A business AI needs the same as a good new colleague: context, real data, clear instructions and a person to review and correct it. The difference isn't made by the tool — it's made by how you guide and supervise it.
How it works (and how it doesn't)
It does not do this
It cannot read your mind, it does not know your intention unless you provide it, and it must not decide on its own about things that commit you (who to write to, what to state, what to send).
Yes, it does this
It carries out what you tell it with the data you give it. The more specific the brief and the more real the data, the more accurate and useful the result. And it always improves if you correct it.
7 common mistakes and how to avoid them
1. Empty brief
Real example: "send emails"
Say who to, in what tone, what you offer and what you want the contact to do (CTA). E.g.: "Brief, friendly email to [contacts] presenting [service], asking for a call."
2. Blank cheque
Real example: "sell [product] to any type of customer, with no limit at all"
Always scope it: sector, size, country, and a contact limit. "No limit" makes it act on people it shouldn't. You set the boundaries.
3. Out of scope
It is the team's sales tool. Use it for prospecting, contacts, trade fairs and messages — not for leisure tasks unrelated to work. Keep the focus where it adds value.
4. Scraping and enrichment "all of it, in one go"
An AI does not work for hours in a loop extracting or enriching thousands of records in one go — it does it in batches, and a giant task gets cut off halfway or comes out incomplete. Ask for limited batches (e.g. 50-100), review each one and ask for the next. It takes slightly more effort to specify, and it is far more reliable to get.
5. Feedback without guidelines
Say WHAT to change and in which direction: "shorter", "less sales pitch and more interest in their brand", "more direct tone". Specific corrections are how you train it.
6. Not checking what it did
Confirm before assuming something has been sent, and give it your context (who you are, from which email address). Always review the result before it goes out.
7. Delegating sensitive actions blindly
Handle sends (email/social) and scheduled ones with human review: generate first, review, and then authorise. Don't ask it to "send" without having seen what is being sent.
The principle that sums it all up
AI amplifies, it does not replace. It saves you hours of searching, writing and organising — but the judgement, real data and final decision are yours. Work with it by giving it good material, explaining why, correcting it when it fails and reviewing before sending.
What never to delegate blindly
- Bulk sends without reviewing the content before it goes out.
- Stating company data (figures, clients, awards) that you haven't given it or that are not in its database.
- Actions in your name (social media, email from your address) without confirming beforehand what it will do.
Ten rules for a good brief
- Who you are and who you represent, in one line.
- Clear objective and the action you expect from the contact (CTA).
- Real data to use — and ask it not to use any other.
- Tone and phrases to avoid ("don't start with 'Hi, I'm…'").
- Limited batches and specific criteria, not "everything / no limit".
- Review and correct with specific instructions: that is how it is fine-tuned.
- Check before sending — you give the final authorisation.
In short: AI is an accelerator, not an autopilot. The team that guides it well, gives it real data and reviews its work gets professional, consistent results.
- IA
- Best practices
- Prompts