Industry estimates project the global AI in the BFSI market will grow from roughly $35 billion today to somewhere between $240 billion and $370 billion by 2032, depending on the source and scope of the forecast.
Yet when most banks talk about AI, they default to the same pitch: reduce costs, automate processes, cut headcount. That framing misses the bigger story.
The banks that are winning with AI aren’t just saving money. They’re making it. DBS Bank in Singapore generated SGD 750 million in economic value from AI in 2024, across 370 use cases powered by over 1,500 models. Their CEO expects that figure to surpass SGD 1 billion in 2025.
At JPMorgan Chase, the asset and wealth management division reported a 20% year-over-year increase in gross sales between 2023 and 2024, with AI-driven tools cited as a contributing factor in that growth.
These are not pilot programs. These are production systems generating measurable revenue at scale. And they point to a pattern: the AI use cases that generate the most value are the ones embedded deepest into banking workflows.
Here are five of them, and what they mean for banks that don’t have 1,500 models or a $17 billion tech budget.
Over 1.4 billion adults globally remain unbanked. In ASEAN alone, MSMEs face a financing gap exceeding $300 billion. Traditional credit scoring, built on formal credit bureau data, systematically excludes anyone without a conventional financial track record.
Machine learning models trained on alternative data (mobile payments, utility bills, e-commerce transactions, behavioral signals) consistently outperform traditional scoring. Multiple peer-reviewed studies report ML models achieving AUC values of 0.82–0.94, compared to 0.68 or lower for conventional logistic regression. Results vary by dataset and geography, but the direction is consistent across studies.
Early adopters report up to a 20% expansion in qualified borrower pools without a corresponding increase in default rates. DBS demonstrated this at scale: AI-powered SME lending with cash flow monitoring predicted 95% of defaults before they occurred, driving accelerated client acquisition and market-share leadership.
Kindgeek perspective: You don’t need DBS-scale infrastructure to start. An AI credit scoring module can be integrated into an existing core banking system as a standalone microservice, using your own transaction data as the training set. Integration with open banking APIs (PSD2, PSD3) provides additional data enrichment. Typical deployment: 3–4 months from data audit to production scoring.
Wealth advisors and relationship managers spend a disproportionate amount of time on research, data retrieval, and administrative preparation. During volatile markets, the gap between client expectations and advisor response capacity widens dramatically.
JPMorgan’s Coach AI tool allows private client advisors to find relevant research and content up to 95% faster. During the April 2025 market volatility, the tool helped advisors handle a surge in client queries with personalized, data-driven responses. The bank cited AI tools as a contributing factor in its 20% gross sales growth, and projects that advisors will expand their client rosters by up to 50% over the next three to five years.
JPMorgan has gone further than any other bank. Their LLM Suite platform, now used by over 250,000 employees, integrates models from OpenAI and Anthropic into a secure, model-agnostic portal that is updated every eight weeks. It generates investment banking decks in 30 seconds, automates 40% of research tasks, and is now moving into agentic AI: autonomous multi-step agents that handle complex workflows across trading, compliance, and client services. The bank has deployed over 450 AI use cases and expects $2 billion in annual AI-attributed value.
On the trading side, JPMorgan uses reinforcement learning to improve trading win rates from 52% to 63% and has saved an estimated $25 million in slippage costs through optimized order routing. Over 50% of the bank’s electronic FX spot volume above $10 million is now handled by proprietary AI algorithms.
This trend extends beyond individual banks. OpenAI launched Project Mercury in late 2025, recruiting over 100 former investment bankers from JPMorgan, Goldman Sachs, and Morgan Stanley at $150 per hour to train AI models on financial modeling: DCF analysis, M&A valuations, LBO models, and IPO structuring. The goal is to build domain-specific financial AI capable of automating the analytical work that junior bankers spend 80+ hours per week performing.
What this means for mid-sized banks: The largest banks are building proprietary financial LLMs. AI labs are training domain-specific models on real financial workflows. This is creating a capability gap that will widen rapidly. Mid-sized banks that partner with fintech engineering teams to deploy RM copilots, risk assessment AI, and workflow-integrated LLMs now will capture the value before it becomes table stakes.
Kindgeek perspective: An RM copilot connects to your CRM, core banking, and market data feeds via API layer. It retrieves client context, generates talking points, and drafts personalized communication. The architecture is RAG-based (Retrieval-Augmented Generation), running on your internal data with no client information leaving your infrastructure. Key requirement: data governance and model validation aligned to local regulatory expectations (EBA, FCA, MAS).
Most banks still segment customers into broad buckets and serve generic product recommendations. Digital-native competitors offer experiences that feel individually tailored. The gap erodes both conversion rates and customer loyalty.
DBS sends over 1.2 billion personalized nudges to more than 13 million customers annually, powered by over 100 distinct AI/ML algorithms analyzing more than 15,000 data points per customer. Customers who engage with personalized AI recommendations save 2x more, invest 5x more, and are nearly 3x more insured than non-users.
BBVA recorded 391 million customer interactions with AI-powered features in Q1 2024 alone, a 49% increase year-over-year. Customers using AI-driven financial health features report satisfaction rates nearly 24% higher than non-digital users.
Source: BBVA, “What AI algorithms does BBVA use to boost its customers’ finances?,” August 2024.
Kindgeek perspective: A next-best-action engine sits between your data warehouse and customer-facing channels (app, internet banking, CRM). It ingests transaction history, product holdings, and behavioral signals to generate ranked recommendations per customer. A digital bank can start with 5–10 decision models covering core products (deposits, cards, lending, insurance cross-sell) and scale from there. Cloud or on-premise, with API-based integration to existing front-ends.
Legacy rule-based AML systems generate massive volumes of false positive alerts. Compliance teams spend the majority of their time investigating legitimate transactions, while sophisticated criminals learn to operate within the static rules. Global AML compliance costs exceed $180 billion annually.
While AML is typically framed as a cost center, AI transforms it into a business enabler. HSBC deployed Google Cloud’s AML AI platform: the system identified 2–4x as much genuinely suspicious activity while reducing total alert volume by 60%. Detection time dropped to just 8 days after the first alert.
Across the industry, AI-driven AML systems report 30–70% reductions in false positives depending on institution size, data quality, and legacy system baseline. A SAS/ACAMS survey of 850 AML professionals found that reducing false positives remains the top priority for AI/ML deployment (38%).
That is not just better compliance. It is faster onboarding, fewer customer friction points, and lower operational cost per transaction.
Kindgeek perspective: AI AML modules deploy as a layer on top of existing transaction monitoring systems (TMS), gradually replacing rule-based alerting with ML-driven scoring. Key technical considerations: model explainability (SHAP/LIME) for regulatory audits, on-premise deployment options for data sovereignty, and continuous model retraining pipelines. Integration with sanctions screening, PEP databases, and your existing case management workflow.
Customer support in banking is expensive, inconsistent, and often frustrating. First-call resolution rates remain low. Collections processes are adversarial and manual. Both represent significant cost and revenue leakage.
DBS deployed a GenAI-enabled CSO Assistant in 2024: call transcription, summarization, service request generation, and knowledge base lookup. Result: 20% reduction in average handling time while improving response quality. Their internal DBS GPT processes over 250,000 queries monthly.
BBVA deployed ChatGPT Enterprise to 11,000 employees: staff saved nearly three hours per week on routine tasks, with over 80% engaging daily. Staff created thousands of custom GPTs for specific banking workflows. The bank is now extending access to all 120,000 employees globally.
Kindgeek perspective: Conversational AI for banking requires secure deployment (on-prem or private cloud), integration with core banking APIs for transactional capabilities, and multi-language NLP. Architecture: RAG pipeline for knowledge base, intent classification layer, handoff logic to human agents, and audit logging for regulatory compliance. Deployable as WhatsApp/Viber integration, in-app chat, or voice channel.
Across all five use cases, the AI initiatives that generate the most revenue share three characteristics:
Before moving to what a practical implementation looks like, it is worth understanding why most AI initiatives stall between pilot and production:
The banks generating real revenue from AI have solved each of these. That is not an AI problem. It is an engineering and organizational integration problem.
DBS has 1,500 models. JPMorgan spends $17 billion a year on technology. You probably have neither. That does not mean AI revenue impact is out of reach. It means the starting point is different.
Consider a bank with 300,000–1 million active customers, 40–60% digital adoption, 5–10 core retail products, and a limited in-house AI team. Deploying a single AI next-best-action engine focused on cross-sell and retention:
| Metric | Expected Range |
| Cross-sell conversion uplift | +2–4% (from baseline ~8% to ~10–12%) |
| Deposit churn reduction | 1–2% reduction in outflow |
| Digital engagement growth | +10–15% in active feature usage |
| Time to production | 3–6 months (data audit → live) |
| Payback period | 9–14 months |
| Models required to start | 5–10 (not 1,500) |
The key is not building 1,500 models. It is identifying the three to five highest-impact workflows and deploying AI where it directly touches revenue. Start with one use case. Measure it. Scale based on results.
Every successful AI implementation in banking follows the same feedback cycle:
Data Layer → ML Model → Workflow Integration → Human Decision → Revenue Metric → Feedback Loop
Banks that treat AI as a one-time implementation miss this loop entirely. The ones generating real revenue understand that AI is a compounding asset: the longer it runs in production, the better it gets, and the wider the gap becomes between them and competitors still stuck in pilot mode.
Research across Southeast Asia, Latin America, and the Middle East reveals a clear pattern:
| Approach | % Banks | Implication |
| Build in-house | ~10% | Requires JPMorgan-scale budgets ($17B/year) and thousands of in-house data scientists. Realistic only for the largest global banks. |
| Buy off-the-shelf | ~56% | Good for exploration. But generic products rarely achieve the workflow integration that drives revenue. |
| Partner to scale | ~34% | Banks that combine domain knowledge with a fintech engineering partner achieve the fastest time-to-revenue and lowest integration risk. |
A management consultancy can advise on AI strategy. A pure-play AI vendor can provide models. But neither will integrate those models into your core banking system, connect them to your payment rails, validate them against your regulatory framework, and maintain them in production.
That is the role of a fintech engineering partner: a team that has spent years inside banking systems, understands core banking architecture, payment processing, compliance requirements, and the operational reality of deploying AI in a regulated environment.
What Comes Next
AI in banking is not a future trend. It is a current reality with measurable revenue impact at the institutions that have moved past pilots into production.
The question for most banks is not whether to invest in AI. It is where to start, what to measure, and who will help them build it right.
Kindgeek is a fintech engineering company with 8+ years of experience building banking systems, payment platforms, and neobanks for clients globally. Our team of 200+ engineers specializes in AI-powered fintech solutions: credit scoring engines, personalization systems, compliance automation, RM copilots, and end-to-end digital banking platforms. We deploy on cloud and on-premise, with regulatory model validation for EBA, FCA, MAS, and other frameworks.
We offer a 4-week AI Revenue Audit for digital and regional banks. One focused engagement. Clear deliverables: data readiness assessment, top-3 revenue use cases mapped to your workflows, implementation roadmap with timeline and KPIs.
Pursuing digital transformation, responding to market pressures, working on customer retention – modern businesses have…
Earlier this year, Kindgeek launched Easyflow - a comprehensive AI automation platform, designed to streamline…
In the fast-paced realm of fintech, the pursuit of seamless user experiences harmonised with stringent…
Decentralized finance, or DeFi, has emerged as a truly disruptive force within Web3, seeking to…
In today’s fast-changing digital finance environment, fintech applications are redefining the way people manage money,…
HyperJar is a world-first payments technology embedded in a digital wallet. It’s the only app…