R&D tax credits for software companies used to be one of those quiet wins bootstrapped founders barely had to think about. Build something new, keep decent records, claim it back at year end, and HMRC would generally wave it through. That world has gone. HMRC has spent the last couple of years cracking down hard on R&D tax relief fraud, and the collateral damage is landing on exactly the kind of genuine, bootstrapped software businesses this scheme was meant to help.
I have claimed R&D tax relief through more than one of my own businesses over the years, so this is not theoretical for me. I have also watched the goalposts move enough times to have a strong opinion about where things stand now, and what founders should actually do about it.
What R&D tax credits actually give you
The basic mechanics have not changed. If your company is doing genuine research and development, you can claim tax relief on qualifying costs such as staff time, software, and certain subcontractor spend. For a loss making SaaS business in its early years, that can mean a cash payment from HMRC rather than just a reduced tax bill, which matters enormously when you are bootstrapping and every pound of runway counts.
The rates and scheme names have shifted around a fair bit too, with the old SME scheme and RDEC being merged into a single scheme for most companies. The detail changes yearly enough that I would not trust any blog post, including this one, as your final source of truth on the numbers. What has not changed is the principle. You are being rewarded for taking on technical risk that a competent professional in your field could not have resolved by simply looking it up.
Why HMRC came down so hard on R&D tax relief claims
The crackdown exists for a reason, and it is worth being honest about it rather than just moaning about HMRC. For a few years, a swarm of no win no fee R&D claim firms sprang up promising inflated claims for almost anything a software company did. Building a normal web app became overcoming significant technical uncertainty. Standard integration work became novel architecture. Billions of pounds went out of the door on claims that would never have survived proper scrutiny, and a meaningful chunk of that was straightforward fraud.
HMRC's response has been to treat the whole population of claimants with far more suspicion, add a mandatory pre claim notification for many companies, require a named agent on every submission, and dramatically increase compliance checks. Genuine businesses are now paying the price for the behaviour of a small number of chancers and the firms that enabled them. That is not fair, but it is the reality, and pretending otherwise will not help you.
What genuinely counts as R&D in a software business
This is where most founders trip up, usually in one of two directions. Either they assume nothing they do counts because it is just building software, or they assume everything counts because it involved writing code. Both are wrong.
The actual test is whether you were seeking an advance in science or technology by resolving a genuine technological uncertainty, one that a competent professional could not have readily worked out in advance. Building a standard CRUD application with an off the shelf stack is not R&D, no matter how hard it was to build on a deadline. Working out how to make a multi tenant architecture handle a genuinely novel scaling problem, or solving a data synchronisation issue with no established pattern to follow, plausibly is.
The honest question to ask yourself is this. Could you have found the answer to your technical problem in a well written blog post, a Stack Overflow thread, or a vendor's documentation? If yes, that part of the work is not R&D, however clever it felt at eleven at night. If genuinely not, and you had to experiment, prototype, and fail your way to a solution, you are in R&D territory.
How to claim without painting a target on your back
A few things I actually do, having been through this properly with my own companies.
Use a specialist who understands software, not a generic accountant bolting R&D onto your year end return, and definitely not one of the volume claim firms that caused this mess. Ask them directly how many of their claims have gone to enquiry and what happened. A good adviser will have a real answer.
Keep contemporaneous technical notes as you go, not reconstructed six months later when the accountant asks for a technical narrative. A few lines in your project management tool describing the uncertainty you hit and how you resolved it is worth more than a beautifully written retrospective justification.
Be honest about what fraction of a project was genuinely uncertain versus normal engineering execution. Claiming the whole build is what gets claims rejected wholesale rather than just trimmed. I would rather submit a smaller, defensible claim than a larger one that falls apart the moment an inspector asks a follow up question.
Is it still worth claiming R&D tax relief?
Yes, if you are doing genuine technical work, which most serious software businesses are at some point. The relief is still real money, and for an early stage SaaS business it can be the difference between hiring your next developer this quarter or next year. I have written before about the numbers that actually matter when you are bootstrapping, and cash in the bank from a properly defensible R&D claim is about as good as SaaS metrics get.
What has changed is that you can no longer treat it as free money for a rubber stamp claim. Treat it the way HMRC now treats it, as a relief that has to be earned and evidenced, and it remains one of the better reasons the UK gives founders to build technically ambitious products here rather than somewhere else. Given everything else the UK tax system currently does to entrepreneurs, I will take the wins where I can find them.
It also fits a pattern I keep seeing in UK policy toward small tech businesses. A reasonable idea, undermined by an implementation that punishes the people doing the right thing to catch the people doing the wrong thing. I wrote about the same dynamic with IR35 a while back, and the parallels are hard to ignore.
