We've completely rebuilt our AI assistant from the ground up. The new agent thinks autonomously, handles complex multi-step questions, and can now answer the "why" behind your numbers.

✨ What's new?

• Autonomous reasoning – The agent now understands your question and intelligently selects the right tools to answer it. Ask multiple things in one sentence, and it handles them seamlessly.
• Real-time thinking visibility – Watch the agent's reasoning process as it works through your question, step by step. No more guessing what's happening behind the scenes.
• Smarter response formatting – Simple questions get simple answers. The agent now returns values as text when appropriate, and only generates tables for complex breakdowns.
• Dimension splits – Ask "Revenue by Country" or "OPEX by Department" and get instant breakdowns by any dimension in your model.
• Main contributors analysis – Ask questions like "Why was OPEX so high in February?" and the agent identifies the top drivers across your dimensions automatically.
• Flexible time periods – Split and compare data by month, quarter, or year. No longer limited to totals.
• Refreshed UI – A cleaner interface aligned with Abacum Intelligence branding.

⚙️ How does it work?

Simply open the AI assistant and ask your question naturally. The agent will:

1. Analyze what you're asking
2. Select the appropriate tools (visible in the thinking panel)
3. Execute multi-step queries if needed
4. Return results in the most useful format

For dimension-based queries (splits or contributor analysis), use the exact dimension name from your model for best results.

Example queries to try:

• "What was revenue in January 2025?"
• "Show me gross margin by region for Q1"
• "What were the main contributors to OPEX in March?"
• "Compare Q4 revenue to Q3 by month"

⚠️ Good to know

Clear the conversation to ensure the Agent uses the new framework

Composition vs. causation – The agent can break down what makes up a number (e.g. which departments contributed to OPEX), but does not yet compare to prior periods to explain why a number changed. True variance analysis is on our roadmap.

Dimension names – For dimension-based queries (splits or contributor analysis), use the exact dimension name from your model for best results. Fuzzy matching improvements are coming soon.

🔜 Coming soon

We're continuing to enhance the agent with additional table features (prior period comparisons, dimension filtering) and chart generation.

📚 Learn more