Mar 6, 2025

getty During the 2010s, numerous startups emerged worldwide—from the United States to Singapore—seeking to transform the accounting industry and significantly reduce labor costs. In the U.S., companies like Pilot and Bench gained traction, while in Singapore and other Commonwealth countries, Osome (where I previously served as head of engineering) and Sleek made similar advancements. The 2020s saw the rise of accounting software-as-a-service (SaaS) solutions such as Pennylane, Puzzle and Kick, where I work as head of analytics. These startups, representing diverse business models, have collectively advanced the industry's understanding of which accounting tasks can be automated through AI and which still require human intervention or additional processing. In this article, I will compare the performance of SaaS and tech-enabled service business models in the accounting industry from two key perspectives: 1. Offering insights into two major business models for AI-driven products in high-skill industries. 2. Examining the primary challenges in achieving full AI automation for high-intelligence occupations. Understanding The Bookkeeping Workflow To ensure consistent validation of financial records before filing tax reports, accountants must complete several essential tasks: Gathering Transactions In most cases, accountants receive all client transactions via software integrated with open banking providers like Plaid or direct integrations with financial institutions. However, some legacy banks remain unconnected to universal providers, forcing clients to manually download and upload bank statements. Validating Transaction Completeness Even in 2025, no open banking provider guarantees 100% accuracy. Many banks deprioritize API reliability, so experienced accountants still rely on PDF statements to verify closing balances. Linking Supporting Documents Categorizing Transactions While AI can automate a large portion of transaction categorization using bank descriptions, supporting documents and client business contexts, achieving 95% automation does not necessarily lead to substantial productivity gains unless other bottlenecks are addressed. Detecting Anomalies Accountants check for irregularities in financial reports to prevent errors or audit triggers (e.g., a restaurant reporting a 10-to-1 income-to-expense ratio). Businesses may need to provide written explanations to tax authorities if anomalies arise. Key Bottlenecks Preventing Exponential Automation Despite advancements in AI and automation, significant obstacles remain: Inaccurate Automated Data Feeds Manual verification is still necessary due to unreliable banking APIs. Banks have little incentive to facilitate external access to their data. AI-based crawlers scanning banking apps could mitigate this issue, but implementation complexities require further exploration. Manual Client Input Clients must upload statements, provide supporting documents and clarify business contexts for accurate categorization. The industry's primary challenge in achieving full automation is minimizing client involvement. Since clients typically prefer to spend as little time as possible on accounting tasks, they often provide information reactively—only when prompted by accountants. Based on my experience, when this process is delegated to a product UI or AI agent, motivation to respond decreases even more. A long-term solution requires software that can extract maximum information without client intervention. Over time, more products will likely integrate automatic data sharing from private sources such as emails and messaging platforms. However, balancing automation with data privacy concerns is crucial. Users will only consent to such data sharing if they are certain their private information is neither stored indefinitely nor misused. In this context, open-source AI agents on blockchain platforms could provide a promising solution, allowing users to verify how their data is processed. Comparing Business Models: Tech-Enabled Services Versus SaaS Let's dive deeper into key differences between business models. Broader Scope Of Client Responsibilities Tech-enabled services often handle tax preparation and filing, whereas SaaS platforms focus primarily on transaction management and financial reporting. Different Client Expectations Service-based customers expect immediate support from dedicated professionals, creating a challenge between maintaining high service quality and preserving profit margins. SaaS customers, on the other hand, prioritize intuitive tools, making scalability more feasible without the same labor cost constraints. Lower Operational Margins Based on my own experience, I've found that while SaaS businesses can achieve operational margins of up to 90%, tech-enabled services typically can't exceed 50% without risking damage to service quality. This is because when service-based companies attempt to increase margins, service quality often declines. Given that late-stage startups spend roughly 30% on sales and marketing and, on top of that, have to allocate a significant portion of the budget into R&D to stay competitive, smaller operational margins make tech-enabled service models less attractive to venture capital. Deferred Reality Check Due To Seasonality Since most tax filings occur at the end of the fiscal year, many clients remain inactive for months after paying for services. This pattern makes it difficult for tech-enabled businesses to assess true margins and service quality until tax season arrives when demand surges. Given these constraints, tech-enabled services have generally underperformed compared to SaaS providers. However, rapid advances in AI may shift this dynamic. As soon as people become comfortable trusting and communicating with AI agents instead of human representatives, tech-enabled services may see a significant improvement in margins. In Summary As AI continues to evolve, the distinction between SaaS and service-based models will likely blur. However, business fundamentals remain unchanged. The key to success lies in solving data ingestion and client interaction challenges without compromising trust or user experience. Future advancements may come from improved open banking standards, AI-driven data extraction or blockchain-powered AI agents ensuring transparency. Ultimately, the drive to reduce human involvement while maintaining accuracy will shape the future of accounting. Companies that master this balance at scale will emerge as industry leaders.