AI can make customer service better - but we mustn't let it make things worse
If you read about customer service on tech websites and on general news websites, you would be forgiven for thinking they exist in two different universes.
Tech websites (including this very newsletter 😊) are full of utopian visions of how Generative AI can deliver amazing customer service at lower cost. Yet plenty of newspaper column inches are taken up with with polemics about how customer service has never been worse, and most of the blame is being pinned on chatbots. Here are just some recent examples:
How did customer service get so bad? (Financial Times)
Chatbot ‘cycle of doom’ making customers miserable, companies warned (Telegraph)
Customer service is surely designed to irritate (i Newspaper)
Now, whilst there is some element of clickbait in these headlines, it’s hard to deny the truth at the heart of them. Customer experience is a lot about perception, and most customers’ perceptions have been formed from bad experiences of interacting with chatbots that don’t exactly live up to their initial promise.
As companies look to roll out the next generation of AI technologies into customer services, there are some lessons from the past that can help ensure the AI doesn’t inadvertently make the outcome worse.
1. AI automates a process that doesn’t work
One of the biggest challenges faced by customer service organisations is that sometimes there isn’t a standard process to fix every issue. There are always exceptional scenarios for which the standard process does not fix the issue - and every Customer Service Representative (CSR) knows which knowledgebase articles to ignore because they don’t provide an effective answer.
Processes that are automated based on what is written on paper rather than how things actually work in practice will fail quickly when they are set loose on real customer issues.
Potential mitigation: work closely with CSRs to build effective processes and don’t rely on what is written in the knowledgebase.
2. Customers get stuck in an AI loop
The situation described above is exacerbated when companies get so confident in the AI that they don’t give the customer any option to get human support, forcing them into an endless loop of frustration. Customers in this situation can choose from three courses of action:
Accept that their issue is never going to get resolved, alongside a permanent loss of trust in the brand
Find any means necessary to get their issue fixed, even if it means writing to the CEO or going to the press
Stop being a customer of the brand
None of which is a desirable outcome for most businesses.
Potential mitigation: always design “off-ramps” into self-service processes. This doesn’t mean that a customer should immediately be able to request human support from the outset, but the option should be made available if a customer has been through a loop more than once.
3. AI gets built, but not maintained
Customer needs change, products change, regulations change, but customer service technology is still too often delivered as one-off projects. Without sufficient resource allocation to maintain the technology, it becomes less effective over time. There is a greater risk with Generative AI tools because the capabilities of large language models also change over time, requiring continuous monitoring of performance.
Potential mitigation: design a future customer service function as a “change” organisation as much as a “run” operation, with the capacity and authority to continually maintain and improve tools.
4. Reduced business case to fix root causes
In this scenario, the AI works perfectly… from the company’s perspective. There is a high level of automation and agents are substantially more productive. The cost to serve customers becomes so low that there is no longer an obvious financial business case to fix the root causes of customer issues and prevent the need for customers to get help in the first place.
Potential mitigation: gain a clear understanding of the effect of metrics like Customer Effort Score or Net Promoter Score on long-term customer value and model this as part of the business case for change.
Part of the reason that interest in Generative AI has grown so quickly is that it is remarkably easy to get started. Pretrained Large Language Models enable companies to stand up proof-of-concept chatbots within a few days. The effort required to ensure that the AI works at scale - and continues to work at scale - is considerably greater.
Recommended news articles
Latest perspectives from BCG
Reimagining the customer service experience with Gen AI
DEEP AI by BCG X brings transformative relief to customer service headaches—for customers and businesses alike