Unique Research · Partner welcome · September 3, 2024
We are pleased to welcome Lanma Technology (澜码科技) as a Unique Partner.
About Lanma Technology
Lanma Technology was founded in February 2023 as an enterprise AI Agent platform company built on large language models. Core team members come from Google, IBM, Tencent, ByteDance, Alibaba, Yitu, and other leading internet and AI companies in China and abroad.
Main products
Lanma has shipped applications across finance, energy, government, retail, education, and more, and has formed strategic partnerships with multiple listed companies and unicorns.
Industry recognition
Lanma leads or co-authors several large-model standards hosted by the China Academy of Information and Communications Technology (CAICT), including as lead unit for the AI Agent standard and co-author for financial AI and large-model appliance standards. It is a member of the China Artificial Intelligence Industry Development Alliance, a finance-group member of that alliance, and among the first members of CAICT’s MaaS working group. In 2024 it was one of nine AI key projects signed by Shanghai Municipality — and the only AI Agent company among them.
Partial awards (incomplete count): since founding, Lanma has received about 17 media and think-tank list placements, 7 government and research-institute awards, 7 product and solution awards, and has co-authored 3 industry standards.
AskXBOT platform
AskXBOT is Lanma’s self-developed enterprise AI Agent platform on large language models. It combines Agent and workflow design, development, use, management, and knowledge accumulation in one system. Enterprise users can design, create, and manage Agents, and quickly customize enterprise AI Agents for many tasks — raising work quality while cutting cost.
Core capabilities:
- Agent design — preset Agent and workflow templates; create personalized, complex-task apps from templates or from scratch.
- Agent management — view and manage all in-enterprise Agents and workflows: status, permissions, usage.
- Agent use — talk to Agents anytime; combine UI to inspect process and results more intuitively. Built-in ready-to-use Agents and workflows in a public marketplace; creators can publish custom Agents/workflows to a team marketplace, with isolation between creator-private data and tenant-shared data. Agents assist knowledge governance alongside digitized expert knowledge; expert knowledge then powers Agents and workflows in a closed loop.
- Simulation platform — multi-dimensional performance tests across accuracy, recall, inference speed, resource use, and more.
Product advantages:
- Fast app building from preset Agent, dialogue-flow, and workflow templates, plus low-code canvas and components.
- Highly modular, atomic functions that compose and encapsulate for diverse scenes.
- Flexible model choice across domestic and international LLMs.
- Full, precise model evaluation to pick the best fit for a scene.
- High usability — out of the box, clear UI, simplified flows.
- Strong security and transparency — full statistical and audit logs.
- High traceability — clear data provenance and real-time records of user–Agent interaction.
- Flexible deployment — API and private deployment.
Distinctive edges versus peers
1. Expert knowledge
Lanma holds that expert knowledge is the key to getting AI Agents into the enterprise. The height of expert knowledge sets the height of value an Agent can deliver; data helps experts iterate that knowledge faster and raises Agent generality.
LLMs lack enterprise domain knowledge and struggle with real business problems — like a top graduate without practice. They also blur their own capability boundaries and still try to answer beyond them, producing “hallucinations.” In enterprise scenes, expert knowledge is therefore decisive; it sets the ceiling for AI Agents.
AskXBOT can accumulate expert knowledge and industry experience into an enterprise knowledge base for sharing and succession. Agents assist knowledge governance; expert knowledge then powers Agents and workflows — a virtuous loop.
2. Model neutrality and atomic capabilities
When LLM vendors ship models they often define parameters only, not a concrete FeatureList. Enterprises then face a “blind men and the elephant” problem: hard to judge fit and efficiency in a given scene, and hard to choose, tune, and optimize cost-effectively without trial and error.
As a model-neutral vendor, Lanma has, over more than a year of practice, accumulated performance data on Agent atomic capabilities across real scenes. That makes it clearer which models work where, so teams can choose, adjust, optimize, and combine models against core enterprise needs and land them in different business contexts.
Reference customer cases
Case 1: Bank loan due-diligence assistant
Background. A domestic bank launched inclusive lending for SMEs in line with national policy. In practice it is hard to fully understand borrower information and repayment capacity — especially for SMEs and sole proprietors with incomplete financial records. Relationship managers spend heavy time collecting materials, running due diligence, and writing reports covering client analysis, financial analysis, and review analysis.
Two pain points: (1) uneven analytical skill among front-line RMs → uneven DD and report quality; the bank wanted a system to assist analysis and catch easy-to-miss points; (2) RMs spend large amounts of time writing reports, while existing system output is rigid and not directly usable. The bank wanted an LLM-based AI Agent to pull credit-related data on applicants and auto-generate analysis reports.
Customer value. Lanma built a due-diligence report Agent that auto-produces conclusions and reports, cutting about 80% of RM report-writing time while reviewing materials more thoroughly and lowering error rates. Templates can be configured for different banks and branches.
Case 2: Insurance brokerage marketing-assist Agent
Background. A national, full-service insurance broker under a state-owned group, with dozens of branches, serves clients on brokerage and risk management. Its impaired / non-standard-risk products have broad demand but face pricing and underwriting difficulty; traditional sales paths hit bottlenecks:
- High acquisition cost and weak coverage of demand. Brokers “hunt for leads”: of 100 people spoken to, about 15–18% are potential targets; at a further ~20% conversion, ~3 closes per 100 conversations — low productivity.
- Scale and expertise pressure: business expansion raises headcount need; unlike product-only insurers, this broker can place almost any product type and tailor combinations — so matching by premium tier, region, and other factors is the hard problem.
Together with Lanma, the firm launched a marketing-assist Agent: auto-extract health and disease information from medical check-up forms; experts enter product-matching rules in natural language; the Agent recommends suitable coverage and helps brokers expand sales on existing clients — breaking the “hunt for leads” pattern. Multi-channel lead intake on the front end; a conversational “expert twin” turns professional content into language clients and agents understand, raising service level and conversion.
Results. As an AI-native support layer for the firm’s products, the Agent shortens pre-underwriting from about 3 weeks to 3–5 business days, expands lead sources, matches and dispatches by expert rules, and on the ops side automates expert-knowledge construction so scarce experts spend more time composing products for client scenes — feeding the Agent loop.
Case 3: State Grid unit — intelligent settlement audit Agent
Background. Central SOEs and public institutions still lean on manual review for supplier payment management — slow, error-prone, weak on large-scale data. For one State Grid unit, materials centralized-procurement settlement covers advance payment, goods-receipt payment, commissioning payment, and quality retention. Annual settlement volume exceeds 100,000 transactions, with many management steps, tight SLAs, highly structured multi-source data, and complex cross-specialty reconciliations.
The intelligent settlement audit Agent builds a dedicated audit model, senses data across fill, approve, analyze, and insight stages of the settlement system from many sources, documents, apps, and flows, and supports finer, smarter audit / expense-review management. Architecture spans data collection, processing, AI model, application service, and user interaction layers.
Core functions include: anomaly detection and risk scoring with LLMs; real-time feedback and batch export of audit results for settlement staff; auto audit reports with visual analytics; customizable audit rules via natural language.
Results. Intelligent audit covers advance, goods, freight, and retention payments across 100,000+ settlements — higher efficiency, lower settlement risk, visual alerts, actionable recommendations, and smarter multi-source reconciliation and root-cause insight.
About Unique Partners — see Unique Research partner program materials.