Hui Wen Goh

Elaine Lau, Markus Dücker, Ronak Chaudhary et al. · mit

Existing AI benchmarks lack the fidelity to assess economically meaningful progress on professional workflows. To evaluate frontier AI agents in a high-value, labor-intensive profession, we introduce BankerToolBench (BTB): an open-source benchmark of end-to-end analytical workflows routinely performed by junior investment bankers. To develop an ecologically valid benchmark grounded in representative work environments, we collaborated with 502 investment bankers from leading firms. BTB requires agents to execute senior banker requests by navigating data rooms, using industry tools (market data platform, SEC filings database), and generating multi-file deliverables--including Excel financial models, PowerPoint pitch decks, and PDF/Word reports. Completing a BTB task takes bankers up to 21 hours, underscoring the economic stakes of successfully delegating this work to AI. BTB enables automated evaluation of any LLM or agent, scoring deliverables against 100+ rubric criteria defined by veteran investment bankers to capture stakeholder utility. Testing 9 frontier models, we find that even the best-performing model (GPT-5.4) fails nearly half of the rubric criteria and bankers rate 0% of its outputs as client-ready. Our failure analysis reveals key obstacles (such as breakdowns in cross-artifact consistency) and improvement directions for agentic AI in high-stakes professional workflows.

Hui Wen Goh, Ulyana Tkachenko, Jonas Mueller

Real-world data for classification is often labeled by multiple annotators. For analyzing such data, we introduce CROWDLAB, a straightforward approach to utilize any trained classifier to estimate: (1) A consensus label for each example that aggregates the available annotations; (2) A confidence score for how likely each consensus label is correct; (3) A rating for each annotator quantifying the overall correctness of their labels. Existing algorithms to estimate related quantities in crowdsourcing often rely on sophisticated generative models with iterative inference. CROWDLAB instead uses a straightforward weighted ensemble. Existing algorithms often rely solely on annotator statistics, ignoring the features of the examples from which the annotations derive. CROWDLAB utilizes any classifier model trained on these features, and can thus better generalize between examples with similar features. On real-world multi-annotator image data, our proposed method provides superior estimates for (1)-(3) than existing algorithms like Dawid-Skene/GLAD.

Hui Wen Goh, Jonas Mueller

In real-world data labeling applications, annotators often provide imperfect labels. It is thus common to employ multiple annotators to label data with some overlap between their examples. We study active learning in such settings, aiming to train an accurate classifier by collecting a dataset with the fewest total annotations. Here we propose ActiveLab, a practical method to decide what to label next that works with any classifier model and can be used in pool-based batch active learning with one or multiple annotators. ActiveLab automatically estimates when it is more informative to re-label examples vs. labeling entirely new ones. This is a key aspect of producing high quality labels and trained models within a limited annotation budget. In experiments on image and tabular data, ActiveLab reliably trains more accurate classifiers with far fewer annotations than a wide variety of popular active learning methods.

Hui Wen Goh, Jonas Mueller

Structured Outputs from current LLMs exhibit sporadic errors, hindering enterprise AI efforts from realizing their immense potential. We present CONSTRUCT, a method to score the trustworthiness of LLM Structured Outputs in real-time, such that lower-scoring outputs are more likely to contain errors. This reveals the best places to focus limited human review bandwidth. CONSTRUCT additionally scores the trustworthiness of each field within a LLM Structured Output, helping reviewers quickly identify which parts of the output are wrong. Our method is suitable for any LLM (including black-box LLM APIs without logprobs such as reasoning models and Anthropic models), does not require labeled training data nor custom model deployment, and works for complex Structured Outputs with many fields of diverse types (including nested JSON schemas). We additionally present one of the first public LLM Structured Output benchmarks with reliable ground-truth values that are not full of mistakes. Over this four-dataset benchmark, CONSTRUCT detects errors from various LLMs (including Gemini 3 and GPT-5) with significantly higher precision/recall than other scoring methods.