How can AI take repetitive work off your plate?

What automation means

An automation is a set of steps that runs on its own when something happens. You have probably used simple ones already: an out-of-office reply, an online form that emails you each new submission, a calendar that sends reminders.

Every automation has three parts:

• Trigger: the event that starts it. A new email arrives, a form is submitted, a booking is made, a certain time of day comes around.
• Steps: what happens next, in order. Copy the details into a spreadsheet, send a confirmation, create a task, notify someone.
• Review: the point where a person looks before anything important goes out or gets changed.

Plain automation follows fixed rules: if this, then that. AI adds a new kind of step, one that can read and write language. It can read an incoming message and decide which category it belongs to, pull the key details out of a messy email, summarize a long thread, or draft a reply.

Finding the right first task

Good candidates share four traits:

1. It repeats often. Many times a week, not once a quarter.
2. The steps are the same each time. You could write them on an index card.
3. Mistakes are easy to catch and fix. A wrongly tagged email is a small problem. A wrong invoice is a bigger one.
4. It eats real time. Saving two minutes on a monthly task is not worth the setup.

Common examples: sorting incoming inquiries by topic, copying form details into a spreadsheet, drafting first replies to common questions, sending booking confirmations and reminders, and summarizing the week's customer feedback.

Build it in small pieces

Write the task out step by step before you touch any tool. Then automate one step at a time.

Google's guide for writing AI instructions suggests breaking complex work into smaller prompts, one instruction each, rather than packing everything into one request [1]. The same thinking applies to automations. A chain of small, clear steps is easier to test and fix than one big step that tries to do everything.

For any step that uses AI, write the instruction as carefully as you would for a person (Lesson 2 covers how), and tell it exactly what to do when something does not fit: "If the message does not match any category, label it 'Other' for a person to review."

Test before you trust

NIST, the U.S. standards agency, advises that AI systems be tested before they are put to use and checked regularly while in operation [2]. For a business owner, that means:

• Run it on past examples first. Feed it last month's emails or forms and compare its results to what you actually did.
• Watch it closely at the start. For the first week or two, check every result.
• Keep checking after it settles. Spot-check a sample each week. Tools and inputs change over time.

Keep a person at the review point

Decide in advance which outputs a person must approve. A useful rule: anything that goes to a customer, changes money, or commits your business gets a human review at first. Internal sorting and note-taking may need only spot checks once they prove reliable.

Be honest about the review. NIST describes "automation bias," the tendency to trust automated output too much [3]. If the reviewer clicks "approve" without reading, the review is not doing its job.

Common mistakes

Automating a messy process. If the steps are unclear when a person does them, automation just makes the confusion faster. Fix the process first.

Starting too big. A ten-step automation across four tools is hard to debug. Start with one or two steps.

No plan for exceptions. Every automation meets something unexpected. Decide where odd cases go: usually to a person.

Never checking again. An automation that worked in March can quietly break in June when a form or a tool changes.

Worked example

A landscaping company gets about 30 quote requests a week through its website form and by email. The owner spends close to an hour each morning reading them, copying details into a spreadsheet and replying.

Trigger: a new quote request arrives.

Step 1 (AI): read the request and pull out the name, address, service wanted (mowing, cleanup, design or other) and preferred timing.

Step 2 (rule): add those details as a new row in the quote spreadsheet.

Step 3 (AI): draft a short reply that thanks the customer, confirms the service requested and proposes two visit times from the open slots.

Review: the owner reads each draft and the new spreadsheet rows, fixes anything off, and sends.

He tests it on the previous month's requests first and finds that requests mentioning several services are tagged poorly. He adds an instruction: "If more than one service is mentioned, list all of them." After two weeks of checking every result, he moves to spot checks on the spreadsheet rows but still reads every reply before it goes out.

Checklist

• I picked one task that repeats often and follows the same steps.
• I wrote out the trigger, the steps and the review point.
• I tested it on past examples before using it live.
• Anything going to customers or affecting money is reviewed by a person.
• Exceptions go to a person, not into a guess.
• I spot-check results regularly, even after it runs smoothly.

Try it

For one week, keep a list of every task you do more than five times. At the end of the week, choose the one that is most repetitive and lowest risk. Write its trigger, steps and review point on a single page. You now have a plan ready to build.

Tips

• Write the process on paper before you build anything.
• Automate one step at a time and test each one.
• Send anything unusual to a person by default.
• Put a reminder on your calendar to check each automation monthly.

Sources

1. Google, "Prompt design strategies," Gemini API documentation, https://ai.google.dev/gemini-api/docs/prompting-strategies
2. National Institute of Standards and Technology, "Artificial Intelligence Risk Management Framework (AI RMF 1.0)," NIST AI 100-1, January 2023, https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf
3. National Institute of Standards and Technology, "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile," NIST AI 600-1, July 2024, https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf