LLMs in Finance: Use Cases, Benefits, Risks & Cost in 2026

LLMs in finance are AI systems that read, interpret, and process financial documents, transactions, and customer data to automate compliance, lending, fraud detection, and reporting tasks. Backed by machine learning, NLP, and neural networks, they can boost data processing speed up to 50x and cut back-office costs by up to 40%.

It’s not just basic automation that defines this shift—LLMs solve practical, day-to-day challenges finance teams face: manual document review, slow compliance cycles, and inconsistent customer service. This blog covers real-world use cases in BFSI, deployment risks, and implementation costs for 2026.

What are large language models?

LLMs are advanced AI systems capable of reading, interpreting, and generating human language without changing the context. If we talk from a finance perspective, it helps businesses effortlessly work with massive data volumes, documents, and customer information. Unlike traditional AI-driven bots, LLMs never follow any predefined rules. Rather, they are programmed so that they can interpret the actual context of the financial information.

This is how these can support the everyday operations your business deals with, including:

• Reviewing loan applications and other types of financial documents
• Answering customer questions via AI assistant bots
• Summarizing compliance and regulatory documents to make your business audit-ready
• Analyzing financial reports and market data for automated task execution
• Flagging unusual transactions that often signal money laundering and fraud

Whether you are a fintech startup owner, a banking product leader, or a wealth management firm’s CTO, LLMs will help you automate repetitive tasks. 

Why is finance the fastest-growing sector for LLM adoption?

For financial service companies across the US, delaying the adoption of large language models in banking, insurance, and other sectors will create a competitive disadvantage. Here’s why!

1. 65% of companies are already planning to increase their investments in generative AI. They took this decision not just to reduce costs but also to create new revenue opportunities. 
2. Financial businesses are not limited to physical operations. Rather, they have to work with contracts, loan files, KYC documents, compliance reports, and research. These are text-rich workflows where LLMs can generate the highest ROIs easily.
3. 68% of insurers are investing in AI chatbots to drive cross-selling and value-added products. 

Once you embed LLMs into compliance, lending, customer operations, or research pipelines, you will gain an excellent advantage in both speed and margin in 2026. 

Key benefits of LLMs for finance

• Improved data analysis & insight generation

With finance LLMs, you can process massive volumes of both structured and unstructured datasets in minimal time. These often include loan documents, credit reports, transaction histories, or regulatory filings. As you get precise, actionable insights, decision-making will speed up and become more consistent across your teams. 

• LLMs can easily flag inconsistencies in US loan applications by cross-checking W-2 forms, bank statements, and stated income data before sending them for underwriting approvals.
• Whether it’s the Experian or the Equifax report, they can analyze credit bureau summaries to highlight risks like increased credit utilization or missed payments.
• You can also compare SEC filings or earnings reports to identify sudden shifts in debt exposure, revenue quality, or liquidity risks across different profiles. 

• Efficiency through cost reduction

LLMs automate high-volume administrative and compliance-related workflows, allowing you to scale operations overall without investing in proportionate back-office staffing. This becomes more crucial as the US financial market functions in a strict regulatory environment, guarded by CFPB, FINRA, SEC, and AML. 

• You can automate KYC processing easily as LLMs will validate US government-issued IDs, SSNs, and address consistency across various onboarding forms. 
• Most LLMs can generate first-pass credit memos for your loan officers. This can reduce the manual effort needed in reviewing consumer or small business loan applications.
• These models also summarize regulatory updates from different US agencies, like the SEC or the CFPB, into a structured compliance checklist. 

• Improving customer experience & engagement

With an LLM in finance, you can deliver faster, more contextual, and highly personalized customer interactions like PayPal, Stripe, or Robinhood. By doing so, you can avoid delays in support or vague communication, as these introduce spikes in churn, acquisition cost, and lifetime value. 

• LLMs answer account-specific queries instantly by referring to transaction history, like explaining overdraft fees or pending ACH transfers.
• You can proactively alert your users about different financial risks, like low balance, before any scheduled bill payment or unusual login activity from new locations. 
• These AI-driven models can generate personalized product recommendations, like suggesting credit limit increases based on cash flow patterns or repayment behavior.

• Human augmentation

An LLM acts as an internal intelligence layer, which reduces manual cognitive workload for your analysts, advisors, and compliance teams. Thus, they can focus more on decision-making and not remain stuck with information-gathering tasks. 

• Relationship managers can pull pre-meeting briefs from the LLMs that usually highlight cash flow trends, credit exposure, and cross-sell opportunities.
• Your finance analysts will receive automated summaries of different portfolio performances, each segmented based on US regions, industries, or risk tiers.
• Compliance officers can easily interpret new regulatory updates and map those directly to internal policy changes and audit requirements.

Key use cases with real-world examples

Universal use cases of finance-based LLMs

• Natural language communication

Whether it’s your employees or your end users, everyone can chat with the LLM model using simple text messages. Since it’s powered by NLP and generative AI, every interaction remains human-like. Besides, it also ensures responses are adapted to suit the user’s language and communication style perfectly. Add speech synthesis capability, and the LLM can even support: 

• Voice communication
• Real-time phone and VoIP call transcription
• Call auto-dispositioning

Take the example of Bank of America’s AI-powered virtual assistant Erica. It’s smart enough to help customers interact directly with the bank using natural language. They can ask questions about account balances, recent transactions, bill payments, spending habits, and even credit scores in conversational English. At least with this bot, no user has to navigate the complex menus of the online banking platform.  Erica has already handled billions of client interactions, thereby reducing call centre volume, lowering customer service costs, and improving satisfaction through 24/7 self-support.

• Customer document parsing

Once you deploy LLMs in finance processes, it will become easier for you to interpret and categorize customer documents. These can be third-party agreements, proofs of solvency, or service applications. These AI models extract the required data and validate it, either by reconciling it against historical records or cross-referencing it with external data sources. By doing so, you get accurate reports on discrepancies across different documents of the same customer.

One of the best examples would be that of Rocket Mortgage. This fintech enterprise uses AI to automatically parse submitted documents, like tax returns, W-2s, pay stubs, and bank statements. It also extracts relevant financial data to accelerate mortgage approvals while minimizing manual document review. Underwriting timelines have been shortened, too, allowing Rocket Mortgage to improve operational efficiency.

• Financial document review

You can also engage the LLMs in reviewing internal system documents, like financial reports, invoices, and service agreements. The models are often trained to flag both stylistic and factual gaps that are not compliant with regulatory standards like SEC, TRID, ECOA, GLBA, and NAIC. Your compliance specialists can also use these to screen BFSI legal documents for operational rules, data privacy requirements, or new formats. 

With an LLM-powered assistant, Morgan Stanley allows its financial advisors to search and summarize thousands of research reports and investment documents instantly. Since they don’t have to spend time on manual data collection, they can easily serve more clients, respond faster, and deliver informed investment recommendations.

• Financial data consolidation

The best LLM use case in finance is data aggregation in relevance to a specific business aspect, like risk assessment, customer due diligence, or mortgage closing. Once the models have all the necessary datasets in one place, they begin summarizing and structuring them according to the pre-defined rules. Speech-to-text LLMs can even capture service-critical information, including requests from different quotes or service term adjustments, during real-time customer calls. 

Intuit has integrated a Gen AI bot into QuickBooks to consolidate invoices, accounting records, payroll, expenses, and banking transactions into a unified financial dashboard. It further helped the internal teams to speed up financial reporting, improve cash flow visibility, reduce reconciliation errors, and spend more time on strategic decision-making.

• Financial fraud detection

These models use semantic analysis to spot discrepancies, which otherwise indicate fraudulent attempts across interactions, financial documents, and transactions. Every suspicious activity gets auto-classified by type, including compliance breaches or document forgery. Thus, your teams can take appropriate actions without wasting time.

You can consider the use case JPMorganChase has introduced. It deployed multiple AI agents to analyze millions of daily transactions and identify unusual payment behavior indicative of fraud. By doing so, the bank has successfully reduced financial losses, minimized false positives, and strengthened customer trust by preventing unauthorized transactions.

• Intelligent recommendations

When you work with BFSI processes, your specialists can get help from LLMs to perform detailed reasoning on acquired information. It can be project loan defaults based on the borrower’s behavior, or the most profitable investment strategies according to the owner’s assets

When you work with BFSI processes, your specialists can get help from LLMs to perform detailed reasoning on acquired information. It can project loan defaults based on the borrower’s behavior or the most profitable investment strategies according to the owner’s assets. With smart recommendations, you can not only keep your users engaged in your business offerings but also guide them towards making better financial decisions.

SoFi leverages LLMs to recommend personalized loans, refinancing options, investment products, and savings solutions based on different customer profiles. With this, it increased product adoption, improved customer retention, boosted cross-selling opportunities, and enhanced the overall member experience.

• Financial data synthesis

As most LLMs are based on Generative AI, they can produce new content from your company’s existing knowledge repositories. For example, they can analyze the call interactions between your users and banking team to draft topical response memos. In addition to this, the models can mine data patterns from historical transactions and generate training information. 

Bloomberg uses generative AI to summarize earning calls, SEC filings, analyst reports, and financial news into accurate, concise insights for investors. This allowed professionals to make faster investment decisions without sacrificing analytical depth and precision.  

Industry-specific LLM use case for finance 

Banking

Even after banks have gone digital, employee productivity still hasn’t improved much. That’s because numerous tasks are still performed manually. Whether it’s extracting information from loan applications or answering customer enquiries, unnecessary efforts often reduce outcome efficiency. This has become one of the major challenges across the entire banking sector, causing operational bottlenecks every now and then.

With an LLM, however, these issues can be addressed right at the core.

• Fraud detection using narrative transcription pattern analysis across real-time account activities and payment flows
• Regulatory alignment by mapping FDIC, OCC, and Federal Reserve updates to internal banking policies and processes
• Credit decision support by standardizing and validating internal credit memos across commercial and retail banking teams
• Operational risk monitoring through aggregation of branch-level incidents, disputes, and service downtime

LLMs allow banking institutions to deliver more personalized customer services faster. Employees can also complete different tasks in seconds, which otherwise would have taken hours when done manually. As a result, it will become easier to improve user retention rates and generate higher ROI in the long run.

Insurance

This industry is still dealing with complicated, jargon-filled policies and restrictive guidelines. As a result, insurers often struggle with delivering precise and regulatory-compliant customer service. For example, the underwriting process is mostly manual and labor-intensive. Risk assessment requires manual scrolling through a large volume of unstructured information.

LLMs designed specifically for insurance operations can help by enabling:

• Underwriting risk validation by cross-verifying application data against external policyholders and claims history signals
• Regulatory compliance mapping across state-specific insurance regulations and product guidelines
• Loss trend analysis by organizing unstructured claims data into actuarial-ready insights
• Policy language risk review to flag ambiguous clauses that often increase litigation exposure in US insurance contracts

Insurers can thus save time with LLMs, minimize human errors, and improve risk evaluation easily. Apart from this, they can also train the model so that it can generate accurate policy recommendations based on customer behaviors and financial habits. When fed with synthetic data, it can help in fraud detection and ensure fair and robust responses across all insurance operations.

Lending

Most lending professionals still rely on manual, tedious methodologies to evaluate credit risk profiles for every loan application. While doing so, they usually rely on a handful of financial data points, which ultimately leads to inaccurate risk assessment. This traditional system never offers a clear picture of the applicant’s creditworthiness.

So, to address these issues, the best approach is to build and deploy an LLM specifically for lending operations. When trained with appropriate datasets, it can help in:

• Alternative credit assessment using cash flow analysis from US bank statements and SMB financial records
• Loan application risk scoring by analyzing borrower narratives in addition to structured credit bureau data
• Early default prediction using behavioral signals from repayment history and account activity patterns
• Underwriting standardization across lenders to ensure consistency in approval logic and documentation quality

LLMs help accelerate loan approval speed by automating most of the manual tasks involved. Easy access to alternative data helps increase the customer acquisition rate. It means that lenders can ensure that even the underserved communities can access different types of financing options through their lending portals integrated with an AI-based LLM.

Investment 

Both portfolio and trade managers need access to large volumes of up-to-the-minute market information. Only by having it can they make accurate and time-sensitive decisions, thereby delivering a better experience to their customers. However, manual reconciliation of information from news, reports, macroeconomic indicators, and social media is time-consuming. Besides, it often causes analysts to miss small but highly impactful data points.

That’s why investing in LLMs is crucial, as these enable:

• Earnings call intelligence by extracting sentiment shifts, risk signals, and guidance changes from US-listed company transcripts
• SEC filing analysis (10-K, 10-Q) to detect material changes in financial health, risk exposure, and disclosures
• Macro-to-equity mapping using Federal Reserve updates, CPI data, and economic indicators
• Portfolio narrative generation for client reporting and advisor communication 

Portfolio managers can thus use LLMs to react quickly to market changes and improve investment performance simultaneously. They also gain a competitive advantage as real-time data analysis gets accelerated, allowing them to deliver accurate and fast responses before their competitors. It becomes easier for professionals to rebalance different investment portfolios in response to macroeconomic market movements without errors.

Challenges and risks of deploying LLMs in finance

Accidental data disclosures

Whether you want to implement an LLM in banking or insurance, your business’s sensitive information will be at risk of unwanted exposures. This will further cause customer trust in your services to erode. So, the best approach will be to include contractual clauses explicitly. It will help you prevent the use of your business-specific data for model training. Apart from this, anonymize and encrypt every data fed to the LLMs. You can also apply privacy-preserving prompt tuning (RPAT) methods. 

Hallucinations 

Most AI LLMs are at risk of generating incorrect, incomplete, or outdated financial responses. So, make sure you use only the latest model versions. Give the LLMs unhindered access to your up-to-date business data using RAG so that they won’t have to work on obsolete information. 

Misalignment with regulatory needs

Sometimes, both the processing and reasoning logic you embed within the LLMs can fail to meet necessary US-specific compliance policies. This can further cause legal penalties and force you to permanently shut down the model. To address this, feed the LLM with up-to-date regulatory information via RAG for financial documents. Also, incorporate compliance rules directly within the prompt templates.

Opaque output logic

When the logic behind the LLM’s functioning is too vague, it will directly clash with the proof of ethical conduct. As a result, customer trust will start dwindling, leading to sudden drop-offs and declining revenue. What you can do is incorporate source citation requirements directly into the prompt templates to signal credibility.

How to choose the right LLM for your finance app?

Different ways to adapt LLMs in financial services 

Prompt engineering

Here, you will design structured instructions to guide the LLMs. These will allow them to perform specific financial tasks without having to change the underlying model. Whether you want to streamline fintech operations or build a lending platform, here are the benefits prompt engineering will yield. 

• Reducing costs of building AI features, as you won’t have to train models or invest in additional infrastructure
• Helping standardize outputs across regulated workflows like lending explanations or compliance summaries
• Allowing quick adaptation to changing CFPB, FDIC, or state-level regulatory updates
• Supporting rapid testing of AI use cases before scaling into production

Domain-specific fine-tuning

This approach helps adapt LLMs using US-specific financial datasets, like credit underwriting records, insurance claims data, or BFSI compliance notes. It ensures the model understands regulatory frameworks, financial language, and decision patterns used across banks, credit unions, or fintech lenders. Below are the benefits this approach offers.

• Improved accuracy in credit decisions by aligning with US lending patterns and FICO-based workflows
• Reduced errors in interpreting banking terminology, regulatory filings, and insurance policies
• More consistency across multi-state US compliance environments, like CFPB, OCC, and so on
• Enabling stronger model performance in industry-specific workflows like mortgage lending or SME credit evaluation

Retrieval-Augmented Generation (RAG)

RAG will connect the LLMs to your internal business systems, like loan databases or transaction logs. This way, the responses will be grounded in real-time, verifiable data. Some of the benefits you can enjoy are:

• Prevention of outdated or incorrect regulatory references will minimize compliance risks
• Changes in US financial policies or products won’t force you to invest in expensive model retraining
• Improvement in audit-readiness as outputs will be linked to traceable internal data sources

LLM retraining and full fine-tuning

You can also fine-tune the LLM on proprietary US financial datasets to customize its specific business operations and reasoning logic. Here are the benefits of this approach you can have for your financial startup entity. 

• Delivering higher accuracy in complex decisions like pricing, underwriting, or risk modelling
• Aligning model outputs with your internal credit policies and US regulatory frameworks
• Improving performance in large-scale portfolios where even the smallest error can cause a huge monetary impact

Integration with existing financial services 

The last approach will be to plan LLM integration for financial services, like core banking systems, CRM tools, and loan platforms. You can use the APIs supplied by the vendors or custom-build one as per your business needs. Benefits include:

• Enabling real-time AI assistance inside loan processing, support, and compliance systems
• Reducing operational delays across distributed US financial branches and teams
• Improving productivity by embedding AI directly into the tools your employees use everyday

Architecture of LLMs for finance

Data collection layer

The model first collects datasets from different internal and external sources. These often include:

• Core banking systems
• CRM platforms
• Loan management software
• Insurance databases
• Payment gateways
• ERP systems

Apart from this, it also pulls valuable financial information from market data providers, SEC filings, customer documents, chat conversions, and call transcripts. It’s only by combining everything that the LLM gains proper business context without leaving any gap.

Data processing and preparation layer

The data collected needs to be cleaned and prepared so that it can be fed to the LLM. For this, you can use multiple techniques as per the business use case, cost factor, and feasibility. OCR can extract text from scanned documents. ETL pipelines help standardize formats and eliminate duplicate records. If the LLM deals with sensitive PII, masking or anonymization will become non-negotiable. By doing so, you can ensure higher data quality while maintaining regulatory compliance.

Knowledge Retrieval Layer (RAG)

Fintech LLMs cannot just rely on pre-trained knowledge. Rather, they need to retrieve information from enterprise knowledge bases. That’s because these repositories store crucial data, including:

• Internal banking policies
• Underwriting guidelines
• Investment research
• Compliance documents
• Regulatory publications
• Historical financial records

This RAG approach allows the LLM to produce responses based on the latest and most relevant financial and organizational information.

Finance-tuned LLM layer

This acts as the main intelligence layer of the entire LLM. You will have to fine-tune the model for financial use cases properly. Only then can it understand industry terminology and complex financial language, thereby reducing the risks of hallucination and vague responses. Besides, fine-tuning the LLM enables it to perform complex tasks across banking, lending, insurance, and investment operations, like:

• Document analysis
• Report summarization
• Customer query answering
• Risk assessment
• Recommendation generations

AI governance and compliance layer

Every AI-generated response needs to pass through a governance layer before it becomes available for the users. Here, responses are checked for hallucinations and biases. Apart from this, it’s also responsible for applying regulatory rules, recording audit logs, and providing explainable outputs. With these controls in place, you can ensure the LLM adheres to financial regulations like FDIC, OCC, SEC, FINRA, PCI DSS, SOX, and AML requirements.

Recommended: Enterprise AI Governance & Compliance

Business logic and decision layer

The LLM works alongside other enterprise systems to generate accurate, consolidated outputs. However, to ensure the responses are tailored to financial use cases, you will need to implement validation protocols based on:

• Business rules
• Credit scoring models
• Fraud detection engines
• Underwriting platforms
• Recommendation systems

Monitoring and continuous learning layer

The last layer continuously evaluates the model once you deploy it to the production. Here, you will have to monitor response quality, detect model drift, collect user feedback, optimize prompts, and retrain the LLM using new financial datasets and regulatory updates.

Cost to implement LLMs in financial services

The finance LLM development cost in 2026 starts from $45K and can climb to $600K+. It mostly depends on how complex your product is. For example, a basic LLM needs $45K to $100K, which is ideal for small- to medium-scale businesses. On the contrary, if you want highly customized solutions with extensive pre-trained LLMs, the development cost will be between $400K and $600K+. 

Key factors affecting the implementation cost of LLMs in financial services 

Model complexity

When you choose an LLM operating using an API, your implementation costs will remain low, around $45K to $60K. However, if your financial organization needs a custom or highly fine-tuned LLM for fraud detection in banking, lending, or investment analysis, costs can exceed $500K+. That’s because the more complex the LLM is, the more effort will be needed in training, analysis, testing, and infrastructure maintenance.

Data preparation

You can expect an overall expense of $20K to $300K on this aspect, depending on the quality of your existing financial information assets. For example, let’s assume your transaction histories, loan documents, or customer data are spread across multiple systems. Therefore, cleaning, organizing, securing, and normalizing them to ensure the LLM can deliver relevant responses will be expensive.

Integration needs

Integrations of the LLMs with your existing financial systems can incur additional costs of $25K to $400K+. You can either use third-party APIs often provided by different external vendors or build custom ones as per your business requirements. The key here is to plan for integrations from day one, especially if data sits within core banking platform, CRM tools, loan origination software, or payment gateway.

Regulatory compliance

To build compliance directly into the LLM’s architecture, you will have to invest about $15K to $250K additionally. Here, the investments will cover building features like explainable AI, audit trails, data encryption, access controls, and governance mechanisms. By doing so, you can ensure the LLM adheres to the industry regulations, like CFPB, SEC, FINRA, and AML requirements.

Scalability 

As you need to design an LLM for a growing customer base, average scaling costs will be around $30K to $600K+. This will then help you support thousands of users, real-time transaction volumes, and enterprise-level workloads easily. 

Maintenance & updates

As part of ongoing LLM maintenance and update rollout, an annual spend of $10K to $200K will be required. This will cover prompt optimization, performance monitoring, model updates, security patches, and changes necessary to stay aligned with the US evolving regulatory landscape.

Cost-saving strategies for LLM solution implementation in financial services 

Whether you own a lending platform or are a fintech startup founder, the key is to invest in areas where the AI solution can deliver maximum results. Rather than going all out with a fully fine-tuned LLM model, you can use the tips to save on unnecessary LLM development expenses. 

• Start with one high-impact use case where the LLM can generate maximum value in terms of efficiency boost, time savings, and revenue. This can be loan document analysis, customer support, or compliance reporting.
• Use pre-trained LLMs before investing in a custom model for your US financial business. It will not only reduce the initial development costs but also help you validate business outcomes before you spend in expensive fine-tuning.
• Adopt the RAG approach instead of full model retraining. That’s because connecting the LLM to your company’s internal policies, customer records, and financial documents is more affordable.
• Extend your core banking system, CRM, or loan management platform to your LLM via APIs. This will reduce both implementation and infrastructure expenses, otherwise necessary when you plan for a full-scale replacement.
• Prioritize high-quality data over massive volumes. When you invest in cleaning and normalizing your financial data assets, the LLM can generate appropriate responses. This will automatically drive the ROI up. 

Finance LLM consulting and implementation with GMTA Software

Your US finance business needs more than a generic AI solution. Rather, it requires an LLM that aligns with the existing financial systems, US regulations, and measurable business goals. That’s where GMTA Software Solution steps in to help you with the entire process, from planning to development and launch of the LLM. Our experts will be there with you to:  

• Identify high-impact LLM use cases
• Select the right implementation approach
• Integrate AI into banking, lending, insurance, or investment workflows

Whether it’s data preparation, model development, or RAG implementation, our focus stays fixed on delivering secure, compliant, and scalable AI solutions. So, if you are planning to adopt LLMs in your finance operations, GMTA will be your execution partner from day one.

FAQs

What can LLMs do for financial services?

LLMs can help with automating document processing, generating financial insights, improving customer support, detecting compliance risks, assisting with underwriting, and summarizing regulatory updates for your US financial business.

How much does it cost to build an LLM?

The average cost to build a finance LLM in the US is between $45K and $600K+. It will depend on model customization, data preparation, integrations, compliance requirements, infrastructure, and ongoing maintenance needs.

What are the best ways to adopt an LLM for financial services?

Start with prompt engineering or RAG for quick deployment of the LLM. Use domain-specific fine-tuning if you want to implement the model for specific financial tasks. Full-scale model retraining will be best suited for highly customizable, enterprise-grade model.

How long does it take to build an LLM for financial service?

Building a finance-focused LLM requires about 3 to 12 months. The exact timeline depends on project complexity, data readiness, regulatory requirements, system integrations, and if you are building a custom model or using a pre-trained LLM.

What compliance standards are necessary for finance LLMs?

Your finance LLM needs to comply with regulations like GLBA, SEC, FINRA, CFPB, AML/KYC requirements, SOC 2, PCI DSS, and state-specific regulatory laws. 

Anjali Upadhyay

Founder 

Anjali Upadhyay is the Founder of GMTA Software Solutions, a mobile and web application development company she built from the ground up in 2019. Under her leadership, GMTA has delivered 500+ production applications across healthcare, fintech, and on-demand services for clients in the US, UK, Singapore, and UAE. She leads GMTA’s AI practice, which has shipped production AI systems — including HIPAA-compliant healthcare workflows, LLM-integrated logistics platforms, and fintech automation tools — for US-based enterprise clients. Her writing covers AI product strategy, build-vs-buy decisions for AI systems, and the operational realities of moving AI from proof-of-concept to production at scale.