← Sticker Wall AGENT · DETAIL

Mulligan AI Financial Compliance Review Agent Generating HK$80k Monthly

Workflow: Automatically connect to customer communication records of wealth management and insurance institutions every day, inclu

AGENT

Key Fields

FIELD STAMPS

IndustryFintech

RegionChina（中国香港）

ScaleSME

ChannelOnline

🔧 Workflow

Automatically connect to customer communication records of wealth management and insurance institutions every day, including voice-transcribed text, online chat logs, and emails. Use large models to compare regulatory rule libraries and internal compliance manuals sentence by sentence, outputting compliance risk tags, non-compliant phrasing summaries, and rectification recommendation reports. Human compliance officers only spot-check high-risk items, and after confirmation, generate daily reports and monthly compliance dashboards. The system batch-processes all communication records from the previous day every day at midnight, pushes high-risk cases to the compliance officers' email before 8 AM, completes human review before noon and triggers rectification work orders, and automatically generates monthly compliance evidence packages that can be submitted to regulators by the afternoon.

🛠 Setup Requirements

Requires familiarity with financial regulatory rules and insurance sales compliance requirements, as well as proficiency in using large model APIs and rule engine tools. Technically, open-source models can be selected to build a local compliance knowledge base, or cloud-based large models combined with rule engines can be used for hybrid judgment. Building the core prototype takes about two to three weeks, and full integration into institutional systems takes a month. Initially, Python scripts can be used to call large model API to process text records, and then gradually connect to the voice transcription pipeline. A traceable audit log module needs to be prepared to ensure that every AI judgment can be traced back to specific regulatory clauses, which is a prerequisite for gaining the trust of institutions.

🧰 Toolchain

🔧 Large Model API
🔧 Rule Engine
🔧 Speech-to-Text Tool
🔧 Compliance Knowledge Base
🔧 Audit Log Database

💰 Revenue

① Insurance brokerage institutions subscribe by seat (main revenue): institutions pay monthly based on the number of connected seats, at HK$1,600 per seat. 50 seats equal HK$80,000 per month, which is the sole income item of the current month, accounting for about 100% (derived from figures in the case, case caliber, not externally verified), and the proportion of this channel in revenue is not separately disclosed; ② Elastic billing based on the number of reviewed items: institutions pay based on the number of reviewed records, at HK$200 per thousand items. The number of items reviewed each month is not specified, and the volume of this revenue cannot be verified either (case caliber, no external review seen), and its proportion is also missing; ③ Value-added items such as compliance evidence packages and rectification work orders: institutions pay service fees per project, but quotes, project quantities, and revenue proportions do not have data yet; ④ Opportunity item - Cross-border bilingual review: about 30% of 377 financial enterprises in Taiwan have adopted AI, with the life insurance industry adoption rate reaching 62% (media estimate, without independent review), and real-time quality inspection accuracy exceeding 90% (caliber given by merchants, without independent review). Bilingual reviews can serve as a differentiated entry point, with no figures for the proportion yet.

💸 Cost

Mainly large model API calling fees and cloud storage costs, about HK$3,000 per month. Voice transcription is billed by the minute, processing about 100,000 minutes of calls per month, with transcription costs of about HK$1,500. Vector database rental and backup storage for the compliance knowledge base cost about HK$800, and the rest are rule engine server and monitoring/alerting fees.

⏱ Time Investment

2 to 3 hours per day maintaining the rule base and reviewing high-risk cases. An extra half-day spent every week having meetings with institutional compliance officers to align on new regulatory guidelines and update thresholds and keywords in the rule engine. One day spent every month doing omission retrospective analysis, adding model-missed cases found in manual spot checks into the training set.

🚀 Getting Started

Start by cutting into a small and medium-sized insurance brokerage company, providing a two-week compliance review pilot for free, and accumulating hit cases and omission comparison data. Train the rule base with real non-compliant phrasing and regulatory tickets, and then sign contracts with a billing model based on the number of reviews. The first step is to organize regulatory inquiry letters and internal compliance circulars received by target institutions over the past two years, extract high-frequency non-compliant phrasing and penalty clauses from them, establish an initial rule base, and then bring this rule base to conduct pilot demonstrations.

🔑 Keys to Success

✅ Continuous update of the financial regulatory rule base, proactively integrating the latest regulatory guidelines every month
✅ High-risk case human arbitration closed-loop, with human-confirmed results flowing back to train the model
✅ Integration with the institution's existing quality inspection work order system, leaving no data silos
✅ Audit logs traceable to specific regulatory clauses to meet regulatory evidence chain requirements
✅ Bilingual review capabilities covering Hong Kong's mixed Chinese-English communication records

⚠️ 风险

⚠️ Frequent changes in regulatory rules, which may cause the model to lag and lead to omissions
⚠️ Financial institutions are sensitive to data crossing borders, resulting in high barriers for local deployment
⚠️ High misjudgment rate of large models for Cantonese and financial jargon
⚠️ Internal compliance officers of institutions may resist automated systems out of fear of being replaced

📌 Real Cases

📌 Guotai Junan International's launched Archer intelligent advisory system has been equipped with an AI compliance verification mechanism for pre-review of investment advice compliance, proving that top-tier brokerages are willing to pay for AI compliance capabilities
📌 Zhongguancun Science and Technology Leasing's Dezhu Intelligent insurance smart quality inspection solution has been implemented in multiple insurance companies to build a compliance firewall for insurance sales, validating the willingness to pay for insurance sales phrasing reviews
📌 Fubon Property & Casualty Insurance launched the AI claims smart assistant in 2026, introducing AI Agents into the core insurance operation process, indicating that insurance institutions' acceptance of AI automated review is rapidly increasing

Sources