PLAYBOOK ATLASPlaybook Atlas

Finance

AI in finance: explore 49 applications.

Turn financial records and supporting evidence into decisions that can be reconciled and reviewed.

Explore payment matching, lending preparation, financial-crime investigation and planning workflows for finance and control teams.

Plan your first AI move See the 49 applications

Where the 49 applications sit

Explore the domains
  1. 14Lending Operations
  2. 13Fraud Prevention
  3. 11Regulatory Compliance
  4. 4Risk Assessment
  5. 3Customer Service

01Why AI now

Why finance is moving on AI, in three figures.

Each figure links to its source.

  1. Global fraud losses: over $485B in 2023

    Nasdaq estimated that fraud scams and bank-fraud schemes caused nearly $500 billion in global losses during 2023.

    Source: Nasdaq

  2. 2023 enforcement values rose 57% YoY

    Fenergo reported that global financial-institution enforcement actions reached $6.6 billion in 2023, up 57% from 2022.

    Source: Fenergo

  3. 75% of UK finance firms already use AI

    A 2024 survey of 118 financial firms found that 75% already use AI and another 10% plan to adopt it within three years.

    Source: Bank of England

How to evaluate a first project

Start with one evidence packet, allocation or exception queue. Reconcile amounts to authoritative records, expose uncertainty and define who may approve or post a change. Evaluate accuracy, missed exceptions and downstream outcomes against the existing process; a plausible narrative cannot establish a correct financial decision.

02Capability map

Where finance puts AI to work, domain by domain.

Pick a domain to see its applications. Each one opens a full report.

Inside Lending Operations

Applications in Lending Operations

What the use cases ask AI to do

Share of the finance use cases, by the kind of work the AI does.

  1. Reasoning

    Analyzes, predicts and decides: scoring, forecasting, classifying.

    86%

  2. Agentic

    Acts: plans, uses tools and completes multi-step tasks.

    14%

Under 1% each: perception, generation and optimization.

03Workflow ideas

Ways AI could change the work, in finance.

Atlas groups these applications by the changes they propose to a workflow. These preliminary classifications help you explore ideas; each application still needs a review of its sources and operating requirements.

Workflow categories in this catalog

Share of the 41 applications assigned a category.

How to evaluate an application

  1. Start with a bounded task

    Identify the inputs, the person responsible, and the decision or output the workflow must produce.

  2. Check the support for each method

    Look for source-backed details about the specific implementation, including its limits and human review.

  3. Measure the pilot

    Record the current baseline, agree on acceptance criteria, and test on representative cases before expanding.

Explore applications

04What gets built

The approaches finance uses most, and when to skip them.

Knowing when not to use one matters as much as knowing when to.

  1. 01 · 8 applications

    Workflow Automation

    Use AI inside orchestrated workflows to automate multi-step business processes.

    Use it when

    • When you have repetitive, rule-heavy workflows that still require human judgment or document understanding.
    • When the process spans multiple systems (ERP, CRM, ticketing, email) and manual coordination is slow or error-prone.
    • When large volumes of semi-structured or unstructured data (emails, PDFs, forms) must be processed consistently.

    Skip it when

    • When the process is highly creative, exploratory, or non-repeatable, with no stable sequence of steps.
    • When decisions are extremely high-stakes (e.g., life-critical medical decisions, major legal judgments) and regulations require full human control.
    • When you lack reliable access to the necessary systems or data sources to complete the workflow end-to-end.
    Workflow Automation, in depth

  2. 02 · 6 applications

    Classical-Supervised

    Train models on labeled tabular data to predict well-defined outcomes.

    Use it when

    • When you have a clearly defined target variable (label) and a well-scoped prediction problem (classification or regression).
    • When your data is primarily structured/tabular (numeric and categorical features) rather than raw text, images, or audio.
    • When you have sufficient labeled historical data that is representative of future production data.

    Skip it when

    • When you lack labeled data or labels are extremely sparse/expensive, making supervised training impractical.
    • When the core data is unstructured (e.g., raw text, images, audio) and you are not converting it into meaningful features first.
    • When the task is exploratory or descriptive (e.g., clustering, anomaly discovery without labels), where unsupervised or self-supervised methods are more appropriate.
    Classical-Supervised, in depth

  3. 03 · 6 applications

    RAG-Standard

    LLM + retrieval layer that injects relevant documents into the prompt at query time.

    Use it when

    • You need an LLM to answer questions based on proprietary or private documents that are not part of its pretraining data.
    • Your knowledge changes frequently (e.g., policies, product docs, pricing) and you want updates to be reflected without re-training the model.
    • You have a large corpus of unstructured or semi-structured text (wikis, PDFs, tickets, emails, manuals) that users struggle to search effectively.

    Skip it when

    • Your task is primarily structured data querying or analytics (e.g., reporting over a relational database) where SQL or BI tools are more accurate and efficient.
    • You have very small, static knowledge (e.g., a short FAQ) that can fit directly into a prompt or be encoded as rules without a retrieval layer.
    • You require strict, deterministic behavior with no tolerance for probabilistic errors (e.g., certain financial calculations, safety-critical controls).
    RAG-Standard, in depth

05The applications

49 finance applications, each with a full report.

The widest first, by the number of use cases each covers. Every report shows what changed, how it works, who has done it and what can go wrong.

  1. Financial Crime and SAR Intelligence HubThis AI solution uses AI to detect, investigate, and report suspicious activity across banks, wealth managers, and other regulated financial institutions. It combines transaction monitoring, crypto tracing, fraud detection, and regulatory analysis to streamline AML reviews and generate higher-quality Suspicious Activity Reports. The result is faster detection of financial crime, reduced compliance cost, and lower regulatory and reputational risk.Batch → RT17 use cases
  2. OriginationFlow AIAI solution grouping for lending application processing that accelerates bank software delivery with GitLab-assisted development and improves cash-flow underwriting through resilient multi-aggregator bank-data routing.Expert → AI8 use cases
  3. Auto Loan Credit ScoringCompliant gradient-boosted credit scoring for auto loan underwriting, improving default prediction and approval decisions while supporting Basel, Federal Reserve, and ECB model governance expectations.Expert → Platform7 use cases
  4. Financial Asset Tracing InvestigatorThis AI solution uses agentic AI to trace financial assets across accounts, instruments, and institutions while continuously monitoring for fraud, money laundering, and other illicit flows. It ingests and links transactional, customer, and third‑party data to surface hidden relationships, automate investigations, and guide analysts with risk-aware recommendations, reducing losses and improving regulatory compliance.Batch → RT7 use cases
  5. AI Credit Underwriting IntelligenceAI Credit Underwriting Intelligence uses machine learning and generative agents to analyze borrower data, financial statements, documents, and alternative data to assess creditworthiness in real time. It automates and augments credit analysis for commercial, CRE, C&I, and agricultural loans, enabling faster decisions, more consistent risk modeling, and fairer, data-driven lending outcomes. Lenders gain higher throughput, reduced manual review effort, and improved portfolio performance through better, earlier risk detection.Expert → AI6 use cases
  6. Sentinel PulseMonitors social media sentiment and emerging events during market episodes to help finance risk teams quickly detect, interpret, and respond to fast-moving signals at scale.Expert → AI6 use cases
All 49 applications, A to Z

Work with us

The catalog shows what is possible. A sprint decides what you should fund.

Reading every report still leaves the hardest question open: which move is worth funding first, here, given your constraints. That is the sprint: two weeks, fixed scope, and a candid answer either way.

Book a 30-minute callHow the sprint works

Two weeks Fixed scope A candid fit-or-no-fit answer

  1. 01

    Three ranked opportunities

    A decision-ready shortlist scored against value, feasibility, evidence, risk, and time to proof.

  2. 02

    One pilot-ready workflow

    A bounded operating design with actors, handoffs, data needs, controls, and success measures.

  3. 03

    A defensible 90-day plan

    Build-versus-buy guidance, ROI assumptions, risk treatment, sequencing, and executive decisions.

06Rules & standards

Rules and standards to check, with their scope and sources.

Selected US and EU references for these workflows. Applicability depends on jurisdiction, organization, data and intended use. The planning notes are Atlas editorial guidance; they are not official implementation-time estimates or an exhaustive list of obligations.

  1. Model risk management · SR 26-2

    Agency guidance

    United States · covered banking organizations

    The April 2026 interagency guidance replaces SR 11-7 and emphasizes a risk-based approach tailored to model use and bank complexity. The Federal Reserve letter says it is most relevant to supervised banks above $30 billion in assets.

    Federal Reserve: SR 26-2, revised model risk guidance

    Source checked

    Project planning note

    Map model inventory, validation and monitoring to the bank’s risk profile and supervisory context.

  2. GDPR · personal data and automated decisions

    Law or regulation

    European Union / EEA · processing within GDPR scope

    Personal-data processing needs an applicable lawful basis and safeguards. Article 9 adds conditions for special-category data; Article 22 addresses certain solely automated decisions with legal or similarly significant effects. Consent is not the only lawful basis.

    EUR-Lex: GDPR, Articles 3, 6, 9, 22 and 35

    Source checked

    Project planning note

    For credit or customer decisions, identify data bases, special-category use and whether Article 22 applies.

Your next step

Find your first move in finance.

Ask the workspace which move fits your company. Add your size, systems and constraints, and it works from these 49 applications to a plan your team can act on.

Plan your first AI move

Rather talk it through? Book a 30-minute call