How to handle requests to delete an account with an AI agent

Deletion cannot be undone, so the agent never deletes. It verifies the holder, works out whether they want the account closed, their data erased, or simply fewer messages, logs the request with its date, and explains what will be deleted and what the firm must keep. The privacy team approves the deletion, and a controlled system process carries it out.

What is an account deletion request, and who owns it?

Customers ask firms to "delete my account" for many reasons: they no longer use the service, they are worried about their data, they were unhappy, or they want to stop marketing. In data protection terms, a request to erase personal data is a right people hold in many markets, including under GDPR, but it is not absolute. Firms may have to keep certain records by law, for disputes, or to prevent fraud.

The data protection officer or privacy team owns erasure requests and decides what is kept. Operations owns account closure. Engineering owns the deletion process itself. Customer support owns the conversation. The agent is the intake, and the case for keeping it there is made in can an AI agent delete a customer's account? and in our guide to the AI support agent that knows when to say no.

What does the agent prepare, and what does a person approve?

1. Agent, depth 1: verifies the holder to the standard your privacy team sets. A deletion request from someone who is not the holder can destroy a real customer's account.
2. Agent, depth 3: clarifies the request. Close the account? Erase personal data? Stop marketing? Remove a single item, such as a saved card? Many requests are resolved by something smaller than deletion.
3. Agent, depth 3: logs the request with the date the customer first asked, in their own words, which starts the response deadline.
4. Agent, depth 1: checks blockers: a balance, open orders, active subscriptions, an open complaint or dispute, a restriction marker.
5. Agent, depth 3: explains what happens, from approved wording: what will be deleted, what the firm must keep and roughly why, how long the process takes, and that it cannot be undone.
6. Agent, depth 2: drafts the deletion request for the privacy team, with the verification result, blockers and the customer's confirmation.
7. Person, privacy team: decides what is erased and what is retained, and responds formally where required.
8. Person, operations: closes the account first if needed, following handling requests to close an account.
9. System, under engineering control: executes deletion after approval, with its own log.
10. Agent, depth 3: confirms completion to the customer from the system's record.

Which controls matter, and what evidence does each leave?

The agent has no delete capability. Evidence: the role's permission list and refused verdicts for any attempt.

Request date from first mention. Evidence: the timestamp of the customer's words.

Retention decided by the privacy team. Evidence: the decision listing categories erased and kept, with reasons.

Verification matched to impact. Evidence: authentication level recorded on each request.

Deletion has its own log. The system records what was deleted and when, without keeping the deleted data. Evidence: the deletion log, linked to the request.

What should you measure?

• Requests by type after clarification: closure, erasure, marketing, single item. A large share resolved by smaller actions is a good sign.
• Time from request to completion, against your deadline.
• Requests with blockers, and what they were.
• Deletion requests linked to fraud signals, such as a recent contact detail change.
• Complaints after deletion, such as customers surprised by what was kept.

Where does it usually go wrong?

The agent promises everything will be erased. Records the firm must keep are not deleted. Say so plainly at the start.

Deletion is used by fraudsters. Someone who has taken over an account asks for deletion to hide what they did, or to lock the real customer out. Check for recent changes and route through account recovery if access is in doubt.

Open matters are wiped. A complaint, a claim or a refund in progress disappears with the account. Resolve or retain those first.

Backups and copies are forgotten. The main record is deleted, but copies remain in other systems the customer used. The deletion process, not the agent, must reach them; the agent should not claim it has.

The process is hard to reach. Customers who cannot use online forms need another route; see can an AI agent handle accessibility requests?

Regret comes too late. Some customers delete in frustration and later want their history back. Explaining that deletion is permanent, and offering closure as an alternative, prevents some of these.

Where Pryme fits

In Pryme, deleting customer data is not in any support agent's role. The request is drafted at depth 2 and approved by a named person, and the agent's own conversation records follow your retention rules like any other. The append-only record of verdicts keeps what is needed to show how each request was handled.