AI agents
Where AI agents actually save time in a small business
Five repetitive workflows worth automating first.
AI agents are often sold as a way to run a business on autopilot. In a small business, their real value is more modest and more useful: they take the repetitive reading, sorting and drafting off your team's plate, so people spend their time on the work that needs judgment.
An agent, in practice, is software that reads something (an email, a form, a document), decides what to do within limits you set, and acts: it answers, files, drafts or reminds. The best first jobs for it are not the impressive ones. They are the boring ones that happen every day.
A simple test before you automate anything
- It happens every day or every week, not twice a year.
- It follows rules you could explain to a new member of the team.
- The input is text: emails, messages, forms or documents.
- A mistake is cheap to catch before it matters.
If a task passes all four, it is a good candidate. If it fails the last one, keep a person in charge and let the agent prepare the work instead.
Five workflows worth automating first
Answering the same internal questions
How many leave days do I have left? What is the policy on sick leave? An assistant that answers from your own handbook and policies, shows where the answer comes from, and hands over to a person when it is unsure, saves HR from answering the same message all day.
Sorting the shared inbox
An agent can read each incoming email, label it, pull out the key details (who, what, by when) and route it to the right person, with a suggested reply ready for approval.
Turning documents into data
Invoices, CVs, delivery notes and forms can be read into the fields of your system automatically, with anything uncertain flagged for a person to check rather than guessed.
First drafts of routine replies and quotes
From your templates and past answers, an agent drafts the reply, the quote or the follow-up. A person edits and sends it, in a fraction of the time it takes to start from a blank page.
Chasing and reminders
Missing documents, pending approvals, contract renewals: an agent can follow up on schedule, record what it sent, and escalate to a person when nobody answers.
Where not to start
Leave decisions with weight to people: hiring, approving payments, changing salaries, anything with legal or financial consequences. An agent can gather the facts and prepare the decision, but a person should make it, and the record should show who did.
What an agent needs from you
- Real examples: a few dozen past emails, documents or questions, with what the right outcome was.
- Sources that are up to date: if the handbook is wrong, the answers will be too.
- An owner: one person who checks its work at the start and decides when it is ready for more.
- Clear limits: what it may read, what it may do on its own, and what always goes to a person.
Most of the effort in a good agent is not the AI. It is agreeing how the work should be done, which is also the part that makes your team faster even before anything is automated.
How to start in four weeks
- Week one: pick one workflow and measure how it runs today: how many items, how long each takes, how many go wrong.
- Week two: build the agent with real examples from your business, not invented ones.
- Week three: run it beside the team, compare its work with theirs, and fix what it gets wrong.
- Week four: decide, with the numbers in front of you, whether to keep it, adjust it or drop it.
Measure what matters
Count the time each item takes, how often it needs correcting, and how quickly people get an answer. If those numbers don't improve, the agent isn't helping yet, however clever it looks.
Key takeaways
- The best first jobs for an AI agent are frequent, rule-based, text-heavy and cheap to check.
- Start with one workflow, real examples and a person approving what matters.
- Judge it by time saved and errors avoided, measured against how the work runs today.
