Classical, Semantic, Generative or Agentic? Matching the Right AI to the Problem
AI is a spectrum, not a single technology. Leaders who know which kind of intelligence fits which business problem make better investment decisions.
Large language models have made it feel as though AI arrived in 2022. In reality, many of the techniques businesses rely on have been maturing for more than three decades. Generative AI changed the interface by making plain English a way to instruct machines. It did not make the older techniques redundant. The leaders getting the most from AI understand the whole range and pick the right tool for each problem.
How machines become intelligent
It helps to think of intelligence as a cycle: acquire information, structure it, understand it, apply it, and adapt. In practice, useful enterprise AI combines four ingredients: knowledge, data, tools and orchestration.
The four families of AI
Classical or traditional machine learning. Prediction, classification, clustering and forecasting from structured data. Still the workhorse for demand planning, churn, credit risk and fraud.
Semantic AI. Knowledge models, ontologies and rules that give data shared meaning. It improves the accuracy of other AI and is essential where decisions must be explained.
Generative AI. Creating and summarising text, images and code, and understanding unstructured documents, audio and video.
Agentic AI. Systems that plan, call tools and act across workflows, working with other agents and with people.
Start from the business problem, not the technique
Most business questions fall into a small set of generic problem types: classification, regression and forecasting, clustering and segmentation, recommendation, matching, optimisation, search, summarisation and content generation. Name the problem type first and the choice of technique becomes much clearer.
Matching is a good example. Connecting the right customer to the right product, the right candidate to the right role, or the right supplier to the right order is one underlying capability. With good data about your core entities, one matching engine can support personalisation, cross-selling, recruitment and risk screening.
What AI changes about work
Granularity. Decisions can be personalised down to a single customer or asset.
Availability and scale. Intelligence works around the clock and scales with demand.
Dynamism. Models adapt as data changes.
Multi-modal input. Documents, images, audio and sensor data no longer need to be structured first.
Integration. Intelligence can be embedded into devices, tasks and workflows.
A worked example
Imagine a customer asking to switch subscription plans. One request involves several kinds of intelligence:
language understanding to pick out the intent and account details
rules to check eligibility
a compliance and risk check
enterprise systems for billing and orders
workflow automation to make the change and ask for feedback
No single model does all of this. The value comes from orchestration.
Generative AI is the new front door. Behind it, classical, semantic and operational systems still do much of the work.
Readiness checklist
Finish digitising your records, and keep metadata about your data sources.
Create an AI-ready data layer with clear access policies.
Put business, domain and technology people at the same table for each use case.
Build reusable models and share them across teams rather than duplicating effort.
Develop in-house expertise rather than relying entirely on vendors.



Comments