IBFD, the International Bureau of Fiscal Documentation, is – in their own words – the world’s foremost authority on cross-border taxation and the home of international taxation. They manage tax matters for companies, individuals and academic researchers, and offer a wide range of research services and trainings. To maintain this leading position, innovation is necessary to apply the most pertinent information and best online data interactions to tax research. IBFD saw opportunities in their unique text data collection, and was interested in taking this further. With plenty of experience guiding companies to make data work better for them, Aurai was thrilled to help IBFD continue innovating by elevating their data science and applications.
Data Application Is Data Education
Aurai specializes in making data actionable and valuable for each and every partner. Each organization requires its own unique approach, a clear roadmap navigating towards the most important ambitions. We had to understand the organization by getting to know its people.
IBFD is a traditional organization where few people had experience with or affinity for data (science). Also, the content of the tax documentation is very exact, mistakes should be avoided as the interests of many organizations, companies and governments are involved. It takes years of study and experience before the subject experts are competent in their field, and therefore there was some skepticism about whether algorithms can duplicate that expertise. In short, there was no opposition within IBFD to data (science), but it did take a bit of persuasion to get everyone on board the data train.
To reach IBFD’s own goals, it was important to make the process gradual. The transition was made smoother by slowly introducing to IBFD the many possibilities. As IBFD became better acquainted with the advantages of data science, we began to introduce the concepts more broadly in the organization. A global game plan was drawn up to help IBFD become a data-driven company.
Global Game Plan to Become Data-Driven:
1. The team begins with one data scientist. The data scientist uses the most relevant current techniques to examine IBFD’s data collection to articulate where the best opportunities exist for data-driven work.
2. Select the projects with the most potential for accessibility and productivity, start and complete those first projects. This way, the foundation for data science is laid within IBFD and interaction with data builds, which increases trust and support.
3. When the first projects are completed, involve the organization more broadly. This is a good time to build out the team. It is important to be visible and approachable. The employees of IBFD should know that they can contact IBFD’s data science team with (almost) all data related problems. As awareness increases, so does the number of projects.
4. Support is growing, and there is a noticeable difference. It is time to tackle some of the larger, more difficult projects.
A plan is good for preparation, but the real task requires putting prototypes into practice.
Showing How the Data Works
Based on the plan, the team began with just one data scientist who initiated the investigation. They had to classify the type, quality and source of the organization’s data and then match the information with the best techniques and strategies. Based on this research, we determined that IBFD’s unique data collection could create many exciting opportunities across an array of applications.
After reviewing several different directions, our first project aimed to improve the platform’s taxonomy (the structure and classification of documents) with Topic Modeling techniques. After the successful completion of the first project, the results from the research, and the newly enhanced document taxonomy, guided us towards the next projects. The team doubled by the addition of another data scientist, which increased the potential impact and possibilities with the simultaneous development of multiple projects.
Providing Immediate Value
We worked on projects such as finding and tagging economic activities in tax documents with Named Entity Recognition (NER) to improve the search options on the TRP, developing a technique to collect similar news articles with word2vec and doc2vec, and create Extractive Question Answering to quickly and effectively answer customers questions on the TRP, all with models based on transformers. These projects were all fairly easy to explain, and, more importantly, provided immediate value. Now that we had created the models and projects, we needed to integrate them alongside the organization.
Informing and motivating a large traditional company must be approached methodically and patiently. As more people understand the advantages of data science, the more receptive they are towards implementing them. We created interactive workshops open to all IBFD departments and teams to discuss the most relevant and current data science developments. We used many of our initial projects for the company as examples to make the lessons relevant and familiarize the employees with the improved data processes and structures. Contributions to the weekly newsletter also helped increase our visibility; awareness of the new data applications kept growing. As more people got to know our team and the data science field, employees began approaching us with discussions of wider deployment of data science across the company. Every day brought opportunities for new data science projects.
Using Data as Building Blocks
In the end, data science became a highly regarded and exciting tool for IBFD. By researching the organization and the possibilities, immediately demonstrating the value with initial projects, and engaging the organization, we improved technology and operations through education and enthusiasm. IBFD’s spirit of cooperation led them to their goals and brought together a diverse international company driven by durable knowledge systems.