Building a website or app with AI: what is the real cost?

Distinguish a demo from a usable tool

AI can help produce an interface, code and variations quickly. A visible first version makes an idea easier to discuss. It does not yet prove that a website works for its visitors or that an app delivers in difficult cases.

Define what you intend to launch: a brochure website, an internal prototype or a service used by customers. Content, access, payment and continuity needs will differ.

Count preparation time

Work begins before generation: clarify the audience, journeys, pages and information needed. Gather approved text, usable images and business rules before asking for an implementation.

Include the time spent explaining context, comparing proposals and correcting inconsistencies. Repeated attempts have a cost even when you do the work yourself; keep a record of the hours spent on the project.

Check real behaviour

Test important journeys on desktop and mobile: navigation, forms, error states and confirmations. A well-designed button does not prove that an enquiry reaches the right recipient. Also check content, links and the elements needed for search visibility.

If the app manages accounts or internal data, have access and permissions checked by someone qualified. Confidence in a generated answer does not replace testing the implementation. An external review can focus on the areas you cannot check yourself.

Plan costs after launch

List the domain, hosting, external services and subscriptions needed to operate the solution. Examine usage limits and the cost of increased demand. Include backups, updates and incident resolution time.

Keep source files and access under the company's control. Document deployment and how to return to a previous version. The project should be able to continue even if you change the generation tool.

Choose a scope you can support

A limited launch can be a useful way to test an idea, provided you know what is ready and what remains experimental. Reduce the scope rather than adding functions you cannot test or maintain.

Compare working alone, occasional support and outsourced delivery across the same scope. Count spending and your time, then choose according to real use and the consequences of failure.

• Define the expected level of service.
• Count preparation, correction and checking time.
• Test user journeys and access.
• Plan operation, maintenance and rollback.