AI in sales

B2B data for AI agents: what should you consider under GDPR?

Diagram linking B2B DATA by a dashed line to a GDPR shield, then branching to three boxes: PURPOSE, NEEDED DATA and HUMAN APPROVAL. The last box is highlighted with a gold border.

In brief: GDPR applies to B2B data used by an AI agent when that data relates to an identified or identifiable person. Before using it, define the purpose of processing, the lawful basis, the necessary data and each party’s responsibilities. Set separate limits on what the agent may retrieve, store and send.

Using an AI agent in sales? Explore NordicContacts’ MCP connection: your agent can retrieve company and contact data directly, but the connection does not send messages or change your CRM. You can test it with a free trial.

When is B2B data personal data?

A company’s industry, employee size band and general website do not usually identify a person on their own. A name, individual work email address, direct phone number and job title linked to a person, however, are generally personal data. An assessment generated by an agent can also be personal data if it is linked to a specific person.

A professional context does not remove GDPR obligations, and publicly available data is not automatically permitted for every purpose. Searching, scoring and sending data to an AI assistant also count as processing.

We will use a fictional example throughout this article: you sell maintenance planning software to industrial companies. The agent first identifies suitable Finnish companies, then helps find the role responsible for maintenance planning. It does not need to investigate anyone’s private life.

How do you define the purpose before searching?

Describe the purpose precisely enough to approve or reject an individual processing step. “Improving sales with AI” is not enough.

In our example, the purpose could be: “We identify the roles responsible for maintenance planning at industrial companies and assess whether our maintenance planning software is professionally relevant to the company.”

Document at least:

  • which product or service the data will help you sell
  • which types of companies and roles belong to the target audience
  • what data the agent processes and where it comes from
  • whether the agent may only suggest actions or also carry them out
  • when the data will be reviewed and deleted.

If you later want to use the data to train a model, assess that purpose separately. Data collected for sales is not freely reusable as training data.

What lawful basis can you use to process the data?

Choose and document an appropriate lawful basis before using personal data. Legitimate interests may apply to certain processing activities in B2B sales, but a sales purpose does not automatically make it appropriate.

Assess the interest you are pursuing, whether the processing is necessary, and its effects on the individual’s rights and freedoms. Contact related to someone’s professional responsibilities differs from profiling that combines work data with their behaviour outside work.

The lawful basis for processing under GDPR and the rules governing a communication channel are separate questions. Electronic direct marketing is also subject to national rules and depends on the nature of the recipient and the address. The rules in Finland, Sweden and Estonia should not be treated as a single permission to message every work email address.

Assess data collection, the contact channel and the recipient’s role separately. This is a practical checklist, not case-specific legal advice; if anything is unclear, confirm your approach with a data protection officer or lawyer.

How do you narrow down the data you need, from companies to contacts?

Start with companies. Once you know which companies are a good fit, you need to process personal data for a smaller group. Do not collect contact details and look for a purpose afterwards.

The following figures illustrate a workflow. They are not database results or a sales forecast.

StageSelection criteriaCompanies remaining
Initial company groupSelected industrial target audience40 companies
Needs assessmentMaintenance planning is relevant to the company’s operations16 companies
Preparing for contactThere is a timely, professionally relevant topic to discuss8 companies

Only then do you select contacts. If you need two roles at each of the eight companies, you will process 16 contacts. Justify the need: for example, the person responsible for maintenance and the person responsible for technical compatibility.

To begin with, a name, company, job title, necessary work contact details, source and date checked are usually enough. The agent does not need the entire CRM history to draft a message. Nor does a job title alone prove someone’s decision-making authority: confirm their role in conversation.

Who is responsible for processing by the agent and service providers?

Your company is usually the controller when it determines the purposes and essential means of processing personal data for sales. An external service’s role depends on what it actually does, not just the label used in a contract.

If a service provider processes data on your behalf and under your instructions, you need a GDPR-compliant data processing agreement. If the provider also uses the data for its own purposes, assess that processing and the provider’s role separately.

Map the data flow: data source → agent application → model service → CRM. Check storage, access rights, subprocessors, deletion, any transfers outside the EEA, and whether data is used to train models.

NordicContacts provides B2B company and contact search for Finland, Sweden and Estonia, along with REST API and MCP access. The MCP agent connection requires you to log in and have a trial or subscription, and each user needs their own account. MCP only returns data: it does not send messages or change your CRM. External actions remain subject to your own application’s controls and human approval. Every delivered contact has a work email address, a phone number, or both; you can require email, phone, either, or both. Exportable email addresses have been checked within the past 91 days.

What can the agent do without human approval?

Separate reading data, drawing inferences and taking external actions. A limited search carries a different level of risk from sending a message or overwriting CRM data.

Agent actionKey data protection questionPractical control
Shortlisting companiesIs personal data needed at this stage?Use company-level data first
Identifying a professional roleIs the information accurate and up to date?Record the source and check anything unclear
Drafting a messageDoes the input contain unnecessary data?Provide only fields relevant to the message
Sending a messageIs the contact permitted and appropriate?Require human approval initially
Updating dataCould an error spread to other systems?Restrict write permissions and retain a change history

The first sales question in our example could be: “Are you responsible for maintenance planning, and would making it easier to anticipate maintenance shutdowns be useful?” The agent should not state as fact that the company has maintenance problems unless there is a proper basis for that claim.

Not all automated sales prioritisation falls under GDPR’s specific rule on automated individual decision-making. Assess this separately if a decision is based solely on automated processing and produces legal or similarly significant effects on a person. Also check whether the processing requires a data protection impact assessment, particularly if it is likely to pose a high risk.

How do you ensure transparency and respect the right to object?

Build privacy information into the workflow. If you obtain data from somewhere other than the individual, assess GDPR’s information requirements, what you need to disclose and the applicable deadlines. Explain the purpose of processing, lawful basis, data source, retention principles and how people can exercise their rights.

People have the right to object to the processing of their personal data for direct marketing. When someone objects, processing for that purpose must stop. Use a minimal suppression record and access controls to ensure the agent does not put the person back on a contact list. A suppression record does not require you to retain all previously collected data.

How do you start in a controlled way and manage data retention?

Start with a limited group of companies and keep external actions subject to human approval. Before expanding, check company fit, data accuracy, the reasons for each message, and whether objections are handled correctly.

Define review and deletion criteria for each workflow: search results, contact lists and customer records do not automatically need the same retention period. Test how you handle job changes, corrections to inaccurate data, rights requests and a change of service provider. Weekly monitoring of target companies, for example, can help you spot changes in employers and roles promptly.

Changes must reach all relevant destinations. Deleting a CRM record is not enough if the data remains in an export file, the agent’s memory or the model service’s logs. Account for these when planning your CRM import.

Try narrowing down your own target audience: describe your target audience and see the number of matching companies and contacts without an account. The free trial includes a one-time download of up to 50 contacts and lets you test the MCP connection in your own agent.