# Methodology for usage estimates

> Methodology behind the Autopilot usage estimates: model classification categories, relative cost ratings, and project complexity tiers.

This page describes the methodology behind the [Autopilot usage estimates](autopilot-licensing.md): how models are classified and how project complexity is determined.

## Model classification

Models are grouped into three categories based on their capabilities and the workloads they are best suited for. Each model also has a relative cost rating from 1 — the lowest relative usage cost — to 5 — the highest.

| Category | Best suited for | Relative cost rating |
| --- | --- | --- |
| **Cost-efficient** | Focused, repeatable, and high-volume tasks | 1–2 |
| **Standard** | Broad, general-purpose capabilities; suited to most tasks | 3 |
| **Premium** | The most complex tasks, including long-running, multi-step work | 4–5 |

The relative cost rating is an ordinal scale that compares usage-pool consumption across models for comparable tasks. It does not represent a monetary amount or an exact cost multiplier. For example, a rating of 4 does not mean twice the consumption of a rating of 2. Actual consumption also depends on the task's length and complexity. Ratings may change as underlying model costs and the Autopilot model lineup evolve.

## Project complexity

Autopilot usage estimates use three complexity tiers for UiPath RPA projects. The classification follows this order:

1. A project is **Complex** if it meets any Complex criterion.
2. If no Complex criterion applies, a project is **Simple** only if it meets every Simple criterion.
3. A project is **Medium** if it does not meet every Simple criterion and no Complex criterion applies.

## Simple projects

A Simple project is a focused automation with limited integration and UI risk. A project is classified as Simple only when **all** of the following are true:

- Uses 1 identifiable application or system
- Has fewer than 10 estimated process steps
- Has no decision branches or exception-handling logic
- Has fewer than 3 UI selector changes
- Uses structured inputs only, with no Document Understanding, optical character recognition (OCR), Intelligent Xtraction Processing (IXP), or machine-learning extraction signals

## Medium projects

A Medium project is a standard departmental or cross-functional automation requiring moderate design, testing, and integration effort. A project is classified as Medium when it is more involved than Simple but does not meet any Complex criterion. Typical indicators include:

- Uses 2–3 applications or systems
- Has 10 or more estimated process steps
- Includes some decisions, branching, or exception-handling logic
- Has a moderate number of UI selector changes
- Processes documents or other unstructured inputs
- Requires additional testing and process hardening, but does not contain a high-complexity trigger

## Complex projects

A Complex project is a strategically important automation that is likely to require advanced solution design, stronger governance, broader testing, and ongoing operational management. A project is classified as Complex when **any one** of the following is true:

- Uses 4 or more applications or systems
- Includes 5 or more decision or branching steps
- Interacts with a fragile, dynamic, or frequently changing user interface
- Includes agentic automation
- Requires human-in-the-loop activity, such as review, validation, approval, or exception resolution

## Methodology terms

| Term | Definition |
| --- | --- |
| System | A distinct application, platform, API, or app-specific connector. Generic HTTP or UI automation counts as one unknown system. Standard UiPath packages do not count by themselves. |
| Step estimate | Total activity count, when available. Otherwise, activities added plus one step for every eight property changes. If neither is available, recorded Autopilot events provide the estimate. |
| Structured input | Input with no Document Understanding, OCR, IXP, or machine-learning extraction signals. |