Autonomous Customer-Query Resolution For Product Recalls

By S Ramachandran, Lakshminarasimhan N March 2020   |   Article   |   10 min read   |   Email this article   |   Download
Manufacturers face demand spikes in customer queries during challenging, unplanned product recalls. Artificial intelligence can autonomously manage elastic demand from customer queries, offering quick setup and resolution. It is cost-effective, it can improve customer loyalty and act as a tool to cross-sell and upsell.

A product recall is action to remove from sale, distribution, or consumption products that may pose a safety hazard. Recalls can occur because of a report from regulators, manufacturers, wholesalers, retailers or customers. Product recalls are growing in volume and complexity.

Product recall campaigns in the U.S. automotive sector, for example, have seen an upward trend in the past decade. Data from the National Highway Traffic Safety Administration shows a peak in 2018, with more than 1,000 recalls affecting close to 35 million vehicles.

Similarly, the consumer goods sector is witnessing an increasing number of recalls, according to the Organisation for Economic Co-operation and Development.1 The 2019 OECD’s global awareness campaign on product recalls reported 3,700 recalls in 2018 from 39 countries.

To prove the effectiveness of recalls, ideally each individual product recall case must be closed with customer sign-off and feedback, documentation of the root cause, the resolution, and corrective action, if any.

According to the NHTSA, however, vehicles that were six to 10 years old in 2018 saw an average recall completion rate as low as 56% — putting the health and safety of drivers at risk.2

As a result, the NHTSA made four suggestions to help achieve 100% completion of product recalls: improve the reliability of data on ownership; improve communication methods to reach customers; incentivize customers, dealers and suppliers to engage in recalls; and engage with suppliers better.

Clearly, a big part of this is communication with customers. But communication can also be the most difficult to manage. Once a recall is announced, a spike in inbound queries needs to be managed. Data from customers has to be recorded thoroughly, securely and accurately, and in a sensitive manner that helps improve rather than aggravate an already-impacted customer experience of the brand.

Given the unpredictable nature of recalls, it is neither economically attractive nor feasible to plan for staff members to handle product-recall queries. Likewise, outsourcing the problem is not simple. It takes time to onboard partner teams with secure access to enterprise systems, and knowledge and systems access will be specific and distinct for each query.

AI could be a solution, however, because it can help automate many elements of the process and can be agile and flexible enough to fit to the needs of each specific recall. But modern AI goes beyond automation. It can use sentiment analysis and neural networks to better react and respond to customer interactions. Overall, it can reduce the amount of human intervention required, and speed up the entire process.

AI for speed and sensitivity

AI systems can be set up in a matter of days, with product manuals, specifications and test data being used to train the system. By using neural networks to sense human emotion, AI recognizes patterns in sequences of images or voice, observing changes in speech tone, loudness, tempo and voice quality. Similarly, sentiment analysis analyzes textual data, such as emails or chat, for emotional information.

A plug-and-play approach can be seamlessly adopted across multiple channels of communication, such as emails, phone calls and chatbots (Figure 1). Robotic process automation can be used to mimic human behavior for access to other enterprise systems for any data need, avoiding back-end integrations.

Figure 1. AI can sensitively manage interactions with customers throughout the recall process

AI can sensitively manage interactions with customers throughout the recall process

Planning for an AI emotion-sensitive recall program

AI enables a comprehensive, omnichannel-based product recall program supported by digital tools (Figure 2). AI can decide the exact communication to be sent, matching customer preference with his or her previous history. AI can take care of response management, with individual interactions helping to understand the emotion of the affected customer.

AI can therefore play a key role in a large scale, emotion-based product recall strategy augmenting human staff members.

Figure 2. AI can be deployed specifically for response management and can feed into the resolution process

AI can be deployed specifically for response management and can feed into the resolution process

There are five areas where AI can be applied to the recall process to help improve customer interactions. The tools can be used either individually or as an integrated platform, enabling seamless switching with no loss of context. However, each tool on its own can provide significant business benefit.

Phone calls for voice-based reach

Voice-based phone calls are the most common medium for reaching out to customers when there is a problem. Recalls are no different — but AI reduces the need for the human element, with voice interfaces having components such as automated speech recognition, text-to-speech converters and natural language generation.

AI can help devise large scale, emotion-based product recall strategy without human assistance

Infosys is currently working with a global telecommunications and mass media company to pilot voice-to-text conversion, where the system will automatically classify the type of call based on historical conversations. For recalls, historical phone calls could be reviewed and classified by the issue described by the customer, the root cause of the problem or the resolution. This will help design the AI system and how it links to the overall recall process for future calls. Going forward, enhancements are possible via call analytics, delivering sentiment and call-duration analyses.

Email workbench for customer query resolution

Autonomous handling of email-based customer queries begins with the classification of incoming emails into categories based on the underlying issue. Multiple technologies, such as natural language processing can be used (Figure 3).

Figure 3. A typical process for email resolution using AI

A typical process for email resolution using AI

Missing data can be queried from enterprise systems such as an ERP or CRM, with an autonomous email response then formulated based on the available data. A confidence level can be provided for each email response and, based on subsequent thresholds, the system can decide whether a manual review is needed before sending the response to the customer.

Infosys implemented an AI-based email workbench to handle over 50,000 emails a month for a client, with most having scanned documents as attachments. The business need was to extract key information from the unstructured emails and documents, classify them and map each email to one of 50 types of service requests to be raised.

The implementation streamlined the case-handling process, delivering a 30% reduction in the need for human customer service agents for product recalls.

Chatbots for real-time response

Gartner research has shown up to a 70% reduction in customer inquiries due to the deployment of virtual customer assistants or chatbots. As a result, Gartner predicts 25% of customer support operations will engage chatbots in 2020.3

At a broad level, bots can be classified into two types:

  • Knowledge bots that respond to a customer query.
  • Action bots that need to perform work, such as filling in a form.

Industry-specific bots with reusable components and accelerators speed up the recall setup process and provide deep domain knowledge.

Chatbots can be configured by building a knowledge tree from documents such as product manuals, frequently asked questions and specifications. Bots build a decision tree to answer queries, while users can configure business logic or conversation flow via a bot studio tool. With very little requirement for coding, bots deliver a seamless end user experience by integrating with existing front-end channels, such as social media, mobile phones or the web.

An order management and invoicing solution Infosys implemented for an electronic manufacturing firm covered 29 use cases. Order-related information had to be extracted from the client ERP system by performing a search for specific attributes.

With AI providing customer support during unplanned product recalls, autonomous customer-query resolution can be dedicated to meaningful customer engagement and quick product recall resolutions

The Infosys chatbot implementation was delivered in two months, reduced ticket volume by 15% and is now used daily by more than 260 users.

Document management

In product recalls, organizations receive large volumes of unstructured documents in different formats from customers, partners and employees. These documents are usually handled manually by operations agents. They are received as scanned images or in other formats, reviewed and validated, with the data stored in databases or transaction management systems.

While organizations consider AI and RPA for intelligent automation of such processes, it remains difficult to integrate these technologies with legacy systems. As a result, organizations fail to extract the underlying value from such implementations. This creates a business need to develop a unified user interface bringing together enterprise data, insight and action.

Infosys implemented an automated document extraction initiative for an automated “know your customer” initiative. Each customer profile was created from unstructured documents, the process starting with the extraction of key attributes from scanned documents. A portal was provided for subject matter experts to view, edit and approve extracted attributes.

In terms of impact, automation shrank the average case time by 83%, from 60 minutes to 10 minutes.

A cognitive workbench to integrate all tools

Research published by automotive industry bodies shows a consistent lack of customer interest in recall case closure.4 The primary reasons are lack of time, not knowing where to take the vehicle for repair and confusion about who pays.

A cognitive workbench, however, is a platform bringing together digital tools in AI to provide a seamless customer query resolution experience. This is delivered via a single platform switching between channels, with no loss of context or continuity (Figure 4).

AI and RPA can effectively manage recall processes and promote cost-cutting while improving customer experience

As a result, not only does a cognitive workbench improve the accuracy and relevance of information shared with individual customers, it can also be a tool to cross-sell and upsell, if the engagement is handled well.

Figure 4. A cognitive workbench for AI based customer query resolution

A cognitive workbench for AI based customer query resolution

Richer and deeper, not just faster

Product makers have seen AI and RPA as a source of cost savings by being an effective alternative to traditional customer service agents. Indeed, according to IDC, automated customer service agents were the largest industry use case for AI investments in 2019.5

While AI can attenuate spikes in customer queries during unplanned product recalls, autonomous customer-query resolution can also build richer, deeper customer engagement via timely resolution of product recalls.

As a result, by driving effective management of the recall process, AI and RPA not only save on cost but also provide an opportunity to improve customer engagement and loyalty.