AI customer service agents: how to comply with Spanish law without removing the bot

If your business uses an artificial intelligence agent to handle customer enquiries, the change to prepare for before 28 December 2026 is the ability for customers to speak to a person when they ask to do so. Spain's Law 10/2025 on customer service establishes that, within its scope, an automated answering system cannot be the only means of support. A human representative must be able to respond in real time when the customer requests one.

This does not mean you need to remove your AI agent. It means you need to design what happens when automation is no longer enough: how customers request a person, who receives the conversation, what information is passed on and how long the team takes to respond. The twelve-month adaptation period ends on 28 December 2026, so it is worth reviewing the entire journey before then.

Check whether the law applies to your business

Not every business becomes subject to these requirements simply because it uses a chatbot for customer service. Article 2 of Law 10/2025 defines its scope by reference to certain essential services of general interest and businesses meeting at least one of the following thresholds:

• 250 employees or more.
• Turnover of more than €50 million.
• A balance sheet total of more than €43 million.

Meeting one of these business thresholds is sufficient within the other applicable criteria of the law. Do not treat this as a rule that automatically applies to every small or medium-sized business. Equally, do not decide that your business is excluded based on its size alone without checking the relevant legal scope.

If your organisation may be covered, confirm its position with a legal adviser. Even if it falls outside the law, a reliable route to human support remains a useful way to prevent customers from getting stuck in automated loops.

What your agent must allow from 28 December

Article 8 addresses two points that should become technical requirements. First, automated answering systems cannot be the only means of customer service. Second, when a customer asks to speak to a person, human support must be available in real time, whenever the request is made.

The law also sets a specific performance measure: 95% of requests must be answered in less than three minutes on average. Adding a button labelled “Speak to a person” is not enough if the request disappears, reaches nobody or remains in an unmonitored queue.

Your agent can still answer frequently asked questions, collect information and help classify an issue. The important thing is that it must not block the route to a person when requested, and the handover must lead to an operational support process.

Design the handover as part of the product

Start by deciding how customers can request human assistance. Do not rely solely on them typing one exact phrase. Recognise equivalent expressions and provide a visible option to request support. Intent detection should be straightforward rather than turning the transfer into a prolonged negotiation with the bot.

Next, determine what the human representative needs to continue the conversation. Design the handover to preserve the reason for contact and a summary of what has already been discussed, along with the information needed to handle the case. The aim is to avoid making customers explain the same problem from scratch.

Do not confuse the time taken by the bot to reply with the time taken for a person to handle the request. Define the start and end events for your measurements precisely, and confirm the appropriate interpretation of the legal requirement with your legal adviser.

Measure human support, not just bot performance

An agent may resolve many enquiries automatically and still provide a poor experience when something goes wrong. Alongside automated resolution rates, track how many customers request human support, how long they wait and how many transfers complete successfully.

Test realistic scenarios before releasing changes: a direct request to speak to someone, an issue the agent cannot resolve, a conversation with several previous exchanges, and a transfer attempt while the team is busy. Check that staff receive enough information to continue without repeating the entire process.

This is part of a wider design challenge: balancing capacity and reliability in AI agents. Automating more steps delivers little value if the route to a person fails when it is needed.

Prepare an adaptation plan before the deadline

This week, establish whether the law applies to your organisation, map the journey from the first interaction to human support, and assign an owner to each stage. Then test the handover and measure response times using trial conversations. Do not wait until the deadline to discover that the agent has no team to transfer to or that the conversation history disappears when the channel changes.

If you are commissioning a new system, include these requirements from the outset alongside the scope of automation and the resources needed for human support. You can also review the factors that determine the cost of an AI agent before deciding which features to build and which to integrate.

At Contrast 3D, we design AI agents and automations connected to real business processes. Tell us how your business handles customer enquiries and we can help you define a workflow that combines automation, human handover and operational measurement.