Earning Interest: AI Fintech Tools
原文链接: synaptic.com/resources/e…
Introduction 引言
生成式人工智能(Generative AI)已经开始在金融科技(Fintech)领域释放真正的价值。领先的公司已经开始部署这项技术来提高效率——例如,Brex公司的AI助手在费用管理方面每年节省了大约4,000小时的人工劳动。像彭博(Bloomberg)和摩根大通(JPMorgan)这样的金融巨头正在开发专有的生成式AI工具——彭博GPT目前专注于改善金融自然语言处理(NLP)任务,如情感分析、命名实体识别、新闻分类和问题回答,而摩根大通的IndexGPT则帮助投资者分析证券。
AI在金融科技领域并不是新鲜事物,到2020年,近90%的金融科技公司在一定程度上利用了AI/机器学习(ML)。然而,生成式AI提供了一条新途径,用于提高那些仍然繁琐、文档密集型的金融工作流程的生产力,比如账单和发票处理,或者合规性审查。例如,一家大型美国分销公司正在使用Vic.ai处理76%的发票,无需人工干预,平均每张发票的处理时间从2.3分钟缩短到36秒。
尽管全球金融科技投资普遍下滑,但以AI为重点的金融科技公司在2023年仍成功吸引了超过121亿美元的资金。接下来将为您提供该领域的概述。
Generative AI has started unlocking real value in Fintech. Leading firms are already deploying the technology to drive efficiencies - Brex's AI assistant for expense management saves approximately 4,000 hours of manual effort annually. Financial giants like Bloomberg and JPMorgan are developing proprietary generative AI tools – BloombergGPT currently focuses on improving financial NLP tasks like sentiment analysis, named entity recognition, news classification, and question answering, while JP Morgan’s IndexGPT aids investors in analyzing securities.
AI is not new in Fintech, nearly 90% of fintech firms were leveraging AI/ML to some degree in 2020. Generative AI however, provides a new path to enhance productivity on the still tedious, documentation-heavy financial workflows like billing & invoicing, or compliance reviews. For instance, a large US distribution company is processing 76% of invoices without human intervention using Vic.ai, averaging 36 seconds per invoice instead of 2.3 minutes.
Despite a general downturn in global fintech investment, AI-focused fintech managed to attract over $12.1 billion in funding in 2023. Read on for an overview of the space.
Use Cases
Use Cases: Back-office automation and chat-based financial advice are prominent LLM use-cases
应用案例:后端办公室自动化和基于聊天的财务咨询是大型语言模型(LLM)的突出应用案例。
金融科技市场为广泛的客户群体提供服务——从个人到公司再到政府——涵盖了各种应用案例。金融科技长期以来一直利用人工智能/机器学习(AI/ML)技术来改进算法交易和风险评估等领域。大型语言模型(LLMs)引入了一波新的创新浪潮,特别是在自动化重复性、以语言为中心的任务方面。
The fintech market serves a wide range of customers – from individuals to companies to governments – across various use cases. Fintech has long leveraged AI/ML to improve areas like algorithmic trading and risk assessment. Large language models (LLMs) introduce a fresh wave of innovation, especially in automating repetitive, language-focused tasks.
金融后端办公室任务涉及大量文件、账单、发票的手动处理——非常适合由大型语言模型( LLM )驱动的自动化。
金融团队的工作量包括处理大量文件,包括财务数据、交易、合规文件、账单和发票。LLM擅长解析、理解和管理这些大量的文本数据。它们可以自动从财务文件中提取重要细节,分类和总结交易信息,并从各种数据点编制详细报告。LLM还增强了客户互动和查询解决,通过AI聊天机器人和知识助手解释和响应客户查询——为金融科技专业人员节省时间,让他们专注于更关键的任务。
在不直接处理敏感财务数据的应用中,LLM的采用更为容易
与医疗保健行业一样,金融科技行业处理私人和敏感的财务细节,需要监管监督和合规。不直接涉及此类敏感信息的使用案例——如个人财务管理、投资研究、后端办公室工作流程(如费用跟踪、文档处理)——对LLM的采用更为直接,因为它们可能需要更少的法规。
LLM 采用的第一波正在以下应用案例中显现:
CFO 工具(会计、簿记、费用管理) :生成式AI可以通过自动化会计、交易编码、费用跟踪以及收据和备忘录的创建来增强财务运营。它还可以理解模式并识别异常,提供准确的预测,并提供个性化洞察。这减少了金融专业人员的手动工作量。一些参与者,如Basis、Arkifi、Vic.ai,正在为金融团队提供“副驾驶模型”。
生成式AI可以处理、总结和从大量财务文件(如年报、财务报表和收益电话会议)中提取相关数据,然后这些数据可以被输入到数据库中——显著减少了人工工作量。像Sensible.so、Rossum.ai、Attri.ai这样的工具正在使用生成式AI从财务文件中提取数据并加快文档处理。
入职、风险与合规(欺诈、承保、 KYC /KYB): 入职和反洗钱流程涉及大量的文件工作。生成式AI可以自动化合规审查、缺失文件的请求和制裁警报。例如,Greenlite利用生成式AI自动化客户入职的合规审查。Coris AI,一个针对金融科技公司的风险API,使用GPT-4自动确定商户的分类代码,这对于监控风险和合规至关重要。
LLM可以通过利用以前未被利用或手动分析的语言资源来增加额外的检查层。例如,Flagright使用GPT-4通过公共平台如Crunchbase和LinkedIn监控商户档案——评估商户的声誉、客户群和财务健康状况。此外,Flagright使用GPT根据标记的活动创建可疑活动报告(SARs),并支持风险分析师使用AI叙事作家,简化常规文档和沟通任务。
投资研究:金融研究正在通过生成式AI得到增强,允许投资者通过聊天界面对股票、公共和私人市场进行研究。StockGPT,一个由AI驱动的金融研究工具,让投资者可以搜索所有标普500和纳斯达克公司的收益发布、财务报告和其他基本信息。FinChat提供了超过100,000家全球上市公司的财务数据,用户可以从中搜索,还包括股票筛选器、数据可视化、竞争对手比较和投资组合仪表板。
个人理财: 个人财富管理和财务咨询工具正在使用生成式AI提供更个性化的沟通和建议。例如,Wally使用GPT总结支出,根据个人目标提供财务建议,并提供财务习惯的洞察。UStreet是一个AI财富顾问,创建个性化的退休计划,提供投资组合审查和税收节省建议。
Finance back-office tasks involve significant manual handling of documents, bills, invoices - well suited for LLM-driven automation
The workload of a finance team involves handling a large volume of documents, including financial data, transactions, compliance documents, bills, and invoices. LLMs excel at parsing, understanding, and managing such vast amounts of textual data. They can automate the extraction of important details from financial documents, categorize and summarize transaction information, and compile detailed reports from various data points. LLMs also enhance customer interaction and query resolution, with AI chatbots and knowledge assistants interpreting and responding to customer queries – freeing up time for fintech professionals to focus on more critical tasks.
LLM adoption is easier in applications not dealing with sensitive financial data directly
Like the healthcare sector, the fintech industry handles private and sensitive financial details, necessitating regulatory oversight and compliance. Use cases that don't directly involve such sensitive information – such as personal finance management, investment research, back-office workflows like expense tracking, document processing – are more straightforward for LLM adoption, as they may require fewer regulations.
The first wave of LLM adoption is being seen in the following use cases:
CFO Tooling (accounting, bookkeeping , expense management ): Generative AI can enhance financial operations by automating tasks such as accounting, transaction coding, expense tracking, and the creation of receipts and memos. It can also understand patterns and identify exceptions, provide accurate forecasts, and offer personalized insights. This reduces the manual workload for finance professionals. A number of players such as Basis, Arkifi, Vic.ai are offering a “copilot model” for finance teams.
Gen AI can process, summarize and extract relevant data from a bulk of financial documents (e.g. annual reports, financial statements, and earnings calls), which can thereafter be ingested into databases – significantly reducing human workload in doing so. Tools like Sensible.so, Rossum.ai, Attri.ai are using Gen AI to extract data from financial documents and expedite document processing.
Onboarding, Risk & Compliance (fraud, & underwriting, KYC /KYB): Onboarding and AML processes involve a hefty amount of paperwork. Gen AI can automate compliance reviews, requests for missing docs, and sanction alerts. For instance, Greenlite utilizes generative AI to automate compliance reviews for customer onboarding. Coris AI, a Risk API for fintech companies, uses GPT-4 to automatically determine a merchant’s classification codes, which are essential for monitoring risk and compliance.
LLMs can add an additional layer of checks by leveraging language-based sources that were previously untapped or manually analyzed. For instance, Flagright employs GPT-4 to monitor merchant profiles using public platforms like Crunchbase and LinkedIn – assessing a merchant's reputation, clientele, and financial health. Additionally, Flagright uses GPT to create Suspicious Activity Reports (SARs) based on flagged activity and supports risk analysts with an AI Narrative Writer, streamlining routine documentation and communication tasks
Investment Research: Financial research is being enhanced with Gen AI, allowing investors to carry out research on stocks, public and private markets on a chat interface. StockGPT, an AI-powered financial research tool, lets investors search through earnings releases, financial reports, and other fundamental information for all S&P 500 and Nasdaq companies. FinChat offers financial data on 100,000+ global public companies that users can search from, along with a stock screener, data visualizations, competitor comparisons, and portfolio dashboarding.
Personal Finance: Tools for personal wealth management and financial advice are using Gen AI to give more personalised communication and advice. For example, Wally uses GPT to summarize expenses, provide financial advice based on personal goals, and provides insights into financial habits. UStreet is an AI wealth advisor that creates personalised retirement plans, offers portfolio reviews and tax saving advice.
Market Map: Fintech players – from incumbents to nascent startups – are innovating with Gen AI
市场地图:金融科技参与者——从老牌企业到新兴初创公司——正在利用生成式人工智能(Gen AI)进行创新。
老牌企业正在为其团队构建专有的 AI 模型
对相关金融数据进行微调模型对于使用大型语言模型(LLMs)创建相关工作流程至关重要。这使得拥有数据和基础设施来训练专有模型的老牌企业处于优势地位——彭博GPT、摩根大通的IndexGPT、Intuit的GenOS、摩根士丹利的AI助手都是由老牌企业构建的。
成熟的 初创公司 正在积极利用生成式人工智能增强其产品
已经建立了客户基础的成熟公司正在使用生成式人工智能改进其产品。提供首席财务官工具的公司正在将AI助手集成到他们的产品中,使用户能够使用自然语言进行交互,以获得财务数据洞察——Vise Intelligence和PilotGPT是主要的例子。投资研究正在通过生成式人工智能得到加强——Public.com的Alpha提供实时投资背景和关于市场动向的主动警报给投资者。
新进入者正在创新 首席财务官 工具、合规和研究工具
金融科技领域的新进入者主要开发能够简化任务并提高金融专业人员效率的工具,释放他们的时间用于更高价值的工作。这些工具自动化了会计、簿记和管理费用等常规任务,以及与合规相关的活动,如检测欺诈、承保和客户入职。
人工智能创新将不得不应对金融科技对数据隐私和准确性的高标准
金融服务中的数据质量、隐私和准确性标准是商业世界中最高的之一。因此,生成式人工智能的采用目前正发生在不会直接影响重大财务决策的选择性领域,因此不需要通过严格的监管流程。
Incumbent firms are building proprietary AI models for their teams
Fine tuning models on relevant financial data is critical to creating relevant workflows with LLMs. This puts established firms at an advantage – BloombergGPT, JPMorgan’s IndexGPT, Intuit’s GenOS, Morgan Stanley’s AI Assistant have all been built by incumbent firms with the data and infrastructure needed for training proprietary models.
Established startups are actively enhancing their offerings using Gen AI
Established firms that have already built a customer base are improving their products with Gen AI. Companies offering CFO tooling are integrating AI Assistants into their products, enabling users to interact using natural language for financial data insights – prime examples are Vise Intelligence and PilotGPT. Investment research is being boosted with Gen AI – Alpha by Public.com provides real-time investing context and proactive alerts about market movement to investors.
New entrants are innovating with CFO tooling, compliance and research tools
New entrants in fintech are primarily developing tools that streamline tasks and enhance efficiency for finance professionals, freeing up their time for higher-value work. These tools automate routine tasks like accounting, bookkeeping, and managing expenses, as well as compliance-related activities such as detecting fraud, underwriting, and customer onboarding.
AI innovation will have to navigate fintech’s high bar for data privacy and accuracy
The bar for data quality, privacy and accuracy in financial services is among the highest in the business world. Gen AI adoption, therefore, is currently taking place in selective areas which do not directly impact major financial decisions and therefore need not navigate strict regulatory processes.
Funding: Bulk of funding going to established players, speculative bets on new age tools
融资:大部分资金流向了老牌玩家,对新时代工具的投机性投注
尽管全球金融科技市场的融资在2023年达到了五年来的最低点,但投资者对金融科技领域内人工智能的兴趣依然强劲。
Despite the funding for the global fintech market hitting a five-year low in 2023, investor interest in AI within fintech has remained strong.
在过去两年中,超过75%的新兴 初创公司 已经筹集了资金
2022-2024年间,新兴玩家的融资表现强劲。在AI原生玩家领域,已经完成了30多笔交易,尽管规模较小。
一些值得注意的交易包括:
-
Slope在2023年推出了SlopeGPT(第一个由GPT驱动的支付风险模型)后,筹集了3000万美元的资金。Sam Altman是Slope的天使投资人。
-
Federato,一个由AI驱动的承保平台,在2023年的B轮融资中筹集了2500万美元。
More than 75% of emerging startups in the space have raised money in the last 2 years
Funding for emerging players has been robust in 2022-24. There have been 30+ deals, albeit small ones, across the sector for AI-native players.
Some notable deals:
-
Slope raised a $30M round in 2023 following their recent product launch of SlopeGPT, the first payments risk model powered by GPT. Sam Altman is an angel investor in Slope.
-
Federato, an AI-powered underwriting platform, raised $25M in Series B funding in 2023.
对新兴玩家的几轮投机性投资,平均每轮约500万美元
投资者对金融科技领域的AI持谨慎乐观态度,对新兴玩家进行了投机性投注。尽管金融科技是一个高增长领域,但其监管环境和AI整合的初级状态创造了一个风险概况,倾向于较小的初始投资。这使得初创公司在吸引更大资金轮之前能够证明其可行性并获得监管批准。
Several speculative rounds, averaging around ~5M, on emerging players in the space
Investors are being cautiously optimistic about AI in fintech, making speculative bets on emerging players. Although fintech is a high-growth area, its regulatory landscape and the nascent state of AI integration create a risk profile that favors smaller initial investments. This allows startups to prove viability and secure regulatory approval before attracting larger funding rounds.
在2022-2024年期间,将生成式人工智能(Gen AI )作为其核心产品基本组成部分的成熟 初创公司 获得了最大的融资交易:
-
Pigment是一个B2B财务规划平台,在2023年完成了8800万美元的C轮融资。这笔资金是在他们推出PigmentAI一个月后获得的,PigmentAI是一个生成式AI助手,可以回答有关财务数据的问题,自动化某些工作流程,生成摘要和洞察,并允许进行情景规划。
-
Vic.ai - 一个用于自主会计的AI平台,在2022年12月获得了5200万美元的C轮融资。
-
Puzzle.io,一个生成式AI会计平台,在2023年筹集了3000万美元。Puzzle将静态会计数据转换为实时财务洞察。
Established startups that embedded Gen AI as a fundamental aspect of their core products secured the largest funding deals in 2022-24
-
Pigment is a B2B financial planning platform that raised $88M Series C round in 2023. This funding comes a month after they launched PigmentAI, a Gen AI assistant that answers questions about the financial data, automates certain workflows, generates summaries and insights and allows scenario planning.
-
Vic.ai - AI platform for autonomous accounting received $52M Series C funding in Dec’22.
-
Puzzle.io, a Gen AI accounting platform, raised $30M in 2023. Puzzle transforms static accounting data into real-time financial insights.
Website Traction: Steady growth in value-driven engagement after AI hype tempered in Q3’23
网站吸引力:在2023年第三季度AI热潮降温后,以价值为导向的用户参与稳步增长
对人工智能金融科技的兴趣与最初的AI热潮向更具价值导向的参与转变的更广泛AI趋势相一致——类似于人工智能客户支持工具部门或人工智能医疗保健工具部门。在2023年第一季度达到高峰后,该领域的网站流量已趋于稳定。
Interest in AI Fintech has been commensurate with the broader AI trend of initial hype transitioning to more value-driven engagement – similar to the AI customer support tools sector or the AI healthcare tools sector. After a peak in Q1’23, website traffic across the sector has stabilized.
B2B tools have seen the most QoQ growth in website traction; B2C tools have seen great volume but deceleration
B2B 工具在网站吸引力方面看到了最大的季度 环比增长 ;而 B2C 工具虽然流量巨大,但增长放缓。
消费者金融工具如Magnifi、FinChat在2023年第四季度看到了显著但逐渐下降的网络流量,这表明用户参与从好奇心驱动转向寻求价值。
Magnifi是为个人投资者设计的AI副驾驶。Magnifi分析连接的账户以识别过高的费用、风险暴露和改进机会。Magnifi的对话式AI是唯一受SEC监管的AI,提供协助、教育以及买卖证券的能力。
FinChat是为投资者提供公共股权研究平台,类似于金融领域的ChatGPT。FinChat通过互动聊天界面,从其金融信息数据库中提供全面的洞察和正确引用。它在2023年4月推出后的一个月内积累了10万用户。
Consumer finance tools like Magnifi, FinChat saw significant but declining web traffic in Q4’23, indicating a transition from curiosity-driven to value-seeking user engagement.
Magnifi is an AI-copilot for individual investors. Magnifi analyzes connected accounts to identify excessive fees, risk exposure, and improvement opportunities. Magnifi’s conversational AI is the only SEC-regulated AI that provides assistance, education, and the ability to buy and sell securities.
FinChat is a public equity research platform for investors, similar to ChatGPT for finance. Finchat provides comprehensive insights from its database of financial information with proper citations through an interactive chat interface. It amassed 100K users within a month of its launch in April 2023.
一些在网站流量上实现显著季度环比增长的初创公司包括:
-
Vise在2023年第四季度的网页流量环比增长超过了500%。Vise利用AI来优化金融顾问的投资管理工作,处理顾问-客户互动的所有方面,从投资组合个性化到持续洞察,使顾问能够更多地专注于客户关系。该公司即将推出“Vise Intelligence”,这是一个由大型语言模型(LLM)驱动的界面,能够根据投资组合数据、交易细节、客户偏好和投资策略生成洞察。
-
Kasisto,一个用于银行业务的生成式和对话式AI平台,在2023年第四季度实现了101%的环比增长。Kasisto是KAI-GPT的创造者,这是一个基础性的金融科技LLM,用于构建用于知识管理和客户服务的对话式AI。KAI为First Financial Bank的CD账户带来了27%的增长,为Meriwest Credit Union成员的盈利能力带来了30%的增长。
-
Unit21是一家为金融科技提供风险和合规基础设施的公司,在2023年第四季度的网页流量环比增长了77%。Unit21提供了一个AI副驾驶和名为“Ask your Data”的对话式AI,合规分析师使用这些工具从警报中获取最相关信息的方向,并通过对数据进行简单的英语查询来获得可操作的洞察。Unit21帮助LINE的合规团队将其误报解决方案自动化了60%。
Some startups with extraordinary QoQ growth in website traffic:
- Vise saw a QoQ web traffic growth of >500% in Q4’23. Vise employs AI to optimize investment management for financial advisors, handling all facets of advisor-client interactions, from portfolio personalization to ongoing insights, allowing advisors to focus more on client relations. The company is set to launch “Vise Intelligence”, an LLM-powered interface that generates insights based on portfolio data, trade details, client preferences, and investment strategies.
- Kasisto, a generative and conversational AI platform for banking, saw a QoQ growth of 101% in Q4’23. Kasisto is the creator of KAI-GPT, a foundational fintech LLM that is used to build conversational AI for knowledge management and customer service. KAI has led to 27% growth in CD accounts for First Financial Bank, and 30% growth in profitability for Meriwest Credit Union members.
- Unit21 is a risk & compliance infrastructure for fintech that saw 77% QoQ growth in web traffic in Q4’23. Unit21 offers an AI Copilot and conversational AI called “Ask your Data” that compliance analysts use to get direction on the most pertinent information from alerts, and get actionable insights by querying the data in simple English language. Unit21 helped LINE’s compliance team automate its false positive resolution by 60%.
Looking Forward 展望未来
-
生成式人工智能正被稳步采用以自动化金融后端办公室工作流程
特别是大型语言模型(LLMs),通过自动化诸如发票处理(Vic.ai)、数据提取和文档解析(Sensible)、承保(Sixfold.ai)以及客户入职时的合规审查(Greenlite)等重复性、非结构化任务,显著影响了金融科技。生成式AI也被用于通过基于聊天的AI助手增强投资研究(FinChat)和个人财务建议(Magnifi)。
-
老牌企业、金融科技独角兽以及 AI 原生 初创公司 正在利用 生成式AI 进行创新
老牌公司正在为其金融团队构建专有的大型语言模型,如彭博GPT、摩根大通的IndexGPT和摩根士丹利的“AI@摩根士丹利助手”——利用他们庞大的数据资源和监管专业知识。许多成熟的初创公司正在将生成式AI添加到他们当前的产品中——Vise和Unit21资本充足,市场吸引力快速增长。新的金融科技初创公司正在设计工具,自动化常规的财务功能,如会计(Basis)、文档处理(Noetica)和合规任务(SixfoldAI、Greenlite),同时增强投资研究(Finchat、Portrait Analytics)。
-
预计更大份额的资金将继续流向市场吸引力强的成熟 初创公司
在2022-2024年,对AI金融科技的投资一直是实质性的,整个领域有50多笔交易。一些最大的融资轮次流向了将AI作为其产品核心的初创公司,如Puzzle和Pigment。Vise、Kasisto和Unit21市场吸引力强劲,很可能是即将到来的融资轮次的候选公司。Pigment在不久的将来成为独角兽的有力竞争者。
-
开源与专有 AI 模型之间的较量值得关注
像彭博这样的老牌玩家投入了大量资源构建像BloombergGPT这样的专有模型,利用私有财务数据和定制训练。然而,金融科技AI模型FinGPT的成功打乱了这一等式。尽管据报道FinGPT的训练成本低于100美元,但在某些领域它的表现优于500亿模型BloombergGPT。有趣的是,我们将看到FinGPT的成本效益是否能在与像BloombergGPT这样拥有高质量专有数据访问权的闭源模型的竞争中占据优势。
- Generative AI is being steadily adopted to automate back-office workflows in finance Gen AI, especially LLMs, are significantly impacting fintech by automating repetitive, unstructured tasks such as invoice processing (Vic.ai), data extraction and document parsing (Sensible), underwriting (Sixfold.ai) and compliance reviews at customer onboarding (Greenlite). Gen AI is also being used to enhance investment research (FinChat) and personal finance advice (Magnifi) with chat-based AI assistants.
- Incumbents, fintech unicorns, as well as AI-native startups are innovating with Gen AI Incumbent firms are building proprietary LLMs for their finance teams, such as BloombergGPT, JP Morgan’s IndexGPT, and Morgan Stanley’s ‘AI @ Morgan Stanley Assistant' – leveraging their vast data resources and regulatory expertise. Many established startups are adding Gen AI to their current offerings – Vise and Unit21 are well-capitalised and seeing rapid growth in market traction. New fintech startups are devising tools that automate routine finance functions such as accounting (Basis), document processing (Noetica), and compliance tasks (SixfoldAI, Greenlite), alongside enhancing investment research (Finchat, Portrait Analytics).
- Expect larger share of funding to continue going towards established startups seeing strong market traction Investment in AI fintech has been substantial in 2022-24, with 50+ deals across the sector. Some of the largest rounds have gone to startups that are making AI fundamental to their offering, such as Puzzle and Pigment. Vise, Kasisto, and Unit21 are seeing strong market traction, and are likely candidates for upcoming funding rounds. Pigment is a strong contender for unicorn status in the near future.
- The battle of open vs proprietary AI models will be worth watching out for Established players like Bloomberg have poured significant resources into building proprietary models like BloombergGPT, leveraging private financial data and custom training. However, the success of FinGPT, an open-source fintech AI model, throws a wrench into the equation. Despite a training cost reportedly under $100, FinGPT outperforms the 50B model BloombergGPT in certain areas. It will be interesting to see if FinGPT's cost-effectiveness can tip the scales against closed-source models like BloombergGPT that have access to high quality proprietary data.