Not everything should be automated. Here's a practical framework for choosing your first AI automation target, plus the tasks you should always keep manual.
Everyone wants to automate everything. Nobody wants to think about what should stay manual.
After 25 years in ICT and building automation systems for two businesses, here is what I have learned: the biggest gains come from automating the right things, and the biggest disasters come from automating the wrong ones.
I use three criteria to evaluate whether a task is worth automating.
Frequency. Does this happen daily? Weekly? If you do it once a quarter, the ROI on automation is near zero. You will spend more time maintaining the automation than you save.
Structure. Is the task predictable? Does it follow a clear pattern with defined inputs and outputs? A structured task like “send a follow-up email 48 hours after a meeting” is perfect. An unstructured task like “figure out why the client is unhappy” is not.
Volume. How much time does this task consume in total? A 5-minute task that happens 20 times a week is 100 minutes. That is worth automating. A 2-hour task that happens once a month is 2 hours. Probably not worth the setup.
Score each task on all three. High frequency, high structure, high volume? Automate it yesterday.
The exact architecture, memory layers, and delegation patterns I use to run 50 agents across two businesses.
Get the AI Agent Blueprint →Based on what I have seen across my own businesses and clients at Allwebzone, these five tasks almost always deliver immediate ROI.
1. Email triage and drafting. Not sending emails automatically. Drafting them. Your AI reads the incoming message, writes a proposed reply, and saves it as a draft. You review, tweak, send. Saves 30 to 60 minutes daily for anyone who gets more than 20 emails.
2. Meeting notes and follow-ups. Record the call, transcribe it, extract action items, and draft follow-up emails to attendees. This one feels like magic the first time you see it work.
3. Content repurposing. Write one piece of content. Let an agent turn it into a LinkedIn post, an email newsletter section, three tweet-sized points, and a summary for your website. What took a content person half a day now takes minutes.
4. CRM updates. After every client interaction, an agent logs the update, tags the contact, and sets the next follow-up date. Your CRM stays current without anyone manually entering data.
5. Report generation. Weekly dashboards, project status updates, financial summaries. The agent pulls data from your tools, formats it, and delivers it every Monday morning before you open your laptop.
This is the part people skip, and it costs them.
Sales calls. The actual conversation. AI can prepare your notes, research the prospect, and draft your proposal afterward. But the call itself? Keep that human. People buy from people, especially at higher price points.
Apologies and sensitive communications. If a client is upset, if something went wrong, if the message carries emotional weight, write it yourself. An AI-drafted apology feels exactly like what it is.
Strategic decisions. AI can present the data. It can show you three options with tradeoffs. But the final call on pricing, positioning, partnerships, hiring? That is your job. Outsourcing your judgment to a language model is not automation. It is abdication.
Creative direction. AI is great at producing content within established guidelines. It is terrible at deciding what those guidelines should be. Your brand voice, your visual identity, your market positioning: these require human instinct.
Anything involving money movement without review. Generating invoices? Fine. Sending them automatically without a human checking the numbers? Dangerous. Any automation that spends money or sends money should have a human approval step.
Before you commit to building a full automation, run this test.
For 48 hours, do the task manually but log every step. Write down exactly what you do, in what order, with what inputs and outputs. Be specific. Not “I check the email.” Instead: “I open Gmail, filter for unread from the ‘partnerships’ label, read the subject and first paragraph, decide if it needs a response today, and if so, draft a reply using a friendly but professional tone.”
If you can write that procedure clearly enough that a new employee could follow it with no training, an AI agent can do it. If the procedure is full of “use your judgment” and “it depends,” it is probably not ready for automation.
I know I listed five. Build one.
Seriously. Pick the one that hurts the most. Get it working. Run it for two weeks. Fix the edge cases. Then, and only then, start the second.
The temptation to automate everything at once is strong. I have watched multiple clients try it. They end up with five half-working automations instead of one solid one.
Compound progress beats parallel chaos every time.
Automation is not about replacing humans. It is about removing the parts of work that humans are not good at. We are not good at remembering to follow up in exactly 48 hours. We are not good at formatting the same report the same way every single week. We are not good at consistency at volume.
We are excellent at judgment, creativity, empathy, and strategy. Automation gives you more time for those things.
If you want a structured approach to figuring out what to automate in your specific business, the blueprint covers this with a scoring framework and priority matrix.
Subscribe to the newsletter for weekly automation breakdowns and the real implementation details.
Subscribe to The Signal
One issue per week. The biggest AI build of the week, the top five videos worth your time, and an operator's read on what actually matters. For builders who ship, not dreamers who scroll.
No spam. Unsubscribe anytime. Just The Signal.