AI in sales: automate the work, not the trust
Everyone is putting AI agents into sales. They read emails, update the CRM, transcribe calls. Where does that leave personality, security, confidential data, GDPR and the other side’s consent?

You know the feeling. There is a problem with your order, you open the chat on the website and a friendly assistant appears. Five questions later you get a link to the FAQ you have already read. I have never left that kind of support satisfied. I get general information, but nobody actually helps me.
The same thing is happening in sales today, only on a bigger scale.
Everything gets automated
Wherever I look, sales is being automated. AI agents read emails, log tasks, fill the CRM, send follow-ups on their own and turn every call into a transcript and a summary. In presentations it looks great: the salesperson no longer has to write anything and the system sells by itself.
I am not against AI. I use it every day for preparation, research, structuring texts and translation. But after more than 20 years in sales and cybersecurity, I ask five questions about every such deployment, questions that are usually missing from the slides.
1. Where did the personality go?
B2B sales runs on trust. A company does not choose a supplier because they sent more emails, but because it trusts them. A generic message generated for hundreds of recipients is recognisable at first glance. And in DACH, where business is done slowly, personally and for years, such a message is more likely to close a door than to open one.
If an agent writes for you, the customer is talking to an agent. Not to you.
2. Where does the confidential data go?
An agent that reads your mailbox and CRM sees everything: prices, proposals, contracts, NDAs, internal notes about customers. Where is that data processed? Inside the EU or outside it? Does the provider use it to train the model? Who on their side has access? Do you have a data processing agreement with them?
If you cannot answer these questions, you are sending trade secrets somewhere you have no control over them.
3. Who else can give the agent orders?
This question comes from my security background. An AI agent that reads incoming emails and can also act, sending messages, changing records or opening attachments, is a new entry point into the company. An attacker does not need to break anything. Sending an email with a hidden instruction that the agent obeys is enough. This is called prompt injection and today it is one of the biggest risks of AI-based applications.
The more permissions the agent has, the bigger the damage when it happens.
4. Did anyone ask the other side?
A call has two sides. If you record it and have it transcribed, you are processing the personal data of someone who may not know about it. In Germany, recording a non-public conversation without consent is a criminal offence under Section 201 of the Criminal Code (StGB). That applies to AI transcription too, because most tools record the audio first.
The same goes for third parties. A customer’s email often contains the names, phone numbers and signatures of colleagues. When the agent sends it to an external service, you are processing their data as well. GDPR requires a legal basis, informing the people concerned and a contract with the processor.
5. Does the customer know they are talking to AI?
Since 2 August 2026, the transparency obligation of the EU AI Act (Article 50) applies. If a person interacts with an AI system, they must be told, unless it is obvious from the circumstances. A chatbot pretending to be a salesperson is therefore not just a question of good manners, but of the rules.
How I do it
I use AI where it saves time and takes away nobody’s privacy or trust:
- in the background, not as the voice: company research, meeting preparation, structuring proposals, translation,
- a person sends every message: no automated sequences in a client’s name,
- confidential data only where there is a contract: and only as much as is really needed,
- consent at the start of the call: if anything is recorded, the other side knows,
- as few permissions as possible: an agent that can only read cannot break anything.
Five questions before putting an AI agent into sales
- Where is the data processed and who has access to it?
- Does the provider train on our data?
- What can the agent do without a human?
- Do all the parties concerned know about the recording and processing?
- Will the customer recognise that they are dealing with AI?
If you cannot answer one of them, wait with the rollout.
In closing
Automate the work, not the relationship. Technology will save you hours on preparation and admin. But deals are still closed by people with people. And at a time when everyone sends the same thing, a personal, human approach is the biggest differentiator you have.
This article reflects my view from practice and is not legal advice.