Per André works with AI implementation at Frontkom and helps organisations move from generic tools to AI that actually understands their business. This is what he sees working in practice.
ChatGPT and similar tools are impressive. But they do not know your customers, your products or the tone that fits your brand. They give general answers to general questions. That is useful, but it is not what creates lasting competitive advantage.
Three levels of AI customisation
Level 1: Prompt engineering. You give the model context in the request itself. Who you are, what the tone is, what the purpose is. It is simple and requires no technical implementation. It improves results significantly, but depends on every user doing it correctly every time.
Level 2: RAG — Retrieval Augmented Generation. You connect the model to your own documents, product descriptions, guidelines and history. The model retrieves relevant information from your knowledge base before responding. The result is answers grounded in your actual context, not general knowledge. This is the level where most organisations should start.
Level 3: Fine-tuning. You train the model on your own data so it adopts your tone, terminology and domain language. This requires more data and more technical expertise, but produces models that are very difficult for competitors to replicate.
What you actually get out of it
Organisations we work with that have implemented RAG-based solutions typically report three outcomes: faster responses to customers because the system retrieves correct information without manual search; higher consistency in communication because everyone draws from the same source; and reduced time on internal questions because employees can ask the system instead of colleagues.
AI is no smarter than the data you give it. Start by getting your own house in order, then we build from there.
Want to understand which level of AI customisation makes the most sense for your organisation right now? Get in touch with PA for a technical walkthrough.
