Generative AI: Enabling a New Wave of Workflow Automation

Generative AI is creating new opportunities within fintech, helping automate workflows that were previously manual and difficult to scale. Its impact is already significant and, in our view, is likely to grow, particularly when applied to processes that depend on interpreting and synthesizing unstructured information where language, narrative and context matter most.

Client engagement, as well as KYC and AML case management applications, offer a few examples of areas where generative AI is effectively applied. Within case management, teams must interpret IDs, contracts and corporate filings, generate investigation summaries and resolve cases efficiently. Generative AI can help streamline these tasks by accelerating document interpretation and narrative construction. Similar gains are emerging in research automation, earnings-call analysis and monitoring of regulatory and rulemaking developments, where large volumes of text must be processed quickly and accurately.

Within client engagement, we are seeing advisors apply AI to translate complex investment strategies into clearer, more digestible explanations. Summit portfolio company Vestmark offers one example of this trend, incorporating conversational and agentic AI capabilities into its platform to help advisors retrieve information, explore scenarios or complete simple tasks without navigating multiple systems. We believe tools like these meaningfully reduce friction in day-to-day workflows, while preserving the central role of the advisor.

‍

Machine Learning: Turbo-Charged by Expanding Data Volumes

Alongside classical mathematical models, quantitative machine learning has emerged as a powerful driver of innovation across fintech. Large language models focus on learning patterns in human language to generate and interpret text, while quantitative machine learning relies on explicit statistical and numerical models to make precise predictions and decisions from structured data. As the volume of transactional and behavioral data has grown, these ML techniques have become more sophisticated and more effective. Unlike rules-based systems, ML models are designed to learn from historical data to identify patterns and anomalies using approaches such as clustering, classification and time-series analysis.

In many structured prediction tasks including numerical modeling, regression, time-series forecasting or portfolio-level stress testing, generative AI remains ill-suited. In our view, quantitative machine learning remains the superior technology for these use cases, offering higher accuracy and lower latency. We already see its impact across areas such as anti-money laundering where abnormal patterns can be detected in near real time and in credit scoring and underwriting where time-series and behavioral models help predict default risk more precisely.

Machine learning is also enabling innovation in customer-facing financial platforms. For example, Summit portfolio company Fundraise Up applies ML to help personalize online donation flows by modeling donor behavior across variables such as time, location, device, language and traffic source. Trained on a large proprietary dataset of historical and real-time behavioral signals, these models are designed to optimize conversion, gift size and campaign efficiency while also supporting fraud detection and payment-risk mitigation.

‍

Mathematical Models: Here to Stay

Despite the attention surrounding generative AI, a large share of fintech continues to run on classical mathematical models. While many of these models were developed decades ago, they remain foundational to modern investment theory and quantitative finance. Options pricing models such as Black–Scholes, performance frameworks like CAPM and factor-based approaches such as Fama–French continue to underpin portfolio construction, risk analysis and regulatory reporting, and such we believe mathematical models will remain essential, particularly in areas of fintech where accuracy and regulatory clarity are non-negotiable.

These mathematical models are typically both deterministic and explainable, characteristics that matter deeply in financial services, in part because regulators often prefer approaches that can be audited, tested and relied upon to produce consistent outcomes under defined assumptions. One area where we continue to see mathematical models effectively applied is market surveillance and risk monitoring. Summit portfolio company TradingHub illustrates this well, applying quantitative market impact models to aid in deconstructing risk sensitivities, market correlations and the price impact of trading activity. TradingHub’s models help to reliably identify specific market-abuse scenarios and to reduce false positives. While generative tools may assist downstream, for example by aggregating signals across dashboards, translating outputs into more accessible formats or automating remediation workflows, they do not replace the core analytical logic that underpins these systems.

A similar opportunity for mathematical models exists within critical financial infrastructure. Core banking platforms, payment systems and trade execution rely on low latency, deterministic behavior and high uptime. We believe these requirements make them unlikely candidates for near-term disruption by generative AI.

‍‍

The AI Data Moat in Fintech

Across all three approaches – generative AI, machine learning and mathematical models – fintech companies benefit from unusually specialized datasets. Transaction data, behavioral signals and market information create what we believe are deep competitive moats in an AI-driven world. This dynamic mirrors what we see across vertical SaaS more broadly where workflow depth and proprietary data can help create differentiation.

In fintech, data are not simply inputs to AI systems, they are a defining source of advantage. The quality, structure and scale of financial datasets often determine the effectiveness of models and ultimately the outcomes and information they produce.