AI
Three tasks where AI saves time from the first week, and two where you're better off waiting.
Start with the boring work
AI is talked about as a revolution. In a company, though, it helps most where the work is most boring. A good first candidate meets three conditions:
- It repeats daily or weekly, so the savings add up.
- It has a clear input and output: an invoice goes in, filled-in fields come out.
- A mistake is easy to check. A person can see at a glance whether AI got the amount right.
1. Reading invoices and receipts
If a task meets all three, you have a good starting point. The most common one is incoming invoices: someone opens them, finds the supplier, amount, due date and account number, and types them into the system. AI pulls those details out of the document by itself, and a person just checks the result. Where AI isn't sure, the system flags it.
2. Sorting incoming emails and enquiries
A shared inbox that collects enquiries, complaints and order-status questions is a classic bottleneck. AI can read an email, sort it by type and assign it to the right person, with a short summary of what it's about. Nobody has to watch the inbox all day any more.
3. Summaries and draft replies
A long complaint thread, a thirty-page contract, meeting notes. AI turns them into a summary of a few sentences and drafts a reply that only needs checking and editing. The decision stays with a person, but writing from scratch goes away.
Where not to use AI yet
- High-stakes decisions without human review. Approving payments, legal assessments, or replies that go out to customers unchecked. AI can get things wrong, and in tasks like these a mistake is expensive.
- Tasks that happen a few times a year. Setting it up, configuring it and checking it would cost more time than AI would ever save.
How to tell it's working
Don't start with a big project. Try a two-to-four-week pilot on a single task and measure before and after: how long the task takes, how many results a person has to correct, and how many mistakes reach the customer.
If the numbers after the pilot aren't clear, you don't need AI for that task. That's a useful result too.
What about the data
First, check where the data goes. For business use, choose business versions of AI services, which under their terms don't train on your data. For sensitive data, consider whether it should leave the company at all, and keep an eye on who can access the results.
AI pays off in a company when it saves boring work that can be checked. Start there and the rest will follow.