Three things, one label
A chatbot answers questions in a conversation. An AI feature adds model capability to software that already does a job. An AI app is an application whose central function would not exist without the model.
The distinction matters because the cost differs by a factor of five, and because most ideas are the middle one.
Which is yours?
- If it would still be useful with the AI removed, it is an app with an AI feature
- If removing the AI leaves nothing, it is an AI app
- If the whole interaction is a conversation, it is a chatbot
“We want an AI app that helps our team find information in our documents” is a retrieval feature inside an internal tool. Building it as a standalone app usually costs more and gets used less, because people do not visit a second destination.
What a real AI app looks like
- A document processing application where extraction is the product
- A matching platform where semantic similarity is the whole value
- A drafting tool where generation is the core workflow
- A monitoring system where classification of unstructured input is the job
In each case, remove the model and there is no product left.
What every AI app needs beyond the model
The model is a small part of the build. What surrounds it is most of the work: an evaluation set, a review and correction interface, failure handling, cost controls, monitoring for quality drift, and an audit trail.
Budget roughly as much for that as for the feature itself. Skipping it produces a demonstration rather than a product.
Cost and time
| Typical cost | Timeline | |
|---|---|---|
| AI feature inside existing software | £15,000–£40,000 | 8–16 weeks |
| Standalone AI app, one workflow, one role | £40,000–£90,000 | 3–5 months |
| Multi-role AI platform with billing | £90,000–£200,000 | 5–9 months |