are technically automatable (McKinsey, 2025). About $35/hr median total cost of a private-sector worker (BLS, 2026).

Step 3: task inventory

List every recurring task that eats your week or your team's. Not the important ones: the recurring ones. Aim for 15 to 50. Include the embarrassing ones, they are usually the expensive ones. If you cannot name the trigger and the output, it is not a task yet.

Prompt: here is my business in 3 sentences: [...]. List the 15 recurring admin and operations tasks businesses like mine usually lose the most hours to, with a likely trigger and output for each. Then ask me which ones apply.

Step 4: find your biggest domino

Score each task on the five dimensions from the talk: impact on your North Star (1 to 10), difficulty to automate (easy under 1 day = 1, medium under 1 week = 5, hard over 1 week = 10, be honest not heroic), frequency per week (literal number, once a year means ignore it), urgency (1 to 10, when does this need to be fixed by), domino size (1 to 10, if this falls how many other tasks get easier or disappear).

Then sort, in this order, exactly as in the talk: highest urgency, then biggest domino, then highest impact, then highest frequency, then lowest difficulty. Automate the top one first.

Worked example (from the talk): a consultant's lead qualification. Impact 9 (30 to 40% of monthly profit), Difficulty 1 (under a day), Frequency 30+, Urgency 10, Domino 9. That is the domino. Automate it first and watch the rest of the line fall.

Then the 80/20: which 20% of tasks drive 80% of your North Star's growth? Automate or delegate those. Eliminate the rest, especially the stuff you kind of like doing.

My top 3 dominoes (bring these to your OpsJuice intro call):

Step 5: the Hot Mess Checklist

10 questions before you automate anything. AI does not fix bad. It multiplies whatever is there. Answer honestly.

• Do people actually pay for what this process supports?
• Have I run it by hand for 2 weeks and written down every step, in order?
• Do I know exactly where it breaks or gets messy?
• Have I deleted every step that does not need to exist (eliminate before automate)?
• Can I name its trigger and its output in one sentence each?
• Is the data going in clean and consistent, not scattered across five tools in five formats?
• Will automating it move my North Star, or does it just feel good?
• Have I compared the cost to a part-time human doing it? (Sometimes the human is the right move.)
• Do I know which steps need AI judgment, and which can be a simple fixed rule?
• Have I decided where a human approves before anything reaches a customer?

Fewer than 8 checked: fix it by hand first. Automate second.

Step 6: the five functions (your tool map)

Do not get attached to tools. Get attached to functions. You only need the function your number 1 domino requires.

Cost reality: starter $0 to 30/mo. Scaling $200 to 500/mo. Running a real business $1,000 to 3,000/mo.

Build vs buy: buy when the task is common to your industry and a mature tool already does 80%. Build when the task is how you actually win, or the off-the-shelf version costs more per month than the problem does. Neither when you could have eliminated it.

Step 7: standard operating procedure to workflow (the Agentic Layer)

For each of your top 3, fill in the recipe. 80% fixed rules, 20% AI: AI only where a rule genuinely cannot do the job.

Step 8: human-in-the-loop decision tree

For every workflow, ask in order:

1. Is it a high-stakes decision? (hiring, firing, lawsuits, refunds over $500) → human approves.
2. Could the customer be upset or in crisis? → route to a human immediately.
3. Is it a sensitive topic? (political, medical, wedding-related) → human approves.
4. Does it move money irreversibly? (sending an invoice is fine, paying one means human approves)
5. Is this one of its first 20 to 50 runs? → human reviews every output.
6. Is data missing or ambiguous? → hold queue: escalate, do not guess.
7. If it goes wrong, can I fix it within 24 hours without losing trust? Yes, let it run. No, keep a human gating it.

Step 9: the Knowledge System

One inbox per business (voice memos, notes, screenshots, call notes) goes to AI, which sorts it into: knowledge base, action items, follow-ups, ideas to revisit. Query it any time.

Starter prompts:

• Sort everything below into knowledge base, action items, follow-ups, and ideas. Tag each by client and project.
• What did I decide about [topic] last month?
• Based on everything captured this quarter, what is the bottleneck I am not seeing?

Step 10: prompt starter pack

5 workflows that pay back first.

1. Instant lead response: you answer new inquiries for [business] in under 5 minutes. Read the message, identify the service needed and urgency, reply in this voice: [paste 2 past replies]. If details are missing, ask one question. Never quote a price, offer a booking link.
2. Follow-up: write a 3-touch follow-up (day 1, 3, 7) for leads who did not reply. Friendly, short, one clear next step each.
3. Scheduling and reminders: write a reminder sequence (48h, 24h, 2h) that confirms, lets them reschedule in one tap, and includes these prep instructions: [...].
4. Post-job review request: write a thank you plus review request sent 2 hours after a job, with this next-step offer: [...].
5. Invoice follow-up: write polite reminders for unpaid invoices at 3, 10, and 21 days overdue. Firm by the third, never rude.

Step 11: your 30 / 60 / 90

Check yourself for Mirage PMF every quarter: is margin growing with revenue? Is profit per employee climbing? Does every new customer need custom work? Can you point to one number AI moved?