BlogAI & Innovation Software development and the R&D Law: when your project qualifies
Many tech companies in Chile wonder whether their software development can qualify under the R&D Law. The answer is that the law expressly includes it, but with a condition that rules out most “ordinary” software.
What the law says
Within its definition of experimental development, Law 20,241 includes the development of software, as long as it:
- produces greater knowledge aimed at systematically resolving a scientific or technological uncertainty, or
- delivers a substantial, innovative improvement in a process, product or service.
In other words, writing new code isn’t enough. The code must be the means to solve a technical problem you didn’t know how to solve.
What usually qualifies and what doesn’t
| Usually doesn’t qualify | Can qualify |
|---|---|
| An app or website with well-known features | A new algorithm for a problem with no known solution |
| Integrating existing systems via APIs | An architecture that achieves performance existing solutions can’t |
| Migrating a system to the cloud | A method to process data of a type or volume current tools can’t handle |
| Design or interface changes | Predictive models that require researching new approaches |
| Maintenance and bug fixing | Software for a physical or scientific process that has never been automated |
The law also expressly excludes cosmetic or minor modifications of existing applications to set them apart from others.
The R&D core inside a product
In practice, almost no software product is 100% R&D. Usually there’s a core with technical uncertainty (a calculation engine, an algorithm, a model) surrounded by standard work (interface, authentication, integrations, deployment).
Our recommendation is to separate that core and apply with it as the R&D project, rather than presenting the whole product. It’s easier to defend and easier to audit.
What if we work with agile methods?
There’s no contradiction. A University of Chile thesis (Olivares, 2014) analyses a telecom company’s project that developed algorithms to detect large-scale network failures and was managed with SCRUM.
What matters is that each cycle leaves evidence of R&D:
- Hypotheses the sprint tests.
- Experiments and their results, including the ones that failed.
- Technical decisions and why they were made.
A feature backlog isn’t enough evidence. An experiment log is. We explain how in How to document technical uncertainty.
Which costs can be included
In a software project, typical costs are:
- Salaries of the development and data science team, according to their time on the R&D core.
- IT services, such as cloud and compute, needed for the project.
- Subcontracts for specific R&D tasks.
- Costs of registering the software that comes out of the project.
For artificial intelligence projects in particular, see Leveraging AI for innovation.
This article is for information only and is not tax advice. Sources: Law 20,241, Corfo Exempt Resolution No. 17 and Olivares (2014), University of Chile.
The R&D Law paradox: a generous incentive almost no one uses
Chile has one of the most generous R&D tax incentives in the OECD, yet only a tiny fraction of companies use it. We look at the data and the likely reasons.
What is the UTM, and why is Chile's R&D Law measured in it?
A 100 UTM minimum, a 15,000 UTM cap, a 10 UTM fee: the R&D Law speaks in UTM. What this unit is, how it is updated and how to use it in your calculations.