Automation pays off where the same judgement is made hundreds of times a month by people who'd rather be doing something else. Combined with AI, that now includes work that used to need a human to read, not just click — emails, PDFs, photos, free-text notes.

Here's how to find good candidates in your own business, five that show up almost everywhere, and how to roll them out without creating new problems.

How to spot a candidate

  • It happens often — daily or weekly, not once a quarter.
  • The input arrives in a predictable place: an inbox, a form, a folder, a system.
  • A good result is easy to recognise when you see it.
  • Today, someone copies information from one place into another.
  • Mistakes are annoying but recoverable — at least at first.

A quick way to find them: ask each team what they do every Monday morning, and what they'd happily never do again.

1. Inbox triage

Shared inboxes collect questions, orders, complaints, invoices and spam in one pile, and someone spends the first hour of every day sorting it. A system can read each message, label it, pull out the order number or account, check it against your records, and route it to the right person with a suggested reply — so humans start from a draft instead of a blank page.

Watch out for: messages that fit no category. Give the system a clear “I'm not sure” route to a person, rather than forcing a guess.

2. Document data entry

Invoices, delivery notes, purchase orders, application forms: the data is all there, just trapped in a PDF or a photo. Extraction turns each document into structured fields, and validation checks them against what you already know — does this supplier exist, does the total add up, does the purchase order match?

The result is that hours of typing become a quick review of anything flagged as unusual. Watch out for: silent errors. Every extracted document should keep a link to the original so a person can check it in one click.

3. Recurring reporting

If someone spends Monday morning pulling numbers from three systems into a spreadsheet, that report can assemble itself. Connect the sources once, calculate the figures the same way every time, and add a short written summary of what changed since last week and why it might matter.

Watch out for: definitions. Agree exactly what “active customer” or “revenue” means before automating it, or the report will be consistently wrong instead of occasionally wrong.

4. Customer onboarding

Welcome emails, account setup, document collection, checks and reminders follow the same steps for every new customer — and steps get missed when the team is busy. Automating the sequence keeps it consistent, nudges customers who stall, and shows your team exactly who needs a conversation instead of an email.

Watch out for: sounding robotic. Automated messages should read like they came from a helpful person, and replies should reach a real one.

5. Internal questions

“Where's the latest price list?” “What's our policy on…?” “Who handles…?” A searchable assistant over your own documents answers these instantly, with links to the source — and quietly reveals which documents are missing, duplicated or out of date, because those are the questions it can't answer well.

Watch out for: access. The assistant should only search what the person asking is allowed to read.

Rolling it out

  • Measure the current process first — how long it takes and how often it goes wrong — so you can tell whether automation helped.
  • Start in “suggest” mode: the system drafts, a person approves. Move to automatic only where the suggestions are consistently right.
  • Keep a log of every automated decision and make it easy to reverse one.
  • Give one person ownership, so problems are noticed and the system keeps improving.