Before we start: what “automation” actually means here
When most people hear “automation,” they picture something large, complex, and expensive: a software project with requirements documents and a go-live date three months away. That is one kind of automation. It is not the kind we are talking about today.
The kind we are talking about is a connected sequence of steps that runs without you touching it. Something happens: a form gets submitted, an email arrives, a file gets uploaded, a time of day is reached, and a chain of actions follows automatically. An AI step reads the input, makes a judgment, drafts a response, routes to the right person, or writes a summary. A human reviews anything that matters before it goes anywhere.
The tools that make this possible for non-developers are called no-code automation platforms. The two most widely used are Zapier and Make (formerly Integromat). Both have free tiers generous enough to build and test everything in this post. Both connect to hundreds of business apps. Both let you add an AI step, powered by ChatGPT or Claude, that reads, writes, classifies, or summarizes at any point in the workflow.
You do not need to choose between them today. Pick one, make a free account, and follow along. The concepts transfer directly.
How to pick the right workflow to start with
Before building anything, the question worth asking is which task in your week most deserves to disappear. The answer is almost always the one that meets all three of these tests:
- It happens repeatedly. The same trigger, the same steps, the same output, over and over. Weekly reports. Every inbound form submission. Every new customer. If it only happens occasionally, the automation does not pay off fast enough to be worth building first.
- The failure mode is recoverable. A miscategorized customer inquiry that goes to the wrong inbox is recoverable. A contract sent to the wrong person is not. Start with low-stakes workflows while you are learning. Save the high-stakes ones for after you have built a few and understand how they break.
- You can name the person it gives time back to. Abstract time savings are politically weak and motivationally weak. Giving three hours a week back to your office manager, specifically, is concrete. That specificity is what gets the automation actually used and maintained.
With that in mind, here are four workflows that consistently pass all three tests for small and mid-sized businesses, with enough detail to build each one today.
Workflow 1: Contact form → qualified lead summary in your inbox
The problem it solves: Someone fills out your contact form. The raw submission lands in someone’s inbox. That person reads it, decides whether it is worth pursuing, copies relevant details somewhere, and either responds or does not. If they are busy, the response is slow. If they are on vacation, it waits. The whole process is manual and inconsistent.
What the automation does:
- Trigger: new form submission (works with Typeform, Gravity Forms, Jotform, Google Forms, or your website’s native form)
- AI step: reads the submission and produces a three-sentence summary: who they are, what they need, and a qualification signal (does this look like a real opportunity or not)
- Action: sends the summary to the right person’s email or Slack, with the original submission attached
- Optional: adds the contact to your CRM automatically
The AI prompt that drives it:
Read this contact form submission and write a three-sentence
summary for our sales team:
1. Who is this person and what company or context are they from?
2. What are they asking for or trying to accomplish?
3. Does this look like a qualified lead for a [describe your
business]? Flag any signals — positive or negative — in
one sentence.
Keep it factual. Do not invent details not present in the
submission.
Submission: {{form content}}Time to build: 45–60 minutes on Zapier or Make, including testing. The AI step is one additional action block with the prompt above pasted in.
Time it saves: 5–10 minutes per submission, every submission, forever. For a business receiving twenty inquiries a week, that is two to three hours returned to whoever was doing this manually.
Workflow 2: Meeting recording → summary and action items in your notes
The problem it solves: Meetings happen. Recordings sit in Zoom or Teams, unwatched. The notes nobody took mean the decisions nobody documented get re-litigated three weeks later. The action items nobody wrote down do not happen. Institutional memory lives entirely in the heads of the people who were in the room, until they are not.
What the automation does:
- Trigger: meeting ends in Zoom, Teams, or Google Meet
- Step: transcript is generated (Zoom and Teams do this natively; for others, a tool like Otter.ai or Fireflies.ai handles it)
- AI step: reads the transcript and produces a structured summary
- Action: saves the summary to Notion, Google Drive, OneNote, or wherever your team stores reference documents
- Optional: posts a brief recap to the relevant Slack or Teams channel
The AI prompt that drives it:
Read this meeting transcript and produce a structured summary
with four sections:
DECISIONS MADE: List any decisions that were reached.
ACTION ITEMS: List tasks with the owner's name and any
deadline mentioned. If no owner was named, write "unassigned."
OPEN QUESTIONS: List anything raised but not resolved.
KEY CONTEXT: One short paragraph summarizing what the meeting
was about and any important background for someone who
was not present.
If a section has nothing to report, write "None noted."
Be specific. Do not pad.
Transcript: {{transcript}}Time to build: 60–90 minutes depending on which meeting and notes tools you are connecting. If your team uses Zoom and Notion, both have direct Zapier integrations and the connection is straightforward.
Time it saves: Variable, but the compounding value is the point. By month six, your team has a searchable archive of every meeting summary, every decision, and every commitment anyone made. That is worth more than the hours saved on note-taking.
Workflow 3: Customer support email → drafted reply ready for review
The problem it solves: Customer emails pile up. Someone has to read each one, decide what kind of issue it is, and write a reply. For a business handling fifty support emails a week, that is a meaningful block of time every day, and the quality of the reply depends entirely on who gets to it and when.
What the automation does:
- Trigger: new email arrives in a designated support inbox
- AI step: reads the email, classifies the issue type, and drafts a reply
- Action: creates a draft in Gmail or Outlook, or adds a note to your helpdesk ticket, flagged for human review before sending
- Nothing goes out automatically. A human reviews and sends every reply.
The AI prompt that drives it:
You are a customer support representative for [COMPANY NAME],
a [describe your business].
Read this customer email and do two things:
1. CLASSIFICATION: Identify the issue type in one phrase
(e.g., "billing question," "order status," "product complaint,"
"refund request," "general inquiry").
2. DRAFT REPLY: Write a professional, warm reply that:
- Acknowledges what the customer said specifically
- Addresses their question or issue directly
- States any next steps clearly
- Does not make commitments about timelines or outcomes
unless you are certain they are accurate
- Stays under 150 words
If the email contains a complaint or frustration, lead with
genuine acknowledgment before moving to resolution.
Customer email: {{email body}}Time to build: 60–75 minutes. Gmail and Outlook both connect to Zapier directly. The AI step is the same as the others, one additional block with the prompt pasted in.
Time it saves: For a team handling fifty emails a week, drafting time typically drops by 60–70%. The human reviewer goes from writing to editing, which is faster and less draining. Response times improve because the draft is already there when someone sits down to work the queue.
Workflow 4: Weekly data → automated status report draft
The problem it solves: Every Friday, someone, often the most senior person who can least afford to spend three hours on it, pulls numbers from a few different places, writes a summary, and sends it up the chain. Same format. Same sources. Same exercise, every single week, indefinitely.
What the automation does:
- Trigger: scheduled (Friday at 9am, or whatever cadence you use)
- Steps: pulls the week’s key numbers from whatever source they live in: a Google Sheet, an Airtable base, a report exported from your CRM, a simple form your team fills out
- AI step: reads the numbers and writes the commentary: what changed, what the trend is, what warrants attention
- Action: emails the draft to the person who normally sends the report, ready to review, adjust, and forward
The AI prompt that drives it:
You are writing a weekly status report for the leadership team
of [COMPANY NAME].
Here is this week's data:
{{current week data}}
Here is last week's data for comparison:
{{prior week data}}
Write a status report with three sections:
HIGHLIGHTS: Two or three things that went well or improved.
WATCH ITEMS: One or two things that declined or need attention,
stated factually without alarm.
NEXT WEEK: One sentence on the primary focus or goal for the
coming week, based on the data above.
Keep the tone professional and direct. Total length: under
200 words. Do not invent context not present in the data.Time to build: 90 minutes to two hours, depending on where your data lives. If it is already in a Google Sheet updated weekly, this is straightforward. If data needs to be pulled from multiple systems first, that step adds complexity, but it is still buildable in an afternoon by someone comfortable with Zapier’s multi-step workflows.
Time it saves: Two to three hours every week for whoever was writing this manually. For a senior person, that is some of the most expensive time in the building being spent on mechanical work. Automation changes that permanently.
The thing that trips most people up the first time
The most common failure mode in a first automation is not the AI step. It is the trigger. If your trigger fires on the wrong condition, or the data it passes to the AI step is formatted differently than expected, the whole chain produces garbage, and it can take a few test runs to figure out where the problem is.
The fix is to test with real data before you turn anything on for production use. Zapier and Make both have testing modes that let you run a workflow on a real recent example without actually triggering the output actions. Use them. Run at least three real examples through before you declare it working. Look at the AI output each time and ask whether you would be comfortable with that response going to a real person or landing in a real document.
And keep the human review step in anything that touches the outside world. Internal summaries can run more freely, if the meeting summary has an error, someone will catch it. Customer-facing output needs eyes before it goes anywhere. That checkpoint is not overhead. It is what makes the automation trustworthy enough to actually use.
The best automation is the one that runs so quietly you stop thinking about it. You only notice it when it is not there, when you are on vacation and you realize someone is doing manually what your workflow usually handles, and it is taking them three times as long.
Where to go from here
Pick one of the four workflows above, the one that addresses the task you most dread this week, and build it today. Give yourself two hours. If you get stuck on a specific step, Zapier and Make both have documentation and community forums that cover almost every integration question you will run into. The learning curve is real but short, and it pays off across every workflow you build after the first one.
Once the first one runs reliably, look at what is next. The pattern: trigger, AI step, human checkpoint, action, applies to an enormous range of business tasks. Once you have built it once, you see it everywhere.
What is coming in June
Next month we are going somewhere that generates more questions than almost any other topic in this space: the dark side of AI. Deepfakes, social engineering, and how to protect your team when the attacks are getting harder to see. If you manage people or run a business, June’s post belongs on your reading list.
This is post nine of a two-year series on AI for real people doing real work. Post one covers what AI actually is. Post two is how I use these tools day to day. Post three covers the five free tools worth trying first. Post four tackles email. Post five covers the AI and security landscape. Post six is prompting without the jargon. Post seven covers phishing awareness on a budget. Post eight is for leaders who just want to know what is relevant to them. Got a workflow you want help automating? Send a note – this is exactly what I build.
