AI in finance: Use cases, benchmarks, & benefits

What is AI in finance?

AI in finance is the use of machine learning, natural language processing, and generative AI to automate and improve financial tasks. Common applications include fraud detection, risk scoring, trading, forecasting, reporting, and customer service.

AI is not a single tool but a set of technologies applied to specific jobs that a bank, insurer, or finance team already does.

Adoption is no longer at the fringes of these fields. The U.S. Department of the Treasury's 2024 report, issued after its request for information on AI in financial services, found that AI use is increasing across the sector. The report also identified real concerns about data privacy, bias, and third-party vendor risk.

Mercor's AI Productivity Index, APEX, takes this a step further by measuring how frontier models actually perform on real professional finance tasks, graded by the people who do that work for a living.

How is AI used in finance: Examples and use cases

AI is creating the most value in finance where work is repetitive, data-intensive, and can be verified by a human. Gartner's evaluation of generative AI use cases in banking found that organizations see the greatest return from applications that are both high-impact and practical to implement.

Below are some of the most common AI use cases across financial services today:

Fraud detection and security

AI-driven fraud detection flags anomalies in transaction patterns, login behavior, and payment flows faster than rules-based systems because it learns from evolving patterns rather than matching against a fixed list.

The tradeoff is false positives. A model tuned to catch more fraud will also flag more legitimate transactions, which means someone still has to review the queue. Teams that skip that step end up with frustrated customers and blocked payments rather than fewer losses.

Risk modeling and credit scoring

AI expands the range of variables a lender can evaluate when scoring credit risk beyond the handful used by traditional models. This potentially improves accuracy for thin-file borrowers, but explainability remains the constraint.

Regulated lenders must justify any denial, and a model that can't produce a clear reason for its output is a liability regardless of how accurate it is. This is where explainability, not raw predictive power, becomes the deciding factor in whether a model ships.

Algorithmic trading and market analysis

Algorithmic trading uses AI to parse market data, news sentiment, and order flow at speeds and volumes that humans can't match. This is why the use of such technology is concentrated among institutional trading firms with the infrastructure to act on signals in milliseconds.

However, this AI solution is not a practical strategy for retail investors. The IMF's 2024 Global Financial Stability Report frames this kind of AI use across investment decisions, execution, and monitoring as both a performance tool and a source of systemic risk that warrants monitoring.

Forecasting and financial planning

AI models improve forecast accuracy in financial planning and analysis by identifying patterns across more variables and historical data than a spreadsheet model typically holds. This shortens the back-and-forth in planning and close cycles, but it doesn't remove the need for an analyst. Someone still has to sanity-check an output that assumes that past performance predicts the future, especially in volatile conditions that the model hasn't seen before.

Reporting and document automation

Automating tasks such as reconciliation, regulatory reporting, and document review frees finance staff from the highest-volume, lowest-judgment part of the job. Tecban, a Brazilian financial-services infrastructure company, reported up to 30% productivity gains after deploying Microsoft 365 Copilot across 98 use cases and more than 3,000 users.

Banco Inter reported similar efficiency gains after applying Azure AI to release management within its core banking operations. In practice, the time saved is typically reallocated toward exception handling and work requiring judgment calls rather than toward doing nothing.

Customer interactions

Virtual assistants and AI-powered chatbots handle routine account inquiries and basic servicing requests, reducing response times for high-volume, low-complexity customer interactions.

Absa used Microsoft Copilot for tasks such as email summarization and meeting action generation as part of a broader push to enhance operational efficiency in banking. More complex, relationship-driven conversations still route to a person, and the effectiveness of that routing decision matters more than the chatbot's fluency.

What are the benefits of using AI in the finance industry?

The benefits of AI in finance are apparent in multiple key areas:

• Speed: Catching fraud, generating forecasts, or drafting reports in minutes instead of days
• Coverage: Reviewing a larger share of transactions or documents than manual sampling ever could
• Consistency: Applying the same criteria across every case instead of varying by reviewer
• Cost: Reducing the need for headcount growth by automating routine, high-volume work
• Insight: Identifying patterns across datasets too large for a human to analyze

However, AI doesn't deliver these benefits by default. Success depends on using it for the right tasks and measuring how it performs on real business work. .

What Mercor model benchmarks reveal about AI in finance

There are several AI model benchmarks that can help distinguish measurable AI performance in finance from marketing claims.

Mercor's APEX AI productivity index

Mercor's APEX AI productivity index measures frontier AI models on real finance analyst work created and graded by professionals from Goldman Sachs, Morgan Stanley, JPMorgan, and Barclays.

Using investment banking analyst tasks as a proxy for AI performance in professional finance work: Across a sample set of 100 different tasks, the latest results show that leading models, such as GPT 5.3 Codex, GPT 5.4, and Gemini 3.1 Pro, each completes roughly two-thirds of the benchmark successfully.

To give an example, one of the sample tasks asks the model to calculate Coupang's weighted average cost of capital (WACC) for a discounted cash flow analysis by identifying the correct debt and market capitalization, retrieving Treasury yields and cost of debt, calculating beta from historical stock returns, and producing all of the inputs required for the valuation model.

The results suggest that today's leading models can accelerate financial research, modeling, and valuation work, but analysts still need to review calculations, validate assumptions, and make the final recommendation.

Mercor's APEX Agents benchmark

Mercor's APEX Agents benchmark evaluates a more agentic finance workflow to evaluate whether models can complete a realistic, end-to-end finance workflow rather than a single assignment.

Using investment banking analyst agentic tasks: Out of the 160 different workflows, one sample task it evaluates requires the model to modify an existing merger model by adding accretion and dilution sensitivity analyses for both BBDC and TVPG shareholders across multiple bid premiums and cash consideration scenarios while incorporating new operating assumptions.

Unlike APEX, this requires editing an existing financial model, maintaining context across multiple steps, and delivering a finished work product.

Even the highest-scoring model completes fewer than half of these workflows successfully, suggesting that end-to-end finance work remains significantly more difficult than individual analyst tasks.

Bottom line: These results highlight frontier models but longer, multistep workflows remain challenging because models can lose track of context, retrieve the wrong files, or struggle with ambiguous instructions.

The results should be treated as directional guidance rather than a prediction of how AI will perform in your organization. The best way to know AI performance for your own finance organization is by benchmarking models against tasks that reflect your team’s actual work.

Mercor can assist you evaluate AI models and AI agents on your own finance workflows so you can make investment decisions based on measured performance, instead of generic benchmark scores.

How to implement AI in finance?

Implementing AI in finance works best when approached as a sequence rather than a checklist. Start with the problem you want to solve, not the model, and then let each step build on the one before it.

Step 1: Prioritize the finance problems AI should solve

Pick the workflow with the most significant volume and clearest cost of error, not the one that makes the flashiest demo. A reconciliation backlog is a better first target than a customer-facing chatbot because its value and failure mode are both easy to measure.

Step 2: Build a data strategy before choosing a model, a tool or an AI agent

The model is the easy part. Clean, well-labeled financial data, transaction histories, past decisions, and documented outcomes determine whether the model produces something usable. Skipping this step is the most common reason that AI pilots fail.

Step 3: Assess internal AI capabilities and partner needs

Be honest about what your team can build versus what requires outside expertise. Financial workflows carry enough regulatory and reputational weight that a mismatch here can quickly result in a stalled project or compliance issues.

Step 4: Set security standards for financial data

Financial data is regulated, sensitive, and attractive to hackers, so access controls, encryption, and vendor data-handling terms need to be established before any pilot uses real customer or transaction data.

Step 5: Start small, test quickly, and scale what works

Begin with a narrow part of your workflow, measure the outcomes against the manual process baseline, and only expand once success is proven. Scaling a pilot that hasn't been accurately measured is how teams end up with a system that nobody trusts.

AI risks and challenges in financial services

AI adoption in financial services carries several risks and challenges:

• Bias: Models trained on historical data can encode past inequities into future decisions, particularly in credit and lending decisions.
• Explainability: A model that can't justify its output may not meet regulator or customer expectations.
• Data exposure: Third-party AI vendors introduce additional points where sensitive financial data can leak, a concern flagged in a recent Treasury report.
• Model drift: A model's accuracy degrades as market conditions or customer behavior shift away from its training data.
• Cybersecurity: AI systems themselves are a target for hackers.

What governance controls apply to AI in finance?

The frameworks governing AI in finance are evolving and differ between the EU and the U.S.

In the EU, credit scoring and insurance pricing are classified as high-risk under the AI Act, triggering obligations for risk management, data governance, documentation, and human oversight. However, the timeline has already moved once, with the Commission's simplification package reaching political agreement in May 2026.

In contrast, the governance approach in the U.S. encourages the use of AI while setting clear expectations. The Treasury and the Financial Stability Oversight Council have run an AI Innovation Series and published sector risk-management resources rather than issuing prescriptive rules.

Regardless of jurisdiction, it's important for your team to ask, "Which of our AI models make consequential decisions, and can we document how?"

Future trends for AI in finance

Expect AI in finance to continue focusing on productivity gains within existing workflows rather than on new customer-facing products, at least through 2026.

Likely key trends include:

• Wider generative AI use across operations: As already demonstrated by Tecban and Absa
• Tighter governance: As regulators shift their attention to generative AI specifically, not just traditional machine learning
• Growing investment: That builds on the 6.5% budget allocation mentioned in the Gartner report
• More reliance on benchmarks: As buyers increasingly look for real evidence, such as APEX scores, rather than vendor claims

Will AI replace finance professionals?

Finance professionals are unlikely to be replaced by AI, as evidenced by benchmark data.

According to Mercor’s APEX AI leaderboards, no current AI model completes a real investment banking analyst's workflow end to end. Instead, the nature of finance roles is changing, with them incorporating less assembly and reconciliation and more review and judgment. Finance roles are being redefined rather than eliminated.