Turning Large Language Models into Business Results: Alibaba Cloud’s Playbook

Observations and Reflections

The rapid rise of AI large models offers unprecedented opportunities, but true enterprise impact requires overcoming many pitfalls. Jiang Linquan, CIO of Alibaba Cloud Intelligent Group, reflects on three years of leading digital and intelligent transformation, emphasizing the need to align production relationships, culture, and expectations between business and IT.

Full Landscape of Alibaba Cloud Large‑Model Applications

Alibaba Cloud has deployed roughly 28 "digital‑human" projects across documentation, translation, customer service, sales, contract review, BI, employee services, and R&D. These agents are integrated into official websites, CRM, support systems, and internal HR platforms, delivering measurable business outcomes.

Case 1: Translation

Translating technical documentation for over 300 products and millions of pages was a bottleneck. Traditional NLP failed, and hiring bilingual technical translators was infeasible. After trying ChatGPT‑3.5 and 4, the team achieved translation quality comparable to professional translators (4.6/5 vs 4.12/5) at 1/200 of the cost, completing full Indonesian translation and dramatically improving NPS.

Case 2: Intelligent Outbound (Smart Calling)

With millions of enterprise customers, staffing thousands of sales and support agents was impossible. By leveraging existing voice and multimodal expertise, Alibaba Cloud built an AI‑driven outbound system whose service capacity equates to hundreds of human seats, dramatically reducing recruitment and turnover challenges.

Case 3: Contract Risk Review

Large B2B contracts require multi‑discipline risk checks that previously took weeks. The team created "digital humans" for finance, credit, and legal review, embedding them at the contract drafting stage to flag risks in real time, cutting review cycles from months to minutes and freeing experts for higher‑value work.

Case 4: Employee Service Digital Human

HR processes (leave, health checks, benefits) are low‑frequency and scattered across systems. By consolidating them into a single AI assistant on DingTalk, the solution saved the equivalent of ten full‑time staff and dramatically improved employee experience with natural‑language commands.

End‑to‑End Pitfalls and Solutions – RIDE

The team distilled two years of experience into a four‑step framework called RIDE: Reorganize, Identify, Define, Execute. Following these steps raises the probability of successful AI adoption.

Reorganize

Traditional production relationships cannot support AI‑driven productivity. Organizations must restructure teams, culture, and incentives to align with AI capabilities, turning the AI era into a new "elevator" that amplifies effort.

Identify

Pinpoint business problems that are language‑centric, repetitive, scalable, and talent‑scarce. Typical targets include translation, voice‑to‑text, SQL generation, code generation, and any task where AI can replace a high‑cost human effort.

Define

Set clear product metrics (accuracy, latency, safety) and operational KPIs such as DAU, query volume, penetration rate, and especially retention. Accurate measurement prevents teams from chasing vanity metrics that do not reflect real business value.

Execute

Drive engineering and data work with product and business goals as the north star. Build robust data pipelines, APIs, and evaluation frameworks; iterate on models only when data quality and evaluation are ready. Training is reserved for cases where base models cannot meet performance or latency requirements.

Key insights include the necessity of a "knowledge‑engineered" intent space, rigorous E2E attribution to locate failures (most often in data or API layers), and the importance of treating AI as a result‑as‑a‑service (RaaS) offering rather than a mere tool.

Conclusion and Outlook

Alibaba Cloud positions itself as a large‑scale RaaS provider, enabling enterprises to ride the AI "elevator" built on a strong cloud foundation (MaaS, PAI, ODPS, databases, ECS, networking). By continuously lowering costs and expanding capabilities, the platform empowers organizations to achieve digital transformation at unprecedented speed.