How to pick your first process to automate with AI

Start with a tiny slice, not the whole process

Your first automation should be a small, clearly bounded slice of work. Think about one repeatable step in a longer process, for example drafting a reply, renaming files or filling a template. A tiny scope reduces technical risk, keeps the number of tools low and makes testing easier. You can still create big value if the slice repeats every day. Ask yourself where you spend time on the same micro task again and again. Aim for something that fits on a sticky note: subject, inputs, one clear outcome.

Use a simple impact vs effort filter

To prioritise candidates, compare impact and effort side by side. Impact means time saved every week, fewer errors or faster response for others. Effort means data cleanup, tool setup and how fuzzy the decisions are. For a first project, you want high impact and low to medium effort. List three to five candidate tasks, then roughly score each on impact and effort from 1 to 5. The best starting point usually has the highest impact score and the lowest effort score, even if it does not feel glamorous.

Favour structured, rule like work

AI struggles with vague, high stakes decisions. For a first workflow, look for tasks that already follow simple rules. Good signs are checklists, templates, standard phrases or if a junior colleague can do the job after a short explanation. Typical examples are categorising messages, extracting fields from text or drafting standard replies. The more you can describe the task in if this, then that language, the easier it is to automate. This lets you learn tooling and prompts without constant surprises.

Avoid hidden traps in your first candidate

Some tasks look perfect but slow you down later. Be cautious with anything that touches sensitive data, needs five different logins, or has unclear ownership. Also be wary of tasks where people “just know” what good looks like but no one can explain it. Legal approval, hiring decisions or custom creative work are poor first candidates. These tasks add stress and require complex safeguards. For your first win, aim for a task where a slightly imperfect result is still safe and easy to fix manually.

Define a clear success test before you build

A first automation feels manageable when you know exactly how you will judge success. Before opening any tool, write down one or two measurable checks. For example, “The workflow drafts replies that need less than 20 percent editing” or “The bot correctly tags 8 out of 10 emails.” Collect 10 to 20 real examples and use them as a test set. Your first target does not have to be perfect, it just has to be clearly better than today and reliable enough that you trust it.

Useful links

• Zapier — Popular no code automation platform to connect apps.
• Make — Visual automation builder for multi step workflows.

Keep learning