Taktile Labs · Applied AI research
Taktile Labs is an applied research group focused on advancing the frontier of AI in financial decisions: exploring, testing and proving new techniques across the areas that matter in banking and insurance, and sharing the evidence.
Financial services will spend $97B on AI by 2027. Yet the distance between what general-purpose AI can do and what regulated institutions can reliably deploy remains vast. Taktile Labs exists to close this gap.
Every day, financial institutions run on a current of critical decisions, from determining if a transaction is fraudulent to calculating the amount of risk to accept when underwriting a loan. At Taktile Labs, we research what it takes to deploy AI in this high-stakes decision-making where errors aren’t an option.
We work closely with development partners and industry experts to create high-quality evaluation data sets, which we use to assess AI performance in financial services-specific contexts.
Research agenda
We’re pursuing five research tracks focused on the most important requirements for trustworthy AI in regulated financial institutions.
Explore our Research Agenda →In collaboration with our development partners, we use real-world data to build trusted benchmarks for model performance in core financial services use cases. Every benchmark is designed around the KPIs business teams care about most: accuracy, cost per decision, and latency.
View UBO-Bench →Not every decision should be fully automated, and not every decision requires a human to intervene. We pursue balanced research that helps teams unlock the benefits of AI in complex decision-making while preserving the value of human judgment and engagement.
Agentic systems based on LLMs are stochastic and hard to inspect, challenging assumptions behind SR 11-7 and traditional model risk management. We work with institutions and regulators to clarify what responsible adoption looks like in practice.
Foundation models are trained on public text, but financial decisioning runs on data with rich sequential and relational structure. We explore transformer-based architectures purpose-built for common data structures in financial services.
The most effective AI systems will be built using hybrid architectures. We help financial institutions navigate AI hype with clarity and choose the right tool for each task while balancing cost, risk, and performance.
Benchmark · KYBench
KYBench evaluates how well agentic AI systems can investigate a business, weigh the evidence they find, and judge regulatory red flags in realistic Know Your Business reviews. Built with real businesses annotated by expert compliance practitioners.
Read the methodology →
Total economic cost per business screened
September 2026 update. All 25 models and methodology on the benchmark page.
People
Taktile Labs is powered by consistent collaboration between a dedicated internal research team and an external Research Council and Advisory Board.
View full team & bios →
Publications
Benchmarks, technical reports, and research.
David Ahn, Maximilian Eber, PhD
A benchmark of AI models that sort small-business bank transactions into the categories underwriters use. Measures accuracy, calibration and cost across 23 methods, from keyword rules and frontier LLMs to decision models fine-tuned on a lender’s own labelled transactions.
Moritz Geist, David Ahn, Maximilian Eber, PhD
A benchmark of AI agents that trace ownership structures for beneficial ownership checks on UK and German registers. Measures tracing accuracy across frontier models, importance of country specific configuration, and how much of the work agents complete on their own.
Koen Roelofs, Jakob Schmitt, Maximilian Eber, PhD
The first benchmark of prompt-injection resistance for agentic underwriting. Measures defense success across 16 frontier models, three providers, and five attack vectors — with and without untrusted-content tagging.
David Ahn, Maximilian Eber, PhD, Sahith Jagarlamudi
The first public benchmark of AI-driven adverse media investigation. Evaluates detection accuracy, evidence quality, reliability across agent runs, and cost efficiency across frontier models.
Nico Klees, Maximilian Eber, PhD
The first public benchmark for agentic financial document processing. Evaluates extraction accuracy, cross-document reasoning, calculation correctness, and structured output quality across seven frontier models. Built on anonymized production data from financial institutions.
Dustin Eaton, Maximilian Eber, PhD
Why AML teams must now apply model risk management standards to AI systems. Published in ACAMS Today, exploring how regulators are extending MRM frameworks to AI deployed in compliance functions — and what institutions need to do to prepare.