Flanders Investment and Trade (FIT) is a Flemish public agency that:
supports the international activities of Flemish companies
attracts foreign investors by highlighting Flanders’s central European location, its strongly developed infrastructure and its renowned innovative clusters
FIT gives Flemish and foreign companies free advice and introduces them to its global network of experts in more than 90 offices around the world. The quality of service and advice provided to these companies is therefore essential.
Investing in AI to gain competitive advantage
Internationalisation is a big challenge for Flemish companies. This is in part due to the very competitive nature of the environment these companies grow in. Flanders’ innovative ecosystems are competing with all their counterparts around the globe, and this is why operational excellence is required for FIT to put Flanders’ in a pole position to expand abroad, and also to attract foreign investors.
To create operational excellence, the agency has been investing a lot in IT and artificial intelligence (AI) in the last few years, as it became a key part of its strategic plans.
Along the way, some questions naturally arose. How to implement AI? Which ideas should be pursued? Which technologies and applications specifically serve the agency’s ambition to become a data-driven organisation and to set an example for the public sector?
Radix, through its Fast Discovery framework, devised tangible areas and use cases where AI could help FIT by increasing efficiency and creating value for FIT’s clients and employees. The specific case at hand: AI-assisted question answering.
One of FIT’s main goals is to give relevant, reliable and actionable information and advice to Flemish and foreign companies. Many foreign companies directly contact FIT to gain insights into the Flemish market. Flemish companies planning to expand internationally also directly ask FIT for counselling on markets abroad.
These contacts mostly come in the form of direct questions to FIT. Every month, FIT offices around the world receive thousands of questions. A lot of the value FIT creates comes from answering these questions. Every answer needs to be accurate, complete and actionable. It takes FIT’s advisors a considerable amount of time to collect the needed intelligence and research in order to answer these questions.
What can be improved with AI?
As the information needs to be sent relatively fast, busy periods can often create bottlenecks, which cause some answers to be sent later than foreseen. Some complex questions or cases demand more research and more time to be answered. External events also have a big impact on operations. The COVID-19 pandemic led to a huge spike in requests relative to the numerous uncertainties worldwide relative to doing business.
Naturally, a certain amount of questions tend to be quite similar. FIT’s advisors sometimes have to answer some questions repetitively, which is not the most enjoyable part of the job. This is why FIT wanted to optimise this high added-value, but highly time-consuming process. The agency wanted to get accurate, actionable information to clients faster, allowing for more time to take on complex cases and helping the advisors focus on more rewarding parts of their jobs. These frequently asked questions can also be gathered and distributed through a broader audience on FIT’s owned channels such as website or social media.
The AI solution
Radix proposed a solution to assist advisors on their question answering tasks. Through AI and machine learning, Radix was to find a way to extract intelligent insights out of FIT’s data, allowing the agency to answer questions more easily and quickly. The entire question-answering process would then be optimised, creating value for FIT’s clients and empowering its employees.
Radix started to work on a minimum viable product (MVP) that could be made quickly to get direct feedback from the end-users: FIT’s advisors. Radix used Natural Language Processing (NLP) to create a question-answering machine learning model.
How does the model work?
NLP is a field of AI and machine learning used to process and analyze data from the human language. The model created by Radix selects valuable information out of the content of the question, screens similar questions answered in the past and automatically provides advisors with the most suited answer: they don’t have to go from memory or do long research to find these answers. The model automatically updates itself with every new question and answer.
When the first iterations of the model were submitted to FIT advisors for evaluation and feedback, interviews showed that advisors were excited to help work on these features. They believed this kind of technology could help them and showed great enthusiasm around AI. 100% of the advisors were open to innovation as they saw the potential of the concept. After the first sprint, 78% found value in the tool and some advisors brought up interesting features which were ultimately added to the solution.
Delivering impact with AI
After just a few weeks, Radix delivered the MVP. The solution showed great results in several key areas:
Saving time and money. The solution saves 27% of time spent on cases. The AI provided help for 60% of all tested cases (cases where no previous answer/data was available included). Ultimately, FIT’s advisors were able to help 36% more clients with Radix’s solution.
User satisfaction. The solution gives FIT advisors superpowers. It allows them to save time by providing them with relevant answers that they can adapt if need be. This process optimisation saves them a lot of research time and allows them to focus on more complex cases and other parts of their jobs.
Continuity optimisation. The solution is also of great help to guide new advisors or when an advisor is unavailable due to sick leave or vacation. As it solves more and more tasks, the solution becomes a knowledge database, gathering knowledge from hundreds of advisors and cases.
Based on results and learnings from the MVP, Radix also proposed ideas and areas to further improve the solution in the future, optimising performance by incrementally improving what is under the hood. There have already been a few suggestions for the next iterations and a further improvement in problem-solving. The next step is to implement the solution globally.