Unlocking opportunities with AI-driven lending

A little over a year ago, Citi Ventures wrote about the attractiveness of investing in embedded lenders. We now believe Gen AI-powered software for lenders is another increasingly attractive category.

Our thesis is straightforward: Gen AI is coming for lending, and the question is not if but how and when. As horizontal LLMs continue to expand their capabilities, the vertical-specific complexity of lending is making it one of the most attractive categories for application layer-value creation. Where lending software is a system of record, deeply embedded in specific workflows, calibrated to complex regulatory, audit and governance requirements, and mission-critical, we see opportunity.

Emerging companies are addressing different parts of the value chain:

• Distribution: Gen AI can help identify potential borrowers before they know they need a loan. Conversational interfaces can guide borrowers through the process and automatically extract information from uploaded documents such as bank statements and identification. Casca and Lama are helping banks meet borrowers whenever and wherever they wish. Parlay and Fuse extend origination and pre-qualification.
• Underwriting: Traditional underwriting relies on a narrow set of data points, like credit scores. AI models can now analyze both structured and unstructured data to produce far more accurate risk assessments and expand credit boxes without increasing risk. Zest AI, a Citi Ventures portfolio company, delivers AI-powered automated underwriting. EnFi deploys agents across deal screening and underwriting, and Taktile lets lenders build and manage AI decisioning workflows across underwriting and onboarding.
• Servicing: Companies are also using AI to monitor loans, enabling early detection of potential defaults and more personalized interactions that support repayment and upsells. Salient has built automated loan servicing workflows, Spinwheel, a Citi Ventures portfolio company, is strengthening the underlying consumer credit data infrastructure, Casap, is an agentic AI company and dispute management platform purpose-built for banks, credit unions and fintechs, and Valon is an AI partner for mortgage servicing to help customers manage their loans.

Adoption of Gen AI software in lending may not happen uniformly, as it will track the complexity of each lending category. Consumer and BNPL lending (small tickets, ample behavioral data, lighter compliance) will move fastest. Auto and asset lending, with standardized documents but more operational steps, will follow. SMB and commercial lending introduce KYB and cash-flow complexity, and mortgage (long lifecycle and heavy documentation and regulation) will take time to evolve, even as innovators like Vesta gain momentum.

These adoption curves don’t come without real challenges, as AI can amplify risk if models generate errors, causing mistakes to scale quickly. Regulatory frameworks are evolving more slowly than technology, creating uncertainty, particularly in complex categories.

We firmly believe, however, that incumbent lenders who leverage the speed, accuracy, and efficiency promised by the emerging class of lending software providers to amplify their distribution, data and capital advantages, will be well-positioned to expand their market share. We believe the software companies that enable these incumbents to bring Gen AI into their processes could be among the most valuable businesses at the application layer.