You run a small company, a software company. You’ve heard of R&D tax credits.
But do you qualify?
​
Maybe you have been experimenting with AI. That sounds like Research & Development, but is it? If it is, How do you write it up?
​
The tax code is complicated. (Some people think intentionally so.) Doing it yourself is difficult. True, specialist tax advisors exist, but perhaps you are too small for them to be interested.
​
Well, I’m here to help. I'm a serial entrepreneur. Over the past 20 years of filing R&D tax credit claims, I’ve claimed back over £1.5m. I’ve had some of my claims queried by HMRC and passed with flying colours.
​
Here are the filters to see if your software project qualifies:
1st Filter: Not routine
Research & Development (R&D) goes beyond the routine. [See Requirement 22.] Yes, some companies are practically 100% R&D, but most of these are VC-backed, who come with professional tax advisors. So for you, this is your first yardstick.
Think of it this way. Ideally:
-
You ran a separate project with a specific goal.
-
When the project launched, you might have been uncertain you would succeed.
-
More importantly, you were uncertain of how to approach the problem.
I use this comparison to clarify the point.
Let’s say you are creating a corporate website. Sure, there are some unknowns, but you know everything you want is possible, and you even know broadly how to do it. Yes, there is work involved. Mistakes may be made. Bugs may need to be fixed. Yes, there are a few places where there is a choice to be made. You may be able to think of 2 ways of implementing something, try the first but then decide the other way is better.
Everything I have just described is routine software development. There is nothing there that you would brag about over a beer with work colleagues.
Actually, that is not entirely true. That process of trial & error was innovation, but your R&D project exists only during the period of uncertainty. [See Requirement 33.] Meaning that in this project you were uncertain for maybe a couple of hours. Sure, you could claim for those hours, but this is a pointlessly small claim.
2nd Filter: Project-level uncertainty
You need to think in terms of R&D projects. [See Requirement 19.] The whole project should have some level of uncertainty. When you resolve this uncertainty, then the R&D is done. So, you'll need large uncertainties that take weeks/ months to resolve.
To clarify, this uncertainty has to be about the technical implementation ... what you do to achieve this goal. [See Requirement 9.]
The uncertainty cannot be: Not knowing whether anyone will buy your product. You have to take the next step. Which feature do you plan to get people to buy your product, and what technical uncertainty was there in implementing that feature.
Or, perhaps lowering the price of your product will attract customers. What do you propose to change to make your software cheaper to operate, and what technical uncertainty was there around implementing that change.
Additionally, your goal needs to be a permanent, appreciable improvement to your business. [See Requirement 23.] Cosmetic changes to packaging certainly don’t count. Understanding your market better via market research doesn’t count.
Success has to mean an improvement in your capabilities. [See Requirement 6.]
Think of it this way. At the end of the project you should have learned something. You don’t need to explain what you learned, but whatever it was, if you had known it at the start of the project it would have saved you weeks or months of effort. Well, those weeks/ months are your R&D claim.
3rd Filter: Competent professionals
Here is another brutal truth. Something is not innovation if you are simply figuring out what everyone else already knows… what is common knowledge to an average, competent software engineer. [See Requirement 14.]
Here are two questions to ask yourself:
-
Could you have hired someone for whom this would have been straightforward?
-
Could you have saved yourself a lot of time and effort if you had read a manual or a blog, or did a bit of research? [See Requirement 20.]
A good starting yardstick is: If you were interviewing for a programming job at another company, would this R&D project be something you would cite as an example of something impressive you accomplished? After all, you are not going to brag about a problem where the answer was pretty obvious.
4th Filter: Uncertainty vs Inventiveness
Please be aware: the above advice risks setting the bar too high.
To be blunt, the relevant tax regulations are vague and open to interpretation. Big companies get away with a lot because they have teams of lawyers and tax advisors to argue with HMRC. Micro companies (like you and me) have to jump higher hurdles because we do not have the time or energy to debate with HMRC.
So, yeah, we need to pass a higher bar, but it may not be as high as you think.
I have a patent in peer-to-peer architectures. I spent £150k on patent litigators looking to enforce my patent. Patents require an inventive step: a guess, an inference, a deduction that would not be obvious to a skilled person in your profession.
Well, R&D does NOT require inventiveness. It only requires uncertainty. What is the difference? Well, let me explain.
All of the approaches you take may be obvious things to try, but the important point is that you either didn’t know before:
-
if any of these approaches would work, or maybe
-
which approach was best.
​
You don’t need a “Eureka moment” of sudden inspiration. Instead, you need to have wasted time figuring it out.
Here is an excellent example straight for the tax regulations. [See Regulation 29.]
You can take two commercial software packages, both of them well documented. And, you can try to integrate them. If there are no discussions online on how to integrate them ... If there are multiple ways that you could integrate them but you do not know which way is best … Even if you are virtually positive that you will be able to integrate them to achieve what you want … then you were doing Research & Development.
Similarly, you know that a particular modelling technique works in one industry but it hasn’t been applied in your industry and there are multiple ways you could adapt the model to your circumstances. Then you are doing R&D. [See Regulation 23.]
Remember, you can only claim for the time and resources used to resolve the uncertainty. If the uncertainty is too small to have justified a separate project then (1) it is not worth claiming, and (2) your claim may fail anyway for not being an “appreciable improvement”.
So, you want to think of big areas of uncertainty that require a good amount of time and effort to resolve. And the uncertainty is around HOW to implement something, not the result: whether customers will like it, will use it, etc.
Experimenting with AI
Let’s say that you have been experimenting with Artificial Intelligence (AI) over the past year. Thinking of the above criteria, you likely have a valid R&D tax credit claim.
Let’s go step-by-step.
First, Not Routine.
AI was new. Chat GPT was launched at the end of November 2022. OpenAI launched its premium service in February 2023. More importantly, Large Language Models (LLMs) are an entirely new way of seeking to solve problems. So, unless you are an AI research company, it is highly likely that any experiments you made with AI were not routine.
Second, Project with a Goal.
If you were just playing around with AI with no desired result, then you were not doing R&D. But, if you had a goal and you were playing around with AI to achieve that goal – e.g. automate a role, reduce the cost of providing some service – then you had a distinct R&D project. Given how new AI is, it is very likely this project involved a lot of uncertainty.
Third, Competent Professional.
No one (not even OpenAI) knows exactly how AIs do what they do. The capabilities are emergent and unpredictable. Best practice with AI is extremely vague and does not provide a "roadmap" to any particular implementation.
A search for "prompt engineer" on Indeed returns no results. So, it is fair to say that the relevant “competent professional” is any competent computer programmer.
So, you had an AI project and a goal. If a regular computer programmer would not have immediately known how to use AI to achieve this goal? Would they have been able to easily deduce how to achieve this goal? [See Requirement 13.] No, then your project qualifies.
Let’s say that your AI experiments involved using a third-party AI chatbot, and the way to get that chatbot working for you was pretty clear. Then it is unlikely there was uncertainty. You would have been following “an established pattern” or following “routine methods”. [See Paragraph 29.]
However, if this tool was not well documented or the documentation applied to a different use case, and adapting the AI chatbot to your use case was not obvious and required some trial & error, then you did have an R&D project.
Bigger than that, if your project was to use AI to solve a problem and you had to experiment with a range of third-party AI chatbots, each of which required some adaptation, and you did not know if any of them would work, and if they did which was best, then that whole project was R&D.
Please be aware that merely using cutting-edge technology is not in itself cutting-edge. It is not R&D. [See Requirement 8.]
So, just because you used AI does not mean that you did R&D. You had to have learned something along the way. Something that would make it easier if you had to do it again.
Next Steps
So, now you have the answer to your first question: Did I do qualifying R&D?
As for the second question: How do I even apply for R&D tax credits … the answer is simple.
We have an AI chatbot that has been trained on the relevant regulations. It will further explain the regulations to you in terms you can understand. More importantly, it will translate your description of your innovation into terms that HMRC can understand.
The AI chatbot is patient, insightful, does not charge by the hour, and has a perfect memory. Visit it. Chat for a while. Sleep on it. Come back, and chat some more.
So, why not start today?