FAQ — Generative AI: what can it do for your business?

Generative AI creates new content — text, images, or code — from natural language instructions. Traditional AI classifies or predicts based on patterns in existing data. The key difference: a language model does not automatically know your internal processes; you must supply context explicitly through prompts or retrieval.

Repetitive text-heavy workflows with clear expected outputs: drafting customer responses, summarising long documents, generating first-cut reports, and answering frequently asked questions. The best candidates have accessible data, a measurable baseline, and a human reviewer to catch errors.

Three stand out: hallucination (confident-sounding but wrong answers), data leakage if sensitive content is sent to external APIs, and over-reliance without human verification. Mitigate them by validating outputs against known ground truth, using private deployments for confidential data, and keeping a human decision point for consequential actions.

For most business tasks, prompt engineering combined with retrieval-augmented generation (RAG) outperforms fine-tuning at a fraction of the cost. Fine-tune only when you need a highly specific writing style or domain vocabulary that cannot be expressed in a prompt, and when you have thousands of high-quality labelled examples.

Define a before-and-after baseline. Track time saved per task, error rate, and the percentage of AI drafts accepted without major edits. Divide measurable time savings by the total cost (API fees, development, review hours). Avoid using qualitative impressions as your primary metric.

A proof-of-concept for one defined task typically takes two to four weeks. A production system with access controls, monitoring, human review workflow, and integration into your existing tools takes two to three months. Scope creep and data-access issues are the main schedule risks.