Is it R&D? A five-question test

Five questions

1. Is it new? The work must aim at new findings (novel).
2. Does it need original ideas? It must rest on original, not obvious, concepts and hypotheses (creative).
3. Is the outcome uncertain? At the start, you must not know whether – or how – it will work (uncertain).
4. Is it planned? The work must be planned and budgeted (systematic).
5. Could someone else build on it? It must lead to results that could be reproduced or transferred (transferable and/or reproducible).

Criteria named in §1.33; the questions are our plain-language reading of the headings of §§2.14–2.20. The full definitions are on the Frascati Manual page.

The usual traps

Many activities sit close to R&D and are done by the same people, but the manual excludes them when R&D is measured (§2.111). The ones that cause most confusion:

Routine testing

Routine testing and analysis of materials, products, soils or air is not R&D (§2.113).

Routine medical work

Normal application of specialised medical knowledge is usually not R&D, although an R&D element may appear, for example in university hospitals (§2.115).

Feasibility studies

Checking an engineering project with existing techniques before deciding to go ahead is not R&D; a feasibility study for a research project is (§2.114).

Consultancy presented as research

The manual warns of “a significant risk that some types of socio-economic consultancy … are inaccurately presented as R&D” (§2.119).

Handing out grants

Raising, managing and distributing R&D grants is financing, not R&D (§2.121).

Software built with known tools

Business software, websites and new features in existing programs built with known methods are not R&D. Software counts only if it leads to an advance in the area of computer software (§§2.70–2.72).

When only part of a project qualifies

A project can contain R&D without being R&D from start to finish. Three rules from the manual help draw the line:

• Writing up the original report of R&D findings is part of R&D, while general information services are not (§2.112).
• Debugging software counts only if it happens before the end of experimental development (§2.72).
• Transport, cleaning, maintenance and central administration are not R&D work, but their cost enters R&D expenditure as overheads (§2.122).

For software projects the OECD's 2025 guidance adds practical advice: keep records that separate eligible R&D from the rest of the project.

Statistics are not tax law

This test is the statistical one: it decides what goes into R&D surveys such as the Polish form PNT-01. Tax relief follows its own rules. In Poland the tax acts have their own statutory definition of R&D, and the Ministry of Finance's 2019 IP Box guidance draws on the manual but applies Polish law. For tax questions the statute decides – details on the Frascati Manual page. This article is not tax advice.