With AI to the next level of tax advisory
At Taxy.io we are working to reduce the research effort in tax consulting. We build up our innovative technology on Natural Language Processing (NLP) and Machine Learning (ML). This is intended to relieve the burden on consultants and provide them with technological support in synchronizing the two information silos of client and law and perspectively create a central digital hub for tax intelligence that can be used also by software. As CFO of Taxy.io, I wrote this article to give an insight into our plans and our development. It appeared first on steuerkoepfe.de.
Based on our own professional experience, we have found that it is very difficult for tax consultants to advise their clients comprehensively in the given time, as not all relevant information — both on the client side and on the tax law side — is available directly in most cases.
In such situations, we have often wondered why important aspects of our client situation were not proactively asked or why relevant new judgments, which we had recently read about, were not included in the consultation. Sometimes we ended up advising our tax advisor on how to advise us. The problem was that the client situation was becoming increasingly complex and that the legal framework was constantly evolving.
Sometimes we ended up advising our tax advisor on how to advise us. That motivated us to tackle this issue.
How much research do you actually charge for?
The tax consultant has to carry out a lot of manual research work to connect the two information silos — client and law — with each other and is barely technologically supported in the process. It is estimated that German tax consultants spend millions of working hours a year on research, but how many of these can they pass on to their clients?
Too much research, too little advice
A high percentage of manual work also encourages human error and reduces the time required for creative or strategic advice. After analyzing this problem through surveys, interviews and “internships” with tax consultancies, we considered what an ideal solution would be: a digital, easily accessible tax intelligence that understands his information needs and presents answers to his specific problem — just like a digital concierge — except that the questions are far more complicated than “Where do I find the next Italian?”. In the end, all economic operators will benefit, as faster, better and more proactive tax advice will have a positive impact on all transactions.
The solution: a digital, easily accessible tax intelligence that understands his information needs and presents answers to his specific problem
Our products are based on Natural Language Processing (NLP) and Machine Learning (ML), two very exciting elements that are associated with artificial intelligence. With NLP we teach software to understand written natural language (and legal texts are considered natural language here). We have so far processed 35,000 paragraphs, 45,000 judgments and 3,000 pages of administrative guidelines from the tax context and built a huge semantic body of text.
Thanks to the different types of documents, this results in a comprehensive view of tax regulation. In order that the finding of information does not function only keyword based, one needs trained NLP algorithms, which understand for example whether the number “three” is a deadline, a paragraph reference or a page number, or NLP algorithms, which recognize all formulation possibilities of exceptions (“does not apply if”, “unless…”, “does not apply if…”). It is not expedient to include a complete list of exceptions in the software, so we teach our software to recognize such and other legal elements independently.
We train our technology not only with the legal primary literature, but also on the basis of anonymous e-mail inquiries from clients, in order to find with a certain probability of success the relevant extracts of laws, judgments and guidelines that a tax consultant needs to answer these inquiries.
In the future, this will allow us to semi-automatically reply to client e-mails using prefabricated text modules, which we make available to our tax advisors at the click of a mouse. Based on this, our results are continuously improved through direct and indirect user feedback.
From meta search to meta understanding
Combining primary literature such as laws and law firm documents in a meta search engine is a very good first step. The question here is: Does my software know what I want to find? Or does it display all documents in which a certain search term appears? Then a larger database will lead to more, but not necessarily to better answers. We are therefore working on a semantic, context-related understanding of question and answer, based on textual documents of the legal framework literature and the client file as well as sufficient training.
Our tax advisors tell us that clients are increasingly expected to support them almost around the clock and to react quickly. At the same time, almost all tax consultants see digitalization as the greatest challenge for future success. But how exactly does digitization work? Our technology offers the opportunity to react appropriately to the changes that tax consulting is experiencing anyway (in addition to digitization, more complex legal situations, complicated client situations, lack of specialists, increasing cost and time pressure, etc.). With our technology we contribute to the digitization of tax consulting and make the industry more sustainable.
We contribute to the digitization of tax consulting and make the industry more sustainable.
With our products it will be possible in the future to extract the relevant information from the primary literature with much less search effort, to react proactively to changes in client and law and to answer client enquiries much faster and more accurately.
Outlook
Currently, our products are in the minimum viable product stage with a focus on German tax regulation and we are validating the first versions with pilot users. These include over 400 employees from small, medium-sized and large tax consultancies. Recently, there have also been very exciting approaches with insurance companies, publishing houses and financial software manufacturers in whose digital products we can integrate our intelligent solutions. We are always very open for further exchange with tax consultancies, publishing houses, companies from the finance and insurance sectors as well as software companies. We are also happy to get in touch with interested business angels and investors.
Currently we are still training our software on the existing corpus which is huge in the German tax context. It will still take a while until our software starts to become creative and develops its own design ideas, so that this will initially remain a proprietary territory of humans.
Until then, we will continue to work hard and with heart and soul on our vision of a central platform for digital tax intelligence, so that taxes become more manageable and fairer.
If you want to catapult tax advisory to the next level with us, contact info@taxy.io.
This story is published in The Startup, Medium’s largest entrepreneurship publication followed by + 377,008 people.
Subscribe to receive our top stories here.
