R&D in software product companies: how to organise it

A straight track of evenly spaced blocks with a branching path of small flasks above it, one branch rejoining in blue

A practical way to run research and development next to everyday product delivery, so that uncertain work gets protected time, clear stop points and records that hold up later.

In short

Software product companies organise R&D best by separating uncertain, exploratory work from routine product delivery, giving it a fixed share of capacity, running it as time-boxed experiments with stop criteria, and moving results into product teams when they are proven. Documenting the work as you go also supports tax relief and accounting decisions.

In a software product company almost everyone "develops", so the term R&D is easily stretched to cover all engineering. That blurs two very different kinds of work: building features you know how to build, and exploring things where you do not yet know whether they can work. Each needs a different way of planning, funding and measuring. This guide from Devies, a Gothenburg-based product and engineering consultancy, is for CTOs, heads of product and finance leaders who want research and development to produce real advances without disrupting delivery.

What counts as R&D in a software company?

R&D is work that aims to create new knowledge or substantially new solutions under real technical uncertainty, done in a planned and recorded way. Building a feature with known techniques is development in the everyday sense, but usually not research and development in the formal sense.

The most widely used definition comes from the OECD's Frascati Manual 2015, the international reference for measuring research and experimental development. It sets five criteria that must all be met: the work is novel, creative, uncertain in its outcome, systematic, and transferable or reproducible. "Systematic" means the work is planned and budgeted, with records kept of both the process and the outcome.

For a software company, that gives a useful test:

Work

Likely R&D?

Why

A new algorithm to match data that current methods cannot handle

Likely R&D?

Often yes

Why

Outcome uncertain, new knowledge

Testing whether a language model can reliably do a regulated task

Likely R&D?

Often yes

Why

Uncertain, needs systematic experiments

Adding a settings page with a known framework

Likely R&D?

No

Why

Known solution, no real uncertainty

Upgrading libraries and fixing bugs

Likely R&D?

No

Why

Maintenance, not new knowledge

Integrating a standard API from a vendor

Likely R&D?

Usually no

Why

Documented, predictable outcome

The Swedish term forskning och utveckling (FoU) is used in the same broad sense, but Swedish tax rules have their own conditions, covered in step 6 below.

How to organise R&D step by step

The steps below build a small, explicit system for exploratory work inside a product company: clear criteria, protected capacity, short experiments and a path into the product.

An experiment loop leading to a decision diamond with three exits: back to the loop, into a product block in blue, or a stop bar
  1. Name the strategic questions. Write down the three to five technical unknowns that would matter most to the business if solved, for example "can we process customer data in real time at ten times today's volume?". Exploration without questions turns into side projects.
  2. Separate R&D from delivery work. Keep a distinct backlog for exploratory work, using the criteria above. Mixing it into the product backlog means it is always the first thing cut when a deadline approaches.
  3. Reserve capacity on purpose. Decide what share of engineering time goes to exploratory work, and whether it is done by a dedicated group, by rotating people from product teams, or both. A dedicated group builds depth; rotation spreads knowledge. Smaller companies often start with rotation.
  4. Run time-boxed experiments. Frame each piece of work as a hypothesis with a time box, a way to measure the result and a stop criterion agreed in advance. Our comparison of a prototype, MVP or pilot helps choose the right vehicle for testing an idea.
  5. Decide at fixed points. At the end of each time box, choose one of three outcomes: continue, move into product delivery, or stop and record what was learned. Stopping is a valid result, not a failure.
  6. Document as you go. Keep a short record per experiment: question, approach, people involved, time spent, result. In Sweden this also supports the employer contribution relief for research and development, described in the next section.
  7. Move results into product teams. When an experiment succeeds, plan the handover: who builds the production version, what must be rewritten, and what the prototype code must not become. Our guide to the product development process for digital products covers what happens next.

What do Swedish tax and accounting rules ask of R&D?

Swedish tax and accounting rules both reward clear separation and good records: tax relief depends on who did qualifying work and for how long, and accounting treatment depends on whether work is research or development.

The research deduction (forskningsavdrag). As of November 2026, Skatteverket's guidance sets out these main conditions:

  • The deduction is 20 percent of the pay of qualifying employees, capped at SEK 3 million per group of companies and calendar month (rules in force since 1 January 2024).
  • An employee qualifies if at least half of their working time is spent only on research or development, and at least 15 hours a month.
  • The work must be systematic and qualified, and done for commercial purposes. Management, support, service, maintenance, quality assurance and marketing do not count.
  • Development here means using research results to create new or substantially improved goods, services or processes, where the research results are decisive. Routine product improvement does not qualify.
  • You must be able to show, for each period, that the work and the time requirements were met, for example with project descriptions, planning documents and time reports.

The deduction is reported monthly in the employer declaration. Much ordinary product work in a software company will not meet these conditions, so check specific cases with your accountant or Skatteverket before claiming.

Accounting. Under the international standard IAS 38 Intangible Assets, research expenditure is recognised as an expense, while development expenditure that meets specified criteria is recognised as an intangible asset. Swedish companies that do not report under IFRS follow Swedish accounting rules, so confirm the treatment with your auditor. In both cases, the same experiment records help.

How do you measure whether R&D is working?

Research is working when it regularly turns open questions into clear answers, and some of those answers into product advantages. Measure learning and transfer, not only output.

Useful measures:

  • Questions answered per quarter, including those answered with "no".
  • Time to a decision from starting an experiment to continue, transfer or stop.
  • Transfer rate: how many experiments led to something in the product within a year.
  • Share of capacity actually used for exploration, compared with what was planned. If delivery keeps eating it, the protection in step 3 is not working.

Avoid judging research by the same delivery metrics as product teams. Uncertain work cannot be estimated like feature work; our guide to software estimation explains why uncertainty should be shown as ranges rather than single numbers.

What are common mistakes when organising R&D?

The most common mistake is letting research become either an unprotected leftover or an isolated lab with no path into the product. Watch for these warning signs:

  • Research time is planned but always borrowed by delivery when deadlines approach.
  • Experiments have no stop criteria, so they run for months.
  • Prototypes are shipped to customers as if they were production code.
  • Nobody records what was tried and why it failed, so the same ideas are tested twice.
  • Results are presented to product teams only at the end, instead of involving them early.

Devies

Want a second opinion?

Devies, a Gothenburg-based product and engineering consultancy, works on questions like this with Nordic product and IT teams. Tell us where you are and we will suggest a sensible first step.

Talk to Devies

Frequently asked questions

What does R&D mean in a software company?

In a software company, R&D means work that aims to solve problems where the outcome is genuinely uncertain and new knowledge is needed, such as new algorithms, architectures or uses of AI. Building features with known methods, maintenance and routine integration are development in the everyday sense but usually not R&D. The OECD Frascati Manual's five criteria are the common reference for drawing the line.

Does software development qualify for the Swedish research deduction?

Some of it can. Skatteverket requires that employees spend at least half their time only on research or development, at least 15 hours a month, on systematic and qualified work. Development must use research results to create new or substantially improved products or processes. Routine feature work, maintenance, support and quality assurance do not count, so check specific cases with an adviser.

How much engineering time should go to R&D?

There is no standard share; it depends on how much your product's future depends on unsolved technical questions. The important part is to decide the share on purpose, protect it from delivery pressure, and review it each quarter. Starting small, with a few experiments at a time and clear stop criteria, is safer than a large, vague R&D budget.

Should R&D be a separate team?

It depends on size and depth. A dedicated group builds deep expertise in hard problems, but risks drifting away from the product. Rotating engineers from product teams into R&D keeps it close to real needs and spreads knowledge. Many companies combine a small core with rotating members, and plan every handover into product teams from the start.

Sources

Here’s more

Join Our Newsletter

We handle your data according to our privacy policy.

Shopping Cart

We use cookies

We use cookies to improve your experience on this website. You may choose which types of cookies to allow and change your preferences at any time. Disabling cookies may impact your experience on this website. You can learn more by viewing our Cookie Policy.

Check out

Contact Form text

We will contact you text

We handle your data according to our privacy policy.