The practical answer
Automation is a workflow question first
Before AI, automation meant rules: if this happens, do that. AI adds a new ability, handling messy inputs like emails, documents, and conversations that never fit clean rules. The best automations combine both. A rule moves the work, AI reads or drafts or classifies, and a person decides where it matters. Start by writing down how work actually flows through your business today, step by step, including the handoffs nobody documents.
Which workflows to automate first
Look for four traits. The more of them a workflow has, the better the candidate.
- Volume: it happens many times a week, so the savings add up.
- Clear inputs and outputs: an inquiry comes in, a reply and a record go out.
- Low stakes per instance: a mistake is fixable, not catastrophic.
- Painful to a person: nobody will miss doing it by hand.
Typical first candidates: inbound inquiry triage, lead follow-up, meeting notes and action items, invoice reminders, report drafting, onboarding paperwork, and content repurposing.
Build with human checkpoints
- Define the trigger: what starts the workflow (a form, an email, a calendar event, a new row).
- Define the AI step precisely: classify, extract, summarize, draft, or recommend. One job per step.
- Define the checkpoint: who approves, what they see, and how they fix errors.
- Define the record: where the result lands, so nothing lives only in a chat window.
- Run it in shadow mode for two weeks, comparing the AI output with what a person would have done.
Tools, in order of preference
Use the automation and AI features inside software you already run. Then use a no-code automation platform to connect tools, with AI steps in the middle. Only build custom software when the workflow is proven, high-value, and stable. Custom builds are where small businesses lose months.
Measure, document, expand
Track three numbers for each automation: hours saved, error rate, and how often a person had to step in. Write a one-page runbook for each so someone else can maintain it. Expand only when the last one has run cleanly for a month. Automation compounds when each piece is boring and reliable.
Where people get stuck
- Starting with the most complex workflow because it would be the most impressive.
- Removing the human checkpoint too early, then discovering the errors in front of customers.
- Chaining many AI steps together, where small errors multiply.
- Building on tools nobody else in the business understands.
- Automating a broken process, which just produces the wrong output faster.
I would spend an afternoon writing down every recurring workflow in the business on one page, with a rough count of how often each one happens. Then I would pick the highest-volume, lowest-risk one and rebuild it with a single AI step and a single approval.
I would not touch a second workflow until the first had run for a month without surprises.
A recommendation from Maasha Kah, not a guarantee. Your situation decides the order.