Title: How we implement AI in businesses | ARKEA IA
Description: See ARKEA IA’s process for taking a use case from diagnosis and architecture through build, validation, production and optimization.

# From a real process to an AI system in production.

An implementation follows five stages that let us understand the problem before building and validate the system before putting it into operation.

[Explore an implementation](contacto.html#proyecto)
## Five stages, one responsibility

### Diagnosis

Understand the current workflow, data, tools, constraints and the outcome that should improve.

### Architecture

Define context, integrations, tools, skills, automation, permissions, UX and oversight criteria.

### Build

Develop the system and connect the required components.

### Validation

Test real scenarios, behavior, usage, errors and user experience.

### Production & optimization

Launch with control mechanisms and prepare technical continuity.

## How long does it take?

Most implementations are structured around an approximately three-month cycle. The final schedule depends on the number of integrations, data availability, scope and technical complexity.

## What we need from the company

- Access to process owners.
- Understanding of the current workflow.
- Required documentation.
- Authorized technical access.
- Feedback during development.
- Validation of scenarios and outputs.
## What we do not do

- We do not choose AI first and then search for a problem.
- We do not add an agent when simple automation solves the task better.
- We do not treat a prototype handoff as the end of an implementation.