Customers are increasingly asking sensitive questions, making decisions inside personal channels like WhatsApp and expecting instant, reliable guidance. Brands want to meet these expectations without taking risks, exposing data or losing control of their voice.

Guardrails are the framework that ensures every AI interaction remains safe, compliant and aligned with your brand. They protect your customers, safeguard sensitive information and give you confidence that Merx will always behave in a way that strengthens trust rather than putting it at risk.

More than anything, they are the reason our clients can adopt conversational AI at speed, with assurance that their data, their brand and their customers are handled with care.

Why Guardrails Matter More Than Ever

AI is entering the centre of digital commerce. Customers now expect support that feels instant, personal and knowledgeable, whether they are comparing two shades, asking about ingredients or seeking clarity on returns. Many of these questions happen inside WhatsApp, a private channel where customers naturally speak more openly.

This environment requires precision and responsibility. A single incorrect claim, a misinterpretation of a medical query or an accidental disclosure of personal information can erode trust immediately.

Brands want to unlock the benefits of AI without risking:

• Unsafe or inaccurate guidance
• Unintended claims about product performance
• Storage or exposure of personal or medical information
• Responses that fall outside regulatory boundaries
• Inconsistency with brand tone or policy

This is the purpose of Merx Guardrails. Instead of relying on a single model to do everything, we have designed a multi-layer system that guides every step of the interaction, from the moment a customer sends a message to the moment a reply is delivered.

It means you get the value of advanced AI, but with the reliability and governance of an enterprise-grade customer experience platform.

The Input Guardrail: Your First Layer of Protection

Every message your customers send is processed through our Input Guardrail, which acts as a safety filter before any AI model begins reasoning.

It plays four essential roles.

1. Protecting customer privacy

Merx does not send personally identifiable information to any LLM provider.

The Input Guardrail identifies and masks details such as names, phone numbers, email addresses, order references and other personal identifiers. This ensures that LLMs only receive the information required to understand intent, never the private details of your customers.

2. Filtering sensitive or medical information

Customers occasionally share more information than intended when messaging a brand. This can include medical conditions, allergies, pregnancy, religious preferences or other personal details that fall under Special Category Data as defined by GDPR (Article 9).

Special Category Data includes information relating to health, biometric or genetic data, religious beliefs, political opinions, sexual orientation and other highly sensitive attributes that require additional protection under European data protection law.

Merx is designed to prevent this type of data from ever becoming part of your customer database or being used for personalisation.

How Merx handles this safely:

• Merx does not store sensitive or special category information in its database
  
• This information is not used to personalise guidance or recommendations
  
• Any detected sensitive content is filtered and redacted at the point of input
  
• Messages containing this information are routed to safe, pre-approved responses that keep the conversation helpful while remaining fully compliant
  

This approach aligns with GDPR principles of data minimisation and purpose limitation. By filtering and redacting sensitive content before it is processed or persisted, Merx avoids the need for explicit consent, extended retention or additional compliance overhead for brands.

In practice, this means your customers can ask questions naturally, without you needing to worry about accidentally collecting or processing information that should never be stored in the first place.

Why This Matters for Conversational Commerce

Conversational interfaces increase openness. Customers are more likely to express personal context in chat than on a website form. Without the right safeguards, this can introduce risk.

Merx Guardrails ensure that conversational commerce remains safe at scale by:

• Preventing accidental collection of special category data
  
• Reducing regulatory exposure for brands
  
• Maintaining customer trust in private channels like WhatsApp
  
• Enabling AI driven journeys without compromising compliance
  

This is a critical distinction between Merx and generic AI or chatbot solutions. Guardrails are not an add-on. They are embedded into how the system works, ensuring that every interaction remains respectful, compliant and aligned with best practice

3. Understanding the question accurately

Before an AI model responds, the Input Guardrail classifies the message. It identifies whether the customer is:

• Comparing products

• Seeking how-to information

• Asking about policies

• Raising a concern

• Expressing intent to buy

• Asking a question that is out of scope

This classification step ensures the customer receives the most appropriate type of support and prevents the AI from drifting into areas it should not handle.

4. Blocking inappropriate requests

If a message seeks advice that is unsafe, overly personal or outside your brand’s authority, the Guardrail intercepts it. This allows us to maintain high-quality conversations without compromising on safety.

The Input Guardrail is a fundamental part of how Merx operates. It reflects our commitment to respecting both customer privacy and brand responsibility.

Multiple Guardrails Working Together

Where some providers rely on a single LLM to manage everything, Merx uses a layered guardrail strategy. Each layer acts as a check and balance for the next, ensuring that no single model has the final say over what the customer receives.

1. Input Guardrail

Screens, masks and classifies customer messages to protect privacy and prevent unsafe topics.

2. Knowledge Guardrail

Restricts AI responses to your approved documents, product information, tone guidelines and policies.

This prevents the model from inventing details or using non-verified knowledge.

3. Reasoning Guardrail

Evaluates whether the proposed answer complies with your brand rules, accuracy requirements and safety boundaries. If not, it is regenerated or replaced with a more suitable response.

4. Output Guardrail

The final safeguard. It checks the tone, clarity and compliance of the message before it is shown to the customer, ensuring it meets your standards and never includes sensitive information.

Together, these layers create a structured and reliable system that delivers high quality responses at scale.

Data Governance You Can Trust

Privacy is a principle, not a feature. We have designed Merx with strict data minimisation practices that protect both customers and brands.

What Merx stores

Merx stores only the information required to improve conversational journeys, such as:

• High-level summaries of customer questions
• Product interests
• Preference signals (for example preferred shades or categories)
• Intent categories
• Non-PII behavioural patterns

All stored data is fully compliant with GDPR and handled within ISO-aligned controls.

What Merx does not store

Merx does not store:

• Medical information
• Pregnancy or health related disclosures
• Diagnoses or treatment details
• Personal addresses or contact details in plain text
• Financial details
• Raw PII tied to individual profiles

If a customer chooses to share something sensitive in conversation, it is immediately masked and never saved to our systems or used to train models.

This approach guarantees that customer privacy is respected at every point and ensures that brands can deploy AI safely even in categories where trust is paramount.

What This Means for You

Merx Guardrails give you the confidence to offer AI powered conversations that are:

They make it possible to scale high performing WhatsApp journeys without increasing operational risk. They ensure that everyday questions are handled instantly and safely, while giving your team time to focus on the cases that genuinely require human care.

Most importantly, they give your customers confidence that they are speaking with an assistant that is helpful, trustworthy and respectful of their privacy.

Guardrails are not an invisible back-office feature. They are a core part of what makes Merx dependable in a rapidly changing landscape. As AI continues to reshape the way customers discover, evaluate and buy, our guardrail system ensures that every interaction remains grounded in safety and transparency.

This is the foundation on which we will build the next generation of message led commerce.