Our FEtch FAQ's, and What You Need to Know

There is a lot of noise around AI, but far less clarity on how it behaves in real business environments. FEtch AI is built to be controlled, predictable, and aligned to clear outcomes, combining different types of AI to support revenue and customer engagement. This FAQ covers the questions we hear most often, explaining how our agents work and how we ensure consistency, control, and reliability.

What is the difference between Generative AI and Agentic AI?

Generative AI creates content based on input prompts and generally operates in a reactive manner, making it best suited as a creative tool. Agentic AI, by contrast, makes decisions based on defined goals or instructions. It is designed to guide conversations, execute workflows, and achieve specific outcomes within controlled constraints. Rather than focusing on creation, Agentic AI focuses on action and decision-making.

Can Generative AI and Agentic AI work together?

Yes, the two approaches are complementary. Generative AI can be used to create content such as drafting an email, while Agentic AI determines when to send it, which data sources to use, and who the recipients should be. Within the FEtch Conversion Engine, this collaboration is demonstrated through multiple agents working together. Drew manages demand generation through email and SMS, Alex handles conversational lead qualification, and Zoe supports sales by generating one-page briefings. Together, they function as a coordinated system.

Are the FEtch agents using Generative or Agentic AI?

Different FEtch agents use different combinations depending on their role. Alex, the Sales Development Representative agent, is a conversational Agentic AI designed to operate within a structured, goal-driven framework. Zoe, the Sales Support agent, uses both Generative and Agentic AI to produce concise briefings but does not engage directly with customers. Owen, the Customer Support agent, and Ryan, the Talent Acquisition agent, both use Agentic AI to carry out their functions.

Can AI agents learn and develop their own behaviour?

FEtch AI agents can adapt based on the data and context provided, but they do not learn independently like humans. During onboarding, the FEtch team trains the agents using relevant language, workflows, and objectives. Knowledge bases are integrated to ensure accurate and informed responses. When this data is updated, the agent’s outputs evolve accordingly. While this may appear as learning, it is actually the result of updated inputs rather than autonomous development, ensuring full control over behaviour and outcomes.

How is the behaviour of an AI agent controlled?

Agent behaviour is governed through a combination of instructions, goals, constraints, and intentional design. Conversations follow structured flows, such as greeting, questioning, responding, and follow-up. This allows the agent to interact naturally while staying aligned with its objectives. Even if a conversation diverges, the agent can provide relevant information and then guide it back toward the intended goal.

How is control maintained over the knowledge the agent can access?

Agents are connected only to approved data sources, including FAQs, websites, documents, and internal materials. This ensures that responses are based on verified information and remain relevant. The agent does not form opinions and relies entirely on the provided knowledge base as its source of truth.

What happens if a customer asks a question that cannot be answered from the knowledge base?

This is managed through configurable access control settings. The agent can be set to operate strictly within the knowledge base or more flexibly by supplementing responses with publicly available information when needed. For example, in a real estate context, property-related questions would rely on internal data, while general questions about local amenities could be answered using external sources. The level of flexibility is defined in collaboration with the client.

How is the conversation controlled, and how does the agent know what to say or avoid?

Conversation boundaries are managed through guardrails, which define what the agent can and cannot say. These include restrictions on sensitive topics, instructions for handling uncertainty, escalation procedures, and rules such as not offering refunds or providing regulated advice. Guardrails also support business processes by identifying trigger phrases or sentiment cues that indicate when escalation to a human agent is required.

Could an agent behave unpredictably or act outside of its intended purpose?

The system is designed to prevent this. Agents operate within strict guardrails and follow predefined instructions. They do not have independent intent or will. Outbound activity is managed through controlled queues, with contact lists, timing, and operating hours defined during onboarding. Additionally, call recordings and transcripts provide full transparency and oversight.

Could an agent insult customers?

No, guardrails are specifically designed to prevent inappropriate or offensive language. These controls ensure that all interactions remain professional and aligned with brand standards.

Could an agent make unauthorised commercial decisions, such as offering discounts or free services?

No, guardrails prevent the agent from making decisions outside its defined authority. This includes restrictions on pricing, offers, and commitments, ensuring that all communications remain within approved business rules.

Could an AI agent breach GDPR guidelines?

FEtch operates as a data processor under GDPR, while clients act as data controllers and are responsible for obtaining and managing consent. Clients must confirm that their data has been collected appropriately. FEtch manages operational responsibilities such as processing opt-out requests and deleting data in line with retention policies, typically within 30 days after campaign completion.

Could an agent introduce bias into decision-making?

FEtch controls the prompts and knowledge bases used by its agents and reviews them before deployment to prevent biased or discriminatory behaviour. Instructions that could lead to unfair treatment are not permitted. The underlying AI models are provided by third-party organisations, and FEtch relies on their responsible AI frameworks while applying its own safeguards at the application level.

How secure is the data shared with the FEtch platform?

Data security is detailed in the FEtch AI Data Security Statement, which covers protections for data in transit, at rest, and across regions. It also outlines infrastructure security measures and certifications designed to protect customer data.

Can the AI agent be hacked?

Security controls and protections are described in the FEtch AI Data Security Statement. These measures are designed to mitigate risks such as unauthorised access and ensure the integrity of the platform.

Are FEtch AI products compliant with current AI regulations?

FEtch operates in compliance with GDPR, the EU AI Act, and applicable local laws. Security practices include encryption, credential rotation, sub-processor assessments, and minimised data retention. Supporting documentation, including risk registers and technical specifications, is available upon request.

Where does the data reside?

Data residency details are outlined in the FEtch AI Data Security Statement, providing transparency on where data is stored and processed.

How does FEtch classify AI risk?

Every AI product is classified prior to deployment, and this classification determines the level of governance applied. Prohibited use cases, such as social scoring or manipulation, are not pursued. High-risk applications are developed only with full documentation, formal assessment, enhanced oversight, and executive approval. Limited-risk systems include transparency measures and appropriate controls, while minimal-risk systems follow standard development and operational practices.