The vertical doesn't matter as much as people think. Whether the business processes insurance claims, hospitality bookings, or e-commerce returns — the underlying system types that create value are remarkably consistent.
This is useful because it means you don't need to know everything about a specific industry to understand where AI creates leverage. You need to understand which of the five types applies to the workflow you're looking at.
Intake and routing systems
Any workflow where the first step is a human reading something and deciding where it goes. Claims, support tickets, applications, leads. The decision criteria are usually encodable; the volume is usually high enough to make automation worthwhile.
Document intelligence systems
Extraction and structuring of information from unstructured documents. Contracts, invoices, applications, reports. The value is converting something a human has to read into something a system can act on.
Internal operations agents
Systems that monitor internal data sources and surface relevant information without being asked. Anomaly detection, exception flagging, status aggregation across disconnected systems. The person who used to check five dashboards every morning.
Customer-facing communication systems
Context-aware outbound and inbound communication at scale. Not generic templates — responses that incorporate the customer's specific situation from structured data. Acknowledgments, status updates, follow-ups, escalations.
Decision-support and scoring systems
Systems that provide structured input to human decision-makers. Lead scoring, risk flagging, recommendation engines. The goal is not to replace judgment — it's to make sure the human making the judgment has the right information at the right time.
Why this taxonomy matters
When someone describes a problem as "we need AI for our operations," the first question is always: which type? Each one has different data requirements, different integration complexity, different success criteria, and different risks.
A document intelligence system fails differently from an intake routing system. Knowing which type you're building from the start changes what you validate in the pilot, what integrations you prioritise, and what you measure in production.
[TODO: refine voice — add a paragraph on how to identify which type applies to a given workflow; the diagnostic questions are the most practical part of this framework and worth spelling out]