• MS Co-Pilot and other general prompting approaches: While Microsoft Copilot and ChatGPT excel as versatile everyday tools, they fall short in addressing the nuanced and specialised demands of composing R&D tax credit reports. These require Tax Claim aware AI guardrails, a deep understanding of guidelines and regulations, as well as the specific context and substance of your client’s R&D projects.
  
  
• Kreoh’s R&D Engine approach: Kreoh’s platform - the R&D Engine - is purpose-built for R&D tax claims, using multiple, specialised AI agents that understand your clients R&D context, follow your jurisdictions regulations, adapt to your firm's style guides, process various data types, and function as dedicated tools to significantly assist R&D professionals reducing enquiry rates and efficiently drafting compliant reports.
  

• Kreoh’s R&D Engine is an advanced AI platform meticulously engineered for R&D tax professionals, distinct from general-purpose AI tools. Our engine employs multiple synergistic AI agents capable of utilising document editors, online search engines, and other specialised tools on your behalf.
• The R&D Engine augments the workflow of innovation consultants, empowering your team to write faster, research deeper and raise quality to the level of your best writer.
• To date, we have assisted advisors throughout Europe and Asia to generate over 26,000 technical narratives.
• Expanding horizons to make Kreoh truly global, we are currently working on solutions for the Americas, Oceania and Africa. Based in these regions? Please reach out to us today to see how we can help.
  

The R&D Engine is easy to learn and easy to use. Our platform's layout and navigation have been designed with intuitive, user-centric workflows. This design has been refined over time through collaboration with existing partners and our own trusted in-house R&D experts. We consistently receive feedback from our users on how intuitive our software is to use, often describing it as an indispensable part of their daily toolkit.

• Kreoh’s R&D Engine offers significant configuration, allowing each firm to control the behaviour and appearance of their engine. The platform has in-built options, settings and parameters, allowing you to form it into a tool that suits your writing styles.
• Every partner benefits from a comprehensive, onboarding programme. During this, we gather feedback in order to capture your firm’s voice and imprint this within your engine. This bespoke approach ensures we establish quality standards precisely aligned with your requirements.
• Post-onboarding, our system continues to leverage your defined expertise to generate reports that uphold your quality benchmarks and voice preferences. The platform can be specifically "grounded" to employ particular phraseology or stylistic conventions, ensuring outputs reflect your company's established standards.

• Yes, our hard data tenancy architecture facilitates extensive customisation for bespoke functionality and seamless workflow integration. Kreoh can also be fully white-labelled according to your specifications.
• For larger enterprises who have unique requirements, we offer an engine customisation service to change the software’s behaviour and add functionality beyond what the standard configuration options allow.

Kreoh is a Collaborative Generative AI and Software Innovation Partner, offering sustained customer support and training. Our support structure includes scheduled face-to-face training sessions during onboarding, regular contact with a dedicated Customer Success Manager, and ad hoc support provided according to our published Service Level Agreements (SLAs). We maintain responsiveness to queries and requests, ensuring users can effectively leverage the full capabilities of the platform.

• Kreoh’s GenAI story started in 2022, where the founders built software with RNN AI models. In March 2022, RNN experimentation moved to Transformer Models, with which Kreoh built its first B2B products. Kreoh harnessed the potential of fledgling GenAI technology into a useful package, which was first released to early customers in September 2022.
• In 2023, one of our co-founders, David McSharry, became a Stanford Machine Language Alignment Theory Scholar. His learnings are being applied throughout all of our products. He has also made significant contributions to AI safety, including founding Ireland's first AI safety fundamentals course and EA community, working with the Centre for AI safety (CAIS).
• This just covers the surface level, as our team of Engineers are developing new AI capabilities on a daily basis, adding new skillsets to their already impressive C.V.s, degrees and PhDs.

• Kreoh’s experience in software development stretches back even further than its expertise in GenAI. This is epitomised by our people.
• Fareed developed and launched his first successful app to 20,000+ users at the age of 16, and has not looked back since. .
• Garry has been making software for friends ever since age 12, founding his home town's first tech club (still ongoing) with Fareed at 15. Since then Garry worked teaching Data Science workshops at Trinity College Dublin and in Release Management for Fraud & Loss in the payments industry.
• David, Fareed and Garry have also been named twice in Business Post Ireland’s 30 under 30, making them a formidable trio.
• Ben brings cloud, SaaS, AI and enterprise solutions experience from VMware, IBM and several high growth start-ups.
• This doesn’t even cover the Engineering Team, which consists of a wide range of software skillsets, meaning that your products are built and maintained to the highest standards.

• Kreoh created the first and only GenAI powered writing assistant available on the market in April 2023. From its conception, it was developed with core input by R&D experts, including Andrew Blackstock, who joined Kreoh in March 2025 as Senior Product Manager. Andrew brings a wealth of R&D experience, from being a materials Researcher specialising in novel cements, to advising over 400 businesses on R&D tax claims throughout the UK and Ireland and leading a team of R&D advisors.
• Further to Andrew’s expertise in all things R&D, Kreoh’s chief strategy officer, Brian Williamson founded one of the UK’s largest and most well known R&D Tax Advisories, Jumpstart (now ABGI). Brian has a significant understanding of the R&D tax landscape within the UK, Ireland and around the world. This is our in-house knowledge base. However, we work closely with our users to gain genuine “at the coalface” knowledge on R&D tax credits in your jurisdiction, meaning we can mould the engine to your particular needs.

• Kreoh’s R&D engine was designed primarily to augment the workflow of innovation consultants. Kreoh’s R&D Engine delivers comprehensive technical narrative support, encompassing:
  
  • Highly accurate project descriptions with precise,
  • Traceable references to source material;
  • Technical narrative drafting and review featuring customisable processes;
  • Technical project assessment via our integrated screening tool;
  • Technical interview support complete with evaluation and recommendation capabilities; and
  • Thorough baseline research across controlled data sources, providing full evidence and reasoning.
• New in 2025, your accountancy team can now optionally benefit from streamlined Additional Information Form (AIF) submission capabilities, reducing.
• Beyond core dedicated functionality, the tool can be used for prospect meeting preparation, allowing your business development team to query technologies, competent professionals and latest industry trends.

• Our platform users have remarked, "editing is nowhere near as easy as it is with Kreoh". Kreoh’s R&D Engine presents a unique user interface offering three distinct working methods:
  
  • A proprietary auto-drafting feature within an editor environment akin to familiar applications like Microsoft Word and Google Docs. This facilitates the composition of content on your chosen topic with one hold of a button. Consider it the ultimate sentence completion tool—one of Kreoh's notable early features.
  • A streamlined process enabling the generation of an excellent starting point for each report section with a single click, which can then be copied into the platform's conveniently located editor for further iteration.
  • An interactive, conversational interface known as "Kreoh Edit mode" for users preferring direct, dialogue-style instructions to meticulously craft the draft.
    

• The Engine enables users to prepare for and analyse technical interviews directly within the system. The integrated Screening Agent evaluates interview quality post-session, offering insightful feedback including:
  
  • Insight on data quality
  • Follow up questions for you to ask clients
  • Red flags that may disqualify a project.
• Kreoh can process technical information meetings in full, meaning there is no need to manually segregate information. Unlike other AI tools, the Engine can take single or multiple recordings and segregate them accurately into the relevant projects. This can be done either through automatic extraction or guided extraction for “back and forth” discussions covering multiple projects.
• All major file formats are supported, including the ability to process handwritten notes (depending on legibility!) and standardising unstructured financial data. The Ask tab facilitates verification of information accuracy before incorporation into draft narratives. An upcoming enhancement will permit Kreoh to join technical meetings (e.g., Microsoft Teams calls) and furnish real-time follow-up questions and red flags to the interviewer, aiding with extracting qualifying R&D information more effectively.

Users can conduct fully controlled and directed searches across multiple data sources. Kreoh surfaces both supporting evidence and potentially conflicting results requiring further investigation. Within seconds, the research engine can furnish comprehensive evidence and reasoning for the selection of specific results, alongside granular control over how each result should be integrated into your technical narrative. This not only improves technical narrative quality and compliance with jurisdictional guidelines, but significantly reduces time taken to verify or query industry information provided by your clients.

Kreoh offers an optional module that allows dedicated users within your team to submit Additional Information Forms (AIF) to HMRC. Uniquely, this interactive autonomous AI Agent utilises a user account within your agent services account within your organisation, maintaining compliance with your HMRC Agency terms and conditions.

Kreoh can integrate seamlessly with your existing KMS solutions such as SharePoint, DropBox, etc. Direct platform integrations for financial data systems (e.g., Quickbooks, Xero) are available upon request.

• Kreoh has spent the last two years building robust guardrails throughout the platform to ensure strict adherence to the ground truth presented in source materials. The accuracy of our output is underpinned by input from seasoned advisors, and references to source material are readily available.
• Our architecture is explicitly designed to prevent the generation of unsupported information ("hallucinations"). Unlike other AI tools, if insufficient information is present in the source material, the engine will simply REFUSE to give an output, requesting more data from the user.
• The platform never accesses the internet EXCEPT when explicitly directed by the user, and incorporates controls over permitted sources and their usage. Significant development over 2.5 years, in collaboration with existing clients, has yielded highly effective anti-hallucination guardrails—a core strength of Kreoh—ensuring responses are consistently accurate and fully grounded in the provided data.

Good AI applications require a collaborative “Human + AI” approach. All outputs generated by Kreoh, including narratives and reviews, should be interrogated and reviewed by a human expert at all stages in the process. The platform is designed to augment, not replace, professional judgment. While Kreoh significantly enhances capabilities, final accountability always resides with the professional advisor.

• Kreoh’s R&D Engine is designed to think AND do. In order to do this, it must undertake dynamic reasoning, thinking and contextualisation. This is achieved through Kreoh’s proprietary engineering methodology.
• The Knowledge Manager Agent allows users to assign categories to uploads, meaning that they can add context documents, such as previous reports, literature reviews and so on, as well as recordings and notes between you, the advisor and your client. This means it absorbs the same information that you, as a technical advisor would also have to absorb, understand and reason with against your local jurisdiction’s legislation.
• Furthermore, the Drafting agent actually provides the user with the ability to view and interact with the engine, highlighting where in the source material it got the information from to produce any selection of wording, and the reasoning of why it interpreted it the way it did. If you don’t like how it interpreted something, you can tell it how to interpret it better and it will apply your logic.

Kreoh’s R&D Engine allows users to prepare technical narratives, interrogate costings and submit AIF forms. However, it is designed to support HUMAN tax advisors in their process, rather than replace them. It assists users by expanding thinking, researching deeper and wider, and drafting faster. This allows you to spend more time focusing on the client, and improving your service to them.

• No. On the contrary, Kreoh provides advisors with insights into how to extract the best information from their client, and gently teaches them how to write like your best writer, without taking time away from your best writer.
• Feedback from users has included phrases like “it helps me think outside the box” and “the Engine thinks in ways and angles I hadn’t considered before”. We believe the use of AI is an Engine, rather than a crutch, powering you to where you want to get to faster.

• No. Enterprise-grade security is fundamental to the Engine’s architecture. Critically, client data is never utilised for training any Large Language Models (LLMs).
• Unlike other offerings on the market, we employ a stringent 'hard data tenancy' model, ensuring complete isolation of your data—including databases, storage, and AI model access—from all other clients within a dedicated, secure deployment. This approach guarantees strict data protection practices, verifiable data residency controls for regulatory adherence, and the capacity to implement client-specific security protocols.
• Data is stored in accordance with SOC2 and ISO standards. Kreoh holds SOC II certification, ISO 27001 accreditation, and is fully GDPR compliant, adhering to the highest data protection standards. This architectural approach means your proprietary approach and knowledge is never transferred to other users, nor do you inherit the approach of our other partners. Further details are available at www.kreoh.com/privacy

You do. You maintain full Intellectual Property (IP) ownership over both the project data supplied and the AI-generated outputs produced through the Engine.

• Kreoh has been purposefully designed with regulatory compliance at its core. Its capacity to accurately reflect official guidelines (such as those from DSIT), utilise professional language derived from authoritative documentation, and uphold rigorous data security standards (SOC II, ISO 27001, GDPR) directly supports your compliance efforts.
• The hard data tenancy model satisfies stringent compliance requirements concerning data isolation and residency for technical compliance. The engine also provides reasoning, alongside references to source material to trace what legislation or guidance it has interpreted. Nonetheless, ultimate responsibility for ensuring that final outputs meet all pertinent legal and regulatory standards remains with the professional advisor.

• Yes. Kreoh has been designed with input from people like Andrew, who have first hand experience of preparing and submitting claims to HMRC. Our design ethos is to maximise your ability to extract useful and pertinent information from your clients. The R&D Engine has been used to produce thousands of technical reports across multiple jurisdictions with high success levels.
• As your R&D Engine is grounded on the best report examples produced by your own company, when used correctly, it raises the quality of outputs of your wider team, to a level similar to your best R&D tax consultant.

• Drawing from inputs from our own internal experts in the industry, Kreoh has worked hard over the last 2 years to implement proprietary engineering methodologies and guardrails throughout the R&D Engine. These are purposefully built to highlight a lack of qualifying projects when asked to draft from incorrect input data. This not only makes technical narratives compliant, but it also provides suggestions on how to fill information gaps. The robust engineer methodologies help mitigate AI hallucinations and most importantly, generate red flags when a project exhibits low likelihood of eligibility.
• Our Engineering Team works daily to continually improve these guardrails, ensuring relevant and timely compliance with local jurisdictional legislation.

• Kreoh’s R&D Engine has a dedicated Live Screening Agent which, should you desire, reviews your conversations with clients and provides pertinent in-call feedback on how the project aligns with your jurisdiction’s regulations. Where it picks up wording, context or suggestions that are contrary to the legislation or identifies activities such as routine improvements, it will produce red flags, suggest follow on questions, and provide you with reasoning on why it believes this is the case.
• Additionally, it will review your conversation to see if there are any information gaps or claims made by the client, and provide suggested follow-up questions, in real time, for extracting this information from your client. This therefore not only drives up the quality of information included in your report, but it also educates users for their next interviews, improving quality over time.
• The Research agent, which is linked to the baseline drafting section, allows you to do in-depth research within seconds. This then raises results that crossover with your project, scoring the conflict as well as providing a breakdown of the value of the result for your baseline section. This further reduces invalid claims where the technological solutions already exist.
• Within the Drafting Agent, if it does not have enough information to provide a legitimate technical narrative, it will prompt you accordingly. It will flag areas that would need to be addressed in order to be able to produce a report, meaning that you have ample opportunity to gather the relevant and correct information.

We undertake training with our clients to ensure that it is used correctly, and with the guardrails described above, purposefully make it awkward to produce invalid claims. How humans use Kreoh is ultimately up to the individual, and so Kreoh cannot completely prevent misuse. As mentioned previously, the R&D Engine is aimed at encouraging users to generate high quality, legitimate R&D tax claims.

• Kreoh’s R&D Engine has been designed to facilitate all claim types, from short and obvious R&D claims, to the most complex, lengthy project write ups.
• Regardless of project complexity, the R&D Engine’s approach is the same - review eligibility against the local jurisdiction’s legislation, and let the user know
  
  1. how good the information is,
  2. what additional information is needed to make it better, and
  3. if there are any red flags that suggest it wouldn’t qualify.
• This gives each user a fair chance to create the best narrative to support the R&D tax claim. This means that it judges objectively, whilst not requiring black and white answers to allow you to progress. Where a project fits within the grey area, it will assist your reasoning in determining eligibility, or ineligibility. Naturally, we like to err on the side of anti non-qualifying projects as a baseline, limiting the ability to accidentally create non-qualifying, convincing sounding reports.
• This raises an important point in that the engine will only ever assist a human tax advisor, and not make the decision for you, as we believe human review is paramount for true eligibility within these complex R&D schemes.

Your unique instance of Kreoh’s R&D Engine is grounded on examples you or your firm provide to the system for tuning This information is NEVER used to improve any other instances of the R&D Engine. It will mimic your firm’s writing style, and so if your examples provide obscure wording that downplay project weaknesses, it will too. However, between the screening agent, research agent and drafting agent, it is likely to pick up on this, raise red flags, suggest the project isn’t novel and suggest that the project sounds routine.

Kreoh met with HMRC in London, showcasing the R&D Engine in the process. However, similar to not rubber-stamping R&D tax advisors, HMRC does not approve or disapprove of the use of Kreoh’s R&D Engine. Not long after this positive meeting took place, HMRC publicly suggested they are looking to implement AI within their own processes. Furthermore, HMRC have said that they have no issue with people using AI to create claims, as long as they pass HMRC scrutiny.

As mentioned previously, anecdotally, Kreoh’s users have witnessed a reduction in enquiry rates since implementing our engine. Whilst we cannot claim sole credit for this, we believe the improvement to quality and freeing up of time to do deeper due-diligence with clients has, by design, improved the quality of R&D tax claims.

It depends what information you feed it (pun fully intended). Presenting it with the truth, and a transcript of an honest conversation, then no, the RDE will rightfully call your project out as not being qualifying and will put this answer within the draft. By putting the source information through the screening agent, it will identify red flags to you, and score the quality of the project as low. See what the R&D Engine said when attempting to create a claim for the development of a Beetroot Noodle Recipe:

Kreoh Limited - Beetroot Noodles Final Report

• The Screening Agent will suggest follow up questions for you to ask your client in order to tease out if there is R&D or not. Once you have the answers to the questions and feed the knowledge manager with the updated information, it will either:
  
  • Raise definitive red flags which at that point it is your choice to:
    
    • Rule out the project and not proceed (recommended),
    • Continue manually (at user discretion)
  • Help improve your narrative and ensure that you have considered and answered the core questions related to R&D tax.