All Posts

GetDot on DABStep: 450+ financial analysis tasks from Adyen

by Rick Radewagen1 min read

DABStep is a benchmark by Adyen and Hugging Face: 450+ multi-step financial analysis tasks. We ran GetDot on it.

You can train GetDot to get better.

Add instructions, examples, and business rules with GetDot's context agent.

Hours of analysis in seconds.

GetDot vs trained human analysts at Adyen on standardized tasks.3, 4

1 DABStep · 450+ multi-step financial analysis tasks by Adyen & HuggingFace.
2 Google Data Science Agent · ~52% weighted overall (Nov 2025).
3 Human baseline from DABStep: 62% accuracy with 3+ hrs of work on easy tasks.
4 GetDot's scores: GetDot 0.3 on the public DABstep leaderboard.

You can train GetDot to get better

Out of the box, with no domain context, GetDot scored 67%. Then we gave it context with its context agent, one step at a time:

  • 1 hour: 73%. A data dictionary and the table relationships.
  • 2 hours: 80%. Golden query examples.
  • 4 hours: 85%. Business logic and domain rules.
  • 8 hours: 90%. Instructions for edge cases.

For comparison, Google's Data Science Agent scores about 52%.

Hours of analysis in seconds

On the easy tasks, trained human analysts at Adyen got 62% right, with more than three hours of work. GetDot got 96% right in about 60 seconds.

On the hard tasks, GetDot got 89% right in about 130 seconds. There is no human baseline for those: they need domain expertise across several files.

Our scores are on the public DABStep leaderboard, as GetDot 0.3.

Rick Radewagen

Rick is a co-founder of Dot, on a mission to make data accessible to everyone. When he's not building AI-powered analytics, you'll find him obsessing over well-arranged pixels and surprising himself by learning new languages.

Read more