ECB enhances Corporate Telephone Survey with AI tools
The European Central Bank is deploying artificial intelligence to improve the efficiency and analytical capabilities of its Corporate Telephone Survey. This initiative aims to streamline data collection and enhance economic analysis.
Automating insights from corporate interviews
The European Central Bank is modernising its Corporate Telephone Survey (CTS) workflow by integrating large language models (LLMs).
With participant permission, speech-to-text AI creates meeting transcripts, which are then redacted for confidentiality and summarised by LLMs.
Initial experiments with off-the-shelf chatbots showed modest gains, leading the ECB to develop its own bespoke LLM tool.
Early results indicate AI-generated summaries are of similar quality to human-written ones, often requiring limited refinement.
This has already saved an estimated 25-31 person-hours in recent survey rounds, also reducing information loss or misinterpretation.
The ECB is now testing LLMs to summarise individual interview reports for overall summary drafting.
Unlocking deeper economic understanding
Beyond operational efficiency, AI expands the CTS's analytical possibilities.
AI-powered agents can automate scheduling and summarise external data like news and company websites, providing economists with crucial background before interviews.
The rich textual data from the CTS also enables new analytical avenues.
A chatbot application is under development to answer ad hoc queries, such as corporate reactions to trade tensions, enhancing insight extraction speed.
The extensive CTS text dataset, with human-written summaries since mid-2007, will also be used to create quantitative indicators of economic sentiment or uncertainty, similar to recent research on corporate earnings calls.
AI as a central bank co-pilot
AI integration into the Corporate Telephone Survey offers clear efficiency gains for central bank research.
Yet, human expertise is paramount for validating AI outputs and ensuring nuanced interpretation, maintaining data integrity.
This human-machine collaboration promises enhanced analytical capabilities, provided robust safeguards and continuous staff feedback are diligently applied.