Yes!

In this series of blog posts, we share our experience using generative AI to enhance productivity. We will highlight practical use cases for macro-economists, sovereign risk analysts, and AML country-risk specialists. We will also discuss where we expected strong results but were disappointed — including what went wrong and the factors behind the failure. We welcome your input: share your success stories as well as challenges.

Monthly Sovereign Rating Change Monitoring

At the beginning of each month, we review sovereign rating actions by major rating agencies (e.g. Fitch, Moody’s, S&P) and compare them to our quantitative, fully automated CountryRisk.io shadow rating range. We then analyse the key drivers behind these changes. While valuable, this process is time-consuming — especially in months with many updates.

For the October review, we tested our CountryRisk.io data management infrastructure and AI tools to support the workflow and make future reviews more efficient.

Here is how we approached it.

Rating Actions Data from CountryData.io

We used our purpose-built data infrastructure to extract all October rating actions via a simple API query. This produced the full set of sovereign actions in our coverage universe — 27 actions across diverse countries and rating categories.

We also asked our AI tools with web and news-search to identify rating actions for October. The results were unsatisfactory — which, in truth, we expected. Whenever we use AI, we first ask:

• Does the exact information we need exist publicly?
• How likely is the AI to locate complete and accurate data?

Sovereign risk analysis is niche. Identifying rating actions requires specific constraints and decisions:

• Rating agencies: Focus on the major firms
• Category: Foreign-currency long-term issuer ratings only
• Period: October
• Issuer type: Sovereigns only (no sub-sovereigns, MDBs, or supranationals)

Lesson: When specificity matters, deterministic processes outperform probabilistic ones. Provide the model with high-quality, structured input — don’t force it to guess.

Search for Relevant Information for Each Rating Action

The next step is to search for context around each rating action — typically by visiting websites, reading releases, and taking notes. Doing this manually 27 times would likely take an analyst 2–3 hours. Using CountryRisk.io AI tools, it took 7 minutes.

Below is the prompt we used inside the CountryRisk.io AI Assistant.

The Assistant leverages specialised tools such as web and news search. It worked through the list systematically. For each rating action, the orchestration engine triggered the search tool, retrieved the relevant sources, and summarised the content.

As you can see in the screenshot. It found relevant content on the web for the Barbados S&P rating upgrade. And then continued to the next on the list Botswana where it found several relevant news or content items.

Users can hover over each link to preview a source and jump to the original article. The AI completed all 27 items this way.

Summarize the Rating Actions

Finally, the Assistant produced concise summaries for each sovereign. Job done!

Go the Extra Mile

We then instructed the AI to consolidate all summaries into a table.

Next, we asked it to extract cross-cutting themes from the 27 rating actions.

I am confident this would have taken far longer to complete manually — and with no meaningful improvement in quality.

Make It Repeatable: Prompt Library

To streamline future cycles, we asked the AI to generate a markdown file documenting the full workflow — research, summary, and thematic extraction — for inclusion in our prompt library.

We pasted the prompt into the library, made minor edits (adding tool names and descriptions), and saved it as “Rating Actions Commentator.”

Next month, we simply select the saved prompt, paste in the new rating list, and run the workflow. Done.

Outlook

As shown in the screenshots, we have already deployed many workflow assistants and continue to expand our library of tools. Future posts will highlight additional workflows such as automated report generation and deep-dive research.

In short, we believe domain-specific AI platforms deliver real value when they combine:

• Robust data and content infrastructure
• Tailored prompts and workflow tooling
• Intuitive interfaces
• Strong evaluation and feedback systems

Exciting times ahead.