Audit yourself before automating

Audit Yourself Before Automating

Insight from GrantBot

Automation tools like Zapier, Make, and Airtable exist so anyone can streamline their processes without the learning curve of code. Combine that with a clever prompt and you can imagine an AI-led 80/20. Sure sounds nice.

However, many teams jump too far, get upset with the output, and abandon the project for a new shiny object.

Take producing content or sending follow-up emails. Generating this text with AI is a great goal. But to get this automation running requires a matched writing style and tone. And now we have OpenAI assistants, GPTs, and classic prompt engineering. Where to begin?

The bigger productivity gain is at the atomic level: simple binary decision-making. One step up? Categorization.

A better goal is to have AI decide if data requires escalation against a set of criteria. Or categorizing an email and forwarding it to the responsible team. Both solutions free a human to execute high-leverage work.

At the end of the day, it’s deep work we need to protect. Using OpenAI to make the simplest decisions reduces our need to make a context switch. Which allows us to stay in leveraged work. Which produces more goods and services for our customers.

So when you’re thinking about your AI implementation strategy, pull a James Clear and go atomic:

1. Audit: Where are humans making simple yes/no decision based on an input. Or where is data categorized and passed to the next stakeholder?

   1. Inspiration: return/refund requests, sales rep inbox, customer segmentation

2. Build: Create a zap that triggers when this data produced. Run it through OpenAI or a ChatGPT module in Zapier or Make.

   1. Binary = prompt against set of criteria

   2. Categorization = this module

   

3. Test: The automation working on your first attempt is not a test. Give your prompt 7-10 different samples so you know it’s working as expected.

This is a more scalable approach to AI adoption. Your productivity gains start at the atomic level by reducing context switching.