Why Compliance Costs of AI Commercialization Maybe Holding Start-Ups Back
What are the compliance costs of AI commercialization? We explore the growth impacts of such financial burdens on AI start-ups through a field deployment perspective.
While Artificial Intelligence (AI) technologies are progressing fast, compliance costs have become a huge financial burden for AI startups, already constrained by tight research & development (R&D) budgets. Complex regulatory processes, that vary across the globe give well-established technology firms an upper-hand over resource-constrained startups.1 If this continues, giant tech firms may monopolize AI technologies, phasing out start-ups in this field. How do compliance costs typically hinder a start-up’s ability to compete with larger tech firms in AI commercial operations? We explore these challenges using insights from our first-hand experience with PerceptIn, an AI startup in the autonomous driving space.
Based on the OECD Regulatory Compliance Cost Assessment Guidance, we quantitatively compare the financial vulnerability of tech giants versus AI startups.2 Conducting a financial statement simulation provides a glimpse at the impact of changes in compliance costs.
We found that start-ups’ operating margins are significantly impacted by compliance costs, in contrast to tech giants. Moreover, compliance costs may be concealed in financial reports, leading to potential underestimations of true costs. Current estimates of AI compliance costs may be insufficient.
The simulation in Figure 1 assumes that gross profit and expense are a fixed percentage of revenue. The compliance cost is split into a fixed cost regardless of revenue and a variable cost that takes 5% of revenue. When the fixed compliance cost increases by 200%, the operating margin of the startup changes from 13% to -7%, causing the firm to lose money. In contrast, such a change only causes a slight dip in the operating margin for tech giants.
According to Accounting Standards, companies are not required to disclose compliance cost explicitly in financial reports when the cost is not material.3 Thus, compliance costs may become hidden by nature and classified into other categories, such as Research and Development expenses and other administrative expenses. Lacking first-hand information, analysts on the macro-economic level tend to underestimate the costs of AI regulations.
The Compliance Trap
AI is a highly regulated industry, but unfortunately, there is no standardized AI regulation framework.1 Most AI entrepreneurs may not even be aware of the existence of compliance costs, let alone the severe impact compliance costs may have on the company’s overall financial health.
First, unlike R&D budgeting, due to varying AI regulatory frameworks across the globe or even across multiple regions within a country, there is no standard method to budget for AI compliance costs. Even estimating the range of AI compliance costs is infeasible.
Second, even with an AI compliance budget, the actual costs may significantly deviate from the budget. AI startups often encounter new compliance issues as they progress through commercialization. In addition, opportunity costs arise as regulators inspect AI products on safety and privacy issues, causing delays in commercial deployments.
Third, varying AI regulations often introduce indirect costs. For instance, a strict compliance environment demands engineers deal with regulatory issues such as responding to various compliance technical inquiries instead of spending time developing products. Such a shift of focus does not reflect in financial reports, as engineers’ costs are categorized as R&D costs.
A Field Deployment Perspective
With more than six years of first-hand experience in deploying commercial autonomous driving services, we delve into the details of compliance costs from a field deployment perspective. Through this, we aim to raise awareness of the adverse impact of the lack of standardized AI regulations.
PerceptIn is an autonomous driving startup founded in California in 2016. It offers autonomous micro-mobility solutions to customers from the United States, European Union, and Asia. The company only budgeted for ordinary compliance expenses, such as the direct labor cost of a safety driver on board and the equipment cost of a waterproof surveillance camera. While facing a broad spectrum of regulatory obstacles across different countries, PerceptIn fell into the compliance trap.
1) No Standard Method to Budget
The AI regulation framework in China was blurry, and when the company first launched the autonomous micro-mobility project in China, it was impossible to budget for compliance costs. For instance, in the absence of relevant regulation, the company had to develop its own testing plan to obtain deployment approval. Without detailed testing standards, the company spent $25,000/month to simulate real-world scenarios in initial stages for testing and demonstration purposes. The $300,000 annual cost was not included in the company’s original budget and imposed a heavy burden on the company’s balance sheet.
2) Deviation from Compliance Budget
PerceptIn was invited to launch an autonomous driving pilot program in a European city. Before rolling out the project, the company was asked to prepare a risk mitigation plan for 40 different scenarios. To cope with the regulatory process, the R&D team shifted its focus to responding to scenarios-based functional specifications and supplemented the mitigation plan with real-time data. During the project budgeting phase, the company had prepared 20 person-days (or one person’s working time for a day) to cope with the AI regulatory process. Despite an original budget of $10,000, the process ultimately consumed 400 person-days and $200,000 to complete.
3) Indirect Costs
Japan is famous for its rigid structure in organizations. To gain the confidence of the Japanese government, the company first debuted a marketing campaign to promote safe autonomous micro-mobility services in a smart city project.5 With a successful local case and globally established brand, the company then discussed operation permits with the Ministry of Land, Infrastructure, Transport, and Tourism.6 The preparation and initiation of the project took over 24 months, costing $500,000 in promotion, material preparation, and marketing campaigns. Traditionally marketing activities were not meant to cope with compliance requirements. However, in this case, marketing was a tool to convince the regulatory body to accelerate autonomous driving operation permits.
In the case of PerceptIn, the compliance cost of one deployment project is $344,000 on average, whereas the average R&D cost is around $150,000, making the compliance costs 2.3 times the amount of R&D costs, far exceeding the 17.6% estimation of the European Union’s Artificial Intelligence Act.
The root cause of the compliance trap is the lack of a standardized AI regulatory framework. A new business model of Compliance-as-a-Service (CaaS) can specialize in dealing with varying AI regulatory frameworks and thus amortize compliance costs across different AI startups. In addition, CaaS reduces the friction between regulatory bodies and AI startups by providing an interface to compile legal terms into technical and operational plans. With the new business model, AI entrepreneurs can adequately budget for compliance when evaluating the potential of an innovative idea.
Weiyue Wu is Chief Operating Officer of PerceptIn, an autonomous driving startup founded in 2016. At PerceptIn, she has been in charge of commercial autonomous driving service deployments in the United States, Europe, Japan, and China. Before PerceptIn, she served as Investment Director of Oxford Seed Fund and Investment Advisor of ARM Accelerator. She began her career as a Multi-National Corporation Compliance Auditor at KPMG and a Senior Automobile Consultant at Deloitte. She received her MBA from the University of Oxford. She is a founding member of IEEE Special Technical Community on Autonomous Driving Technologies, a Certified Public Accountant and a practicing lawyer in China.
Dr. Shaoshan Liu, MC/MPA 20202, is CEO of PerceptIn and has commercially deployed autonomous micro-mobility services in the United States, Europe, Japan, and China. He is the Asia Chair of IEEE Entrepreneurship. Dr. Liu has served on the World Economic Forum’s panel on Industry Response to Government Procurement Policy, is leading the Autonomous Machine Computing roadmap under IEEE International Roadmap of Devices and Systems (IRDS), and is a member of the ACM U.S. Technology Policy Committee. Dr. Liu’s holds an M.S. in Biomedical Engineering, a Ph.D. in Computer Engineering from U.C. Irvine, and a Master of Public Administration (MC/MPA) from the Harvard Kennedy School. He is an Advisory Council member of the Harvard Business Review, a member of MIT Technology Review’s Global Insights Panel, and a member of the Forbes Technology Council. Dr. Liu has published 4 textbooks, more than 100 research papers, and holds more than 150 patents in autonomous systems.
Photo credit: Bowen Chin via Unsplash
 Wu, W. and Liu, S., 2021. Dilemma of the Artificial Intelligence Regulatory Landscape. Communications of the ACM, 2023.
 OECD. Publishing, 2014. OECD Regulatory Compliance Cost Assessment Guidance. OECD Publishing.
 PricewaterhouseCoppers, 2022. Illustrative IFRS consolidated financial statements.
 Renda, A., Arroyo, J., Fanni, R., Laurer, M., Sipiczki, A., Yeung, T., Maridis, G., Fernandes, M., Endrodi, G. and Milio, S., 2021. Study to support an impact assessment of regulatory requirements for artificial intelligence in Europe. European Commission: Brussels, Belgium.
 Fukuoka City conducts demonstration test of compact self-driving car by US company PerceptIn, Inc.. Nikkei, accessed 2023-01-05, https://www.nikkei.com/article/DGXLRSP518592_W9A900C1000000/
 List of Proposal Sectors and Private Companies, etc. . In Seeds proposal for realization of smart island, Japanese Ministry of Land, Infrastructure, Transport and Tourism, accessed 2023-01-05, https://www.mlit.go.jp/kokudoseisaku/chirit/kokudoseisaku_chirit_tk_000309.html