AstraZeneca Cuts Compliance Labor Costs by Over 50% in Three Months with AI-Powered Intelligent Assistant

An intriguing question has been raised: Can enterprises truly use artificial intelligence to replace traditional information consultants? Today, many modern companies are eager to find the answer. The healthcare industry is no exception.

It is hard to imagine that AstraZeneca, an international pharmaceutical giant, previously relied on a team of just over ten people to handle compliance consulting. On a daily basis, they needed to manage more than 8,000 online inquiries from medical representatives, addressing a high volume of repetitive issues related to expense reimbursements, meetings, and client reception. Notably, 80% of medical representatives asked the same 20% of questions. Following these consultations, medical representatives often requested follow-up responses from compliance colleagues via online channels to obtain documented records.

At the end of each month or during promotional campaigns, the volume of online inquiries received by the department peaks. During these periods, more than ten compliance staff members are often so busy that they barely have time to eat.

For enterprises, assigning dedicated personnel to respond to queries around the clock is undoubtedly a waste of talent. The busyness and mechanical nature of such foundational work erode employees’ morale. As staff turnover rates continue to rise, recruiting suitable candidates has become a significant challenge for human resources departments.

Therefore, how to leverage new technologies to enhance corporate efficiency and replace existing service models is an urgent issue for AstraZeneca. The maturation of artificial intelligence technology has made intelligent information consulting possible. For AstraZeneca’s compliance personnel, this may herald a strategically significant transformation...

The medical field is highly specialized, and for a strictly regulated department like AstraZeneca’s Compliance Department, the appropriate use of professional terminology is critical, as it directly impacts the accuracy of response outcomes. Therefore, ensuring accurate language input and maintaining sufficiently professional phrasing in AI products represent the most critical challenges that the AstraZeneca Compliance Bot must address.

A mysterious AI startup accepted the offer extended by AstraZeneca.

In less than a month, the team obtained from AstraZeneca 70 corpora databases of frequently asked questions and three standard business process workflows. After studying and understanding relevant pharmaceutical laws and regulations, the team quickly organized the materials and leveraged AI technology to initially establish a knowledge base containing more than 100 knowledge points.

Upon completion of the first phase, the entry-level AI chatbot was successfully developed.

Subsequently, the chatbot was launched internally at AstraZeneca, undergoing a “clinical trial phase” through interactions with relevant sales personnel. This process resembled placing a shriveled sponge under running water. Over the two-month testing period, new data and emerging questions served as the “nutrient boost” for the conversational AI.

Its core knowledge base continues to expand. When the product reaches “maturity,” the knowledge base will contain approximately 374 knowledge points and index 5,232 questions, covering 82.3% of inquiries raised by employees and medical representatives, with data continuing to rise.

This pre-built AI chatbot comprises three core modules: the Q&A module, the table module, and the task cluster module. These modules correspond respectively to three modes of human-to-human conversation and are centrally managed by an intelligent control console.

Specifically, the intelligent central console categorizes and identifies collected issues, analyzes their dialogue patterns, and dispatches tasks to the modules best suited to handle them.

“Robot Q&A Module” primarily handles simple question-and-answer dialogues, specifically responding to single-context, one-dimensional questions posed by users. For such inquiries, the robot provides precise matches and responses based on the knowledge base, proving effective in resolving common issues such as meeting time queries.

However, most conversations in the pharmaceutical field are not simple one-dimensional dialogues; they often involve product descriptions and comparisons. This means that AI and users must constrain the chat content within a specific scope (such as a table) to facilitate normal communication, which necessitates the use of a “Table Module” for processing.

For example, when inquiring about AstraZeneca’s products, medical representatives often need to obtain drug information within specified price ranges and compare different medications. To address such scenarios, AstraZeneca can pre-load relevant price reports (e.g., in Excel format) into the system, enabling medical representatives to define the scope of their inquiries and conduct multi-turn questioning focused on specific information.

Thus, when users interact with an AI chatbot, the bot can leverage relevant contextual information to accurately interpret the user’s query. It then performs data retrieval, sorting, and filtering based on structured tables, thereby organizing and outputting two-dimensional data to address tasks such as drug price comparisons and expiration date comparisons.

The third module, the “Task Cluster Module,” is designed to address specific issues. Some users may pose concrete questions, such as seeking consultation on project applications through the AI chatbot system. For such inquiries, the AI chatbot will guide users in entering the information required for the application, answer questions that arise during the process, and direct users through the relevant procedures.

Currently, the accuracy rate of this AI chatbot system in handling inquiries has exceeded 90%, with an intelligent recommendation trigger rate of 98%. More than ten former members of AstraZeneca’s Compliance Department now have the time and energy to engage in higher-value work.

This enigmatic AI team is named Laiye. According to its official website, the company was founded in 2015 by a team of PhDs and MBAs returning from Ivy League institutions, with a commitment to becoming a globally influential intelligent robotics company in the era of human-machine symbiosis.

Prior to engaging with AstraZeneca on its compliance bot project, Laiye’s client portfolio already included large-scale B2C enterprises such as Ctrip and Wyeth. However, through discussions with AstraZeneca, Laiye recognized that it differed significantly from those consumer-facing companies.

For To C projects, Laiye typically devotes more effort to consumer experience when designing AI chatbots, making conversations more engaging to boost customer conversion rates. However, as AstraZeneca is a To B enterprise, Laiye places greater emphasis on the professionalism and accuracy of its AI solutions.

From the results, the entire process of building and refining the AI product took only three months, yet yielded cost reductions exceeding 50%. For pharmaceutical and healthcare companies, leveraging chatbots to optimize workforce allocation, reduce costs, and achieve digital management is clearly a highly cost-effective strategy.

Not stopping there, Han Rui, Business Director at Laiye and project lead for the AstraZeneca collaboration, revealed, “In the course of our partnership with AstraZeneca, we have identified additional opportunities for cooperation. In some remote areas, pharmaceutical sales representatives often reduce their visit frequency due to long travel distances. Consequently, local physicians are unable to promptly contact these representatives to address questions or meet emerging needs, which inevitably leads to lost sales for pharmaceutical companies. In response, Laiye is planning to develop an AI-powered chatbot tailored for physicians, delivering professional technical support via mobile devices to doctors at hospitals covered by pharmaceutical enterprises.”

For instance, when physicians have medication-related inquiries, they can first consult an AI-powered robot. The AI bot not only answers basic questions but also notifies the pharmaceutical sales representative responsible for that territory, prompting timely follow-up visits. This model significantly boosts physician engagement and stickiness on the pharmaceutical company’s platform, thereby driving greater conversion of business opportunities.

More notably, Laiye has also pioneered a model that leverages small-scale knowledge bases to address practical problems and commercialize such solutions, thereby gradually shifting AI technology from industrial applications to consumer-facing ones, and progressively resolving the daily challenges faced by consumers and workers.

In response, Han Rui stated, “Currently, robots have become deeply entrenched in industries such as pharmaceuticals and maternal and infant care. The era of digitalized business management will truly arrive only when the AI knowledge base-driven industry becomes increasingly robust and a growing number of enterprises reap its benefits.”