Kreoh: The Startup Making Tax Claims Less Taxing

An interview with Garry Tiscovschi, Managing Director, and Fareed Idris, CTO of Kreoh, by Joe Gorman NDRC Programme Manager.

Garry Tiscovschi, Managing Director, and Fareed Idris, CTO of Kreoh

Garry Tiscovschi and Fareed Idris radiate the sort of thousand-watt intelligence that you normally only see in fictional TV scientists and University Challenge contestants. They can quote Plato, write code in several programming languages, and reference research papers that still bear wet ink. 

Idris built an Android tool at 16 that scaled to tens of thousands of users. Later, while working on his masters in computer science in UCD, he built an edtech platform that was adopted by two universities. Tiscovschi speaks English, German, and Russian, and combines his day job leading Kreoh with a volunteer gig teaching kids at two coding clubs, one of which he established with Idris while still in secondary school. Their third co-founder, David McSharry, is a Stanford Machine Learning alignment theory scholar and theoretical physics graduate of Trinity, where he met Tiscovschi.

So, with the ability to build pretty much any technology product they choose, and the intellectual horsepower to power a small town, why are these three focusing on probably the most boring problem in the world?

Tiscovschi laughs: “Our motto is: motivated by monotony. The more paperwork reduced, the better. That’s what motivates us.”

Their startup, Kreoh, has morphed from an AI advisory business into what the founders call the future of specialist report writing. At present, they are a multi-agent AI engine for R&D tax consultants. They help the people who write tax credit applications do it five times faster, and with much greater accuracy. 

“The government prefers our AI-plus-human reports over reports written exclusively by humans,” explains Tiscovschi. “They’re spending half of the time on reviews and inquiries that they did before. Inquiries are often a major concern, especially in strict jurisdictions like the UK. There’s also significantly less back-and-forth with the research teams, so less customer friction for the tax consultants too.”

Specialist report writers bill per report. These fees can vary from tens of thousands to tens of millions, depending on the value and complexity of the claim. Tiscovschi tells me that the financial incentives of providing faster, more accurate reports have been crucial in speeding up AI adoption. 

This gives me some hope that processes could speed up elsewhere too. I’m drifting off into a vision of a tantalising future without tax claim paperwork, without paperwork in general, even. Rare birds are chirping in this dream. We all have hoverboards. How far away are we from fully automating a company’s reporting?

Idris is quick to assure me that they’re interested in enabling people rather than automating them into obscurity: “Currently, the tech simply isn't there to complete the entire report. We're currently tackling the technical writing piece of it. That’s only one piece of the entire puzzle. There is still a lot of tacit knowledge required to take these reports from 90% to 100%. We’re focused on cutting down the monotonous work in getting from zero to 90% - a strong first draft.”

Garry Tiscovschi (Kreoh MD) pitching to his fellow NDRC Accelerator Cohort, H2 2024

Kreoh Co-Founders Garry and Fareed in 1:1 Mentor Madness during NDRC Accelerator H2 2024

If you write technical documents, that’s a hell of an improvement to your quality of life. I resign myself to a world where AI doesn’t make paperwork disappear entirely, but makes report writers more efficient and probably a bit happier. 

Tiscovschi tells me about his “personal vendetta” with menial work: “I spent some time managing summer schools in Germany. I used to carry these big binders of paperwork back to head office. We had highly paid accountants going through them, line by line. Most of that work isn't what they were trained to do. To me, deleting billions of hours of menial work, that's that that gets me out of bed in the morning.” The consulting companies we’re supplying are spending a big chunk every day producing these reports. Only now they’re using our product, so they’re doing it much, much faster.”

How many of these consultants are there? “We just onboarded one of our largest clients,” Tiscovschi continues, “They have 140 consultants just working on R&D tax claims in the UK and Ireland. Some firms would have 250 just for the US and another 80 for the UK. It’s highly specific to jurisdiction, so there are a lot of them out there across the world.” He smiles: “and we charge per user, per month.”

Currently, there is a lot of untapped value in the market, which is highly labour-constrained. I ask both founders what it’s worth globally. “Less than half of the available £16.5 billion tax relief in the UK is claimed. In the US, the picture is more drastic, with more than $80 billion currently left on the table every year.” 

Now their mission makes more sense, even if the problem doesn’t immediately set the heart racing. Idris chimes in to tell me that Tiscovschi is “very much not the standard human being when it comes to being pumped up by deleting paperwork.” 

As CTO, Idris is more excited by the use of AI in this realm, a “clear cut, tangible application of the world’s most exciting new technology, rather than some ChatGPT wrapper.” Kreoh started life as an AI agency helping corporate clients to implement AI solutions. According to Idris, this gave them the chance to “get a lay of the land and see where the technology had the most potential and impact.” 

Following their noses on different processes, the Kreoh team noticed that AI is currently particularly useful for developing ‘engines’: tools that can optimise the speed and efficiency of your workflow in technical writing, professional services, financial services, and product development. “Our technology is an engine for people writing reports,” explains Idris, “but there are other engines, for IP audits, grant writing, or even developers building accessible tech products, like DevA11y.”

The pair are AI futurists, and I am beginning to see why they have started with this problem, after digging into the product that they’ve built. It’s not as boring as it sounds. Still - tax credits, engines to help you write technical documents or loan applications - these things are the now. What is the future?

Tiscovschi is a big fan of Yann LeCun, Meta’s Chief AI Scientist who said that “machine learning sucks” at the World AI Cannes Festival earlier this year, before telling a stunned audience that “the future of AI is non-generative. It works for text, doesn't work for anything else.” His Meta team is developing AI architectures to create machines that understand the world, and can plan, remember and reason. 

“Generative AI today is autoregressive - it looks back to look forward,” says Tiscovschi, making a noble attempt at a layman’s explanation. “A good writer observing a blacksmith might write the next chapter of what that blacksmith will do. But someone observing a blacksmith does not understand the subtleties and intricacies of her personal opinions, her model of the world, and her internal life. In terms of AI, that level of understanding is what needs to come next to make more exciting breakthroughs.”

The three young men behind Kreoh have been watching, listening, and learning for quite a while. Now they have started building. Theirs will be a journey worth following.

Kreoh will be pitching live at the upcoming NDRC Demo Day this September. Register your interest here to attend.