AI is neither useless nor magic, and the difference between a business that gets value from it and one that does not is usually task selection. These tools are strong at working with text you already have and at getting a first version onto the page. They are weak wherever being nearly right is not good enough.
First drafts, not final copy
Getting from a blank page to something you can react to is where most of the time saving is. A draft advert, a policy outline, a reply to a difficult email. Editing a mediocre draft into a good one is usually faster than writing from nothing, as long as you actually do the editing.
Summarising material you supply
Paste in a long email thread, a set of meeting notes, or a report, and ask for the key points or the decisions and actions. Because the source is in front of it, the model has far less room to invent, which makes this one of the most reliable uses.
Extracting and reshaping
Pulling the dates, names, and amounts out of messy notes and turning them into a tidy table is dull work that these tools do quickly. The same goes for turning bullet points into prose, or a long document into a short brief.
Changing tone and length
Making a blunt reply more diplomatic, cutting a page to a paragraph, or rewriting something technical for a customer are all well suited to a language model, because the substance is already yours and only the wording changes.
Where it does not belong
Do not use it for figures or calculations, because it predicts what an answer looks like rather than working it out. Do not treat its output as legal, tax, or regulatory advice. And avoid using it for anything where you could not spot a wrong answer, because then you have no way to catch the mistakes.
Common mistakes
- Asking it to add up, reconcile, or work out numbers instead of using a spreadsheet.
- Taking employment, contract, or tax wording as advice rather than as a starting draft.
- Using it on a subject you know nothing about, so you cannot judge whether it is right.
- Sending a draft out unread because it looked polished on screen.
- Asking a broad question with no material supplied, then being disappointed by a generic answer.
- Spending longer correcting an answer than the task would have taken you directly.
Check yourself
0/31.Which of these tasks is the AI most likely to do well and save you real time on?
2.A member of staff asks the AI to draft a customer facing explanation of a technical problem. What is the sensible workflow?
3.Why is it a bad idea to use AI for a task in a subject you know nothing about?
This is general guidance, not a substitute for advice on your specific setup. Want a hand putting it into practice? Talk to us or see our care plans.