The strongest automation opportunities usually begin with a small, familiar frustration rather than a sweeping technology initiative. Here is how to find them, and how to build automations your team can trust.
Start with a familiar frustration
Look for work that is frequent, predictable, and disliked: copying form submissions into a CRM, chasing invoice approvals, sending the same follow-up email for the tenth time this week. These tasks are easy to describe, which makes them far easier to automate reliably.
Ask your team where their time goes. The people doing the work usually know exactly which tasks are repetitive, and they're often the first to suggest how they could be simplified.
Map the work first
Before choosing a tool, write the process down as it really happens, not as the policy document says it should. Note where time is lost to waiting or re-typing, and define what success would look like.
- What triggers the task, and how often does it happen?
- What information is needed, and where does it come from?
- Who is involved, and where are the handoffs?
- What are the exceptions, and how are they handled today?
- What would “better” look like: hours saved, fewer errors, faster responses?
Choose the simplest tool that works
Many workflows don't need AI at all. A rules-based automation in a tool such as Zapier or n8n is cheaper, faster, and easier to debug: when a form is submitted, create a contact, notify the right person, and log the request.
Reach for a language model when the task genuinely involves reading, drafting, summarising, or classifying unstructured text, such as sorting incoming emails by topic or drafting a first reply for a person to review.
Keep a person in the loop
For workflows that touch sensitive information or consequential decisions, define review points and a clear fallback when something looks wrong. An automation that drafts a reply is very different from one that sends it.
Limit each automation's access to the data it needs, and avoid sending personal information into tools your business hasn't approved. Good automation design is as much about what a workflow can't do as what it can.
Start small and observe
Pilot one bounded workflow on representative inputs, including the awkward ones. Log failures, review the quality of outputs, and compare the effort saved with the cost of building and maintaining it.
Expand only when the pilot has demonstrated that it's useful. A small automation that works every day builds far more confidence than an ambitious one that needs constant attention.
Plan for upkeep
Automations depend on the tools around them. APIs change, forms gain new fields, and edge cases appear months after launch. Give every automation an owner, simple documentation, and alerts, so failures are noticed in minutes rather than weeks.
Measure what changed
Revisit the success measures you defined at the start. Is the task faster? Are there fewer errors? Do people trust the output? Honest measurement keeps automation focused on outcomes, and tells you which opportunity to tackle next.