Review waiting, please be patient. This may take 3 months or more, since drafts are reviewed in no specific order. There are 4,421 pending submissions waiting for review. If the submission is accepted, then this page will be moved into the article space. If the submission is declined, then the reason will be posted here. In the meantime, you can continue to improve this submission by editing normally. Where to get help If you need help editing or submitting your draft, please ask us a question at the AfC Help Desk or get live help from experienced editors. These venues are only for help with editing and the submission process, not to get reviews. If you need feedback on your draft, or if the review is taking a lot of time, you can try asking for help on the talk page of a relevant WikiProject. Some WikiProjects are more active than others so a speedy reply is not guaranteed. How to improve a draft Wikipedia:Contributing to Wikipedia – a basic overview on how to edit Wikipedia. Help:Wikitext – how to use the markup Help:Referencing for beginners – how to include references Wikipedia:Article development – how to develop your article Wikipedia:Writing better articles – how to improve your article Wikipedia:Verifiability – make sure your article includes reliable third-party sources You can also browse Wikipedia:Featured articles and Wikipedia:Good articles to find examples of Wikipedia's best writing on topics similar to your proposed article. Improving your odds of a speedy review To improve your odds of a faster review, tag your draft with relevant WikiProject tags using the button below. This will let reviewers know a new draft has been submitted in their area of interest. For instance, if you wrote about a female astronomer, you would want to add the Biography, Astronomy, and Women scientists tags. Add tags to your draft Editor resources Easy tools: Citation bot (help) | Advanced: Fix bare URLs Reviewer tools Instructions · What links here · Large Transaction Model (LTM) (talk: + · bio) · (log) · Copyvios report · reFill · Citation Bot · (Search: Google, Wikipedia) · Submitted 15 days ago by ~2026-27850-17 (talk: D · +) · Last edited 15 days ago by Escrime-2004

A Large Transaction Model (LTM) is a proposed class of machine learning models designed to analyze sequences of financial transactions in order to identify patterns, anomalies, and behavioral trends. LTMs are typically based on transformer architectures and are conceptually analogous to large language models (LLMs), but operate on structured financial data such as payments, transfers, and account activity rather than natural language text.

The term has been used primarily in industry contexts to describe models trained on large-scale transactional datasets, particularly for applications in fraud detection, anti-money laundering (AML), and financial risk assessment.^[1][2][3]

The analysis of financial transactions has long been a central function in banking, payments processing, and regulatory compliance. Early systems, particularly in the late 20th century, relied heavily on rule-based approaches. These systems were built around predefined thresholds and expert-defined heuristics, such as flagging unusually large transfers, rapid sequences of transactions, or activity in high-risk geographic regions. While effective for known patterns, rule-based systems were often inflexible and required continuous manual updates to adapt to new forms of financial crime.

As digital financial systems expanded and transaction volumes increased, statistical models and conventional machine learning techniques were introduced to improve detection capabilities.^[4] Methods such as logistic regression, decision trees, and ensemble techniques including random forests and gradient boosting enabled more data-driven analysis. These approaches allowed systems to incorporate a broader range of variables and to learn from historical data. However, they remained dependent on manual feature engineering, meaning that domain expertise was required to determine which aspects of transaction behavior were relevant. This limitation reduced their ability to generalize to novel or complex patterns.

By the 2010s, the increasing availability of longitudinal transaction data encouraged a shift toward modeling financial activity as a sequence of events rather than as isolated records. Sequence-based neural networks, including recurrent neural networks (RNNs) and long short-term memory (LSTM) models, were applied to capture temporal dependencies in transaction streams. These models enabled systems to consider the order, timing, and contextual relationships between transactions, improving the detection of evolving behavioral patterns.

Despite these advantages, recurrent architectures presented practical challenges. Training on long transaction histories was computationally intensive, and the models often struggled to retain information across extended sequences.^[5] Their inherently sequential processing also limited scalability, making it difficult to efficiently handle the large and continuously growing datasets typical of modern financial systems. These constraints led to continued exploration of alternative modeling approaches.

The introduction of transformer architectures in natural language processing marked a significant shift in sequence modeling. Models such as BERT and GPT demonstrated that attention-based mechanisms could effectively capture relationships across long sequences without relying on recurrence. This development prompted researchers and practitioners to explore the application of transformers to structured data, including financial transactions.

In this context, each transaction is treated as an event characterized by attributes such as amount, timestamp, merchant category, location, and transaction type. These attributes are encoded into numerical embeddings that represent both the content of individual transactions and their contextual relationships. Transformer models use self-attention mechanisms to analyze interactions between transactions across time, enabling the identification of subtle patterns such as gradual behavioral changes, periodic spending habits, or anomalous activity.

The ability of transformers to process sequences in parallel and to scale to large datasets made them particularly attractive for financial applications, where both data volume and temporal complexity are significant.

The term “Large Transaction Model” emerged in the early 2020s within industry discussions, reflecting an analogy with large language models and emphasizing the scale of both data and model architecture. In October 2023, the Cambridge-based company Featurespace introduced a model called TallierLTM,^[1][2][3][6] describing it as a transformer-based system trained on extensive transaction data for applications such as fraud detection and financial crime prevention.^[2][3]

The use of the term has remained relatively limited and is not yet widely established in academic research.^[7][8] Similar approaches are often described using broader terminology, such as transformer-based sequence models or deep learning methods for transaction data. As a result, it remains unclear whether “Large Transaction Model” represents a distinct category or a descriptive label for an emerging set of techniques.

Development in this area continues to be shaped by both technical and regulatory factors. Research efforts have focused on improving model efficiency, handling sparse and heterogeneous transaction data, and integrating multiple data sources, including graph-based representations of financial relationships.^[4] Hybrid approaches that combine transformers with other modeling techniques have also been explored to address specific challenges in fraud detection and risk analysis.