Search and Research modes
In Spotter 3, we are introducing Search and Research modes, which are complementary coexisting modes that share the same underlying toolset but deliver different experiences. Both modes utilize the same agentic tool calls— such as executing Python code, querying the data warehouse, or searching unstructured apps like Slack– but they differ fundamentally in their reasoning and prompt complexity.
Speed versus depth
- Search mode (speed)
-
Designed for high-frequency, quick insights. It uses a concise “thinking budget” to quickly translate your natural language query into validated insights. It’s your go-to for questions like “What was the revenue in Q3?”, where speed and a single, accurate analysis are the priorities.
- Research mode (depth)
-
Built for high-stakes, multi-layered investigations. It takes the same tool-calling capabilities but wraps them in a deeper “reasoning loop”. Instead of a single pass, it consumes a larger computational budget to perform a chain of analysis, iteratively querying, validating, and cross-referencing until it can solve broad prompts like “Diagnose the root cause of our Q3 revenue miss and suggest three recovery paths.”
It might help to think of it like this: If Search mode is a high-speed calculator that gives you the right sum instantly, Research mode is the mathematician who shows the entire proof, explores alternative theories, and double-checks for errors before speaking.
AI insights
Both Search and Research mode now surface summaries of your answer, specifically in Spotter 3. These summaries highlight the most important information found during your analysis, and include recommendations, suggestions for next questions, and opportunities for your business.
Research mode
Research mode introduces a process where Spotter reasons through a problem like a human analyst.
-
Extended planning: Unlike a single-pass search, Research mode creates a multi-step query plan to simplify complex questions. It identifies ambiguities, generates testable hypotheses, and allows users to modify the analytical path before execution.
-
Iterative self-correction: If an anomaly is detected, the agent autonomously deep-dives, self-corrects its logic, and refines subqueries until it reaches a grounded conclusion.
-
Automated narrative summary: The final output is a coherent report that explains the “why” and “what’s next,” citing specific data points and visualizations to build trust, structured in a detailed report.
See Research mode in action.
Key capabilities
| Feature | Search mode | Research mode |
|---|---|---|
Tool calls |
SQL, Python, App Search (Instant) |
SQL, Python, App Search (Iterative) |
Reasoning budget ("Thinking time") |
Low: Optimized to be fast |
High: Optimized to be comprehensive |
Prompt depth |
Single-dimensional (“What”) |
Multidimensional (“Why” and “what if”) |
Output type |
Direct answer or visual |
Comprehensive decision report |
