AI is Ready to Transform Customer Service — for Better & for Worse
Plus, a quick update on the AI arms race, and what comes next
February has been a busy month in the ongoing AI arms race. From the controversial, headline-grabbing releases of Google Gemini and OpenAI’s text-to-video tool Sora, to the brewing AI chips showdown between Groq and Nvidia, to Microsoft’s latest $16 million investment in French AI startup Mistral, there have been a whole lot of big AI development to keep up with.
For consumer-facing brands, however, the deployment of generative AI in supplementing customer service has garnered significant traction, thanks to two companies, that warrants a closer look.
AI Customer Reps Have Arrived
From the day that ChatGPT caught on, many have speculated its potential to aid human customer support teams to handle the basic customer service tasks, if not one day replacing the customer support staff. For big companies, customer service is one aspect of the brand experience that has already been largely outsourced to overseas workforces and increasingly automated. In other words, the conditions are ripe for an AI takeover.
According to a 2023 report by IBM, customer service is the primary area where executives are eager to implement generative AI — by the end of 2024, 85% of executives surveyed by IBM say they expect generative AI to be interacting directly with customers, underscoring the significant impact AI is anticipated to have in transforming customer service.
The integration of AI into customer service is not just a future prediction but a current reality, as illustrated by Klarna’s deployment of its AI assistant. The Swedish buy-now, pay-later company has been touting the effectiveness of its OpenAI-powered virtual assistant in managing customer interactions. Within the first month of its deployment, the AI chatbot was responsible for handling two-thirds of all customer service chats, accounting for about 2.3 million conversations, per Klarna’s press release.
Remarkably, the virtual assistant achieved customer satisfaction ratings on par with human agents, suggesting that AI can match, if not exceed, the service quality of human customer reps. In addition, Klarna anticipates profit improvements of $40 million in 2024 due to the efficiencies brought about by its AI assistant, highlighting the financial benefits of embracing this technology.
(One caveat here is that Klarna is set to go public later this year, and will need all the good press it can get to generate some pre-IPO buzz, so all numbers in this case should be taken with a grain of salt.)
Another recent success story is ServiceNow, thanks to an aggressive integration of AI features into core offerings. The Workflow-owned cloud-based IT management platform, recently reported stellar financial results, with subscription revenue exceeding 27% year-over-year growth, largely attributed to their AI initiatives. Moreover, ServiceNow recently announced an extended partnership with Nvidia to develop a suite of generative AI tools specifically designed for the telecom industry. It is worth noting, though, that ServiceNow is an enterprise tool and customer service works a bit differently in the B2B domain than in B2C.
Needless to say, these two success stories also highlight challenges faced by traditional customer service firms. For example, French call-center firm Teleperformance recently experienced a significant slump in its shares, driven by concerns over the impact of AI on its existing business model. This reaction underscores the broader apprehension within the industry about the disruptive potential of AI in customer service.
The advantages of AI in customer service are compelling, particularly in terms of cost efficiency and the ability to manage a high volume of interactions with consistent quality. For businesses, the appeal of AI lies in its scalability and the opportunity to reduce operational costs significantly. The technology’s capacity to handle routine inquiries and processes allows human agents to focus on more complex and nuanced customer needs, potentially elevating the overall customer service experience.
However, the rapid adoption of AI in customer service is not without its downsides. The loss of the human touch in customer service, which AI won’t be able to fully replicate yet, can be particularly detrimental for brands that rely on personalized, high-touch experiences as a part of their value proposition. Luxury brands, for instance, often rely on the white-glove approach of their customer service representatives to cultivate loyalty and uphold exclusivity. Replacing this with an AI customer rep could significantly undermine the perceived luxury-ness and damage customer relationships.
Moreover, the potential for generative AI to “hallucinate”– or generate inaccurate, nonsensical responses–poses a significant risk. Imagine if an auto brand’s AI customer rep misunderstood a customer’s inquiry and ended up providing incorrect care instructions, potentially causing safety issues for the customers — The resulting customer frustration and potential harm to the brand image could have major ramifications.
Therefore, brands need to take a cautious approach when deploying AI in customer service. While it’s tempting to embrace AI for its potential efficiency gains and cost reductions, businesses must carefully weigh those benefits against the risks to their brand value. A balanced strategy that leverages AI selectively for routine inquiries, while preserving human expertise for complex, sensitive issues, offers a more sustainable path to maximizing the strengths of both human and AI agents in customer service.
An “Everyone vs. Google” Battleground
The competitive landscape of generative AI is rapidly solidifying into an “Everyone vs. Google” scenario, as all the non-Google AI stakeholders increasingly collaborating with each other to undermine the Alphabet company’s perceived lead in AI research, underscored by recent developments and strategic moves by major technology companies.
The leading AI news this week is Google’s recent stumble with the release of its Gemini chatbot highlights the challenges even leading tech giants face in the race to dominate the AI market. Gemini’s problematic rollout, featuring inappropriate image generation and controversial chatbot responses, has created quite a PR disaster for Google. For example, Gemini’s since-removed image generator put people of color in Nazi-era uniforms, seemingly a result of the company’s over-correction of AI image generators’ widely known bias against people of color. There is no doubt that Google has rushed the product, as confirmed by reports of Google’s own employees mocking the PR disaster while admitting that the rollout of Gemini’s image-generation tools had been rushed. Google says it plans to relaunch Gemini’s text-to-image function “in a few weeks.”
Meanwhile, Google’s competitors are closing in. Microsoft’s $16 million investment in Mistral, a French AI startup with a foundational AI model that reportedly rivals GPT-4’s performance, grabbed quite a few headlines. Statistically, this investment is almost negligible, as it represents less than 1% share of Mistral, which is now valued at over $2 billion. Strategically, however, this investment represents a significant move by Microsoft to diversify its AI portfolio (to potentially guard against the type of CEO-ousting drama that OpenAI’s non-profit board pulled last year), while also gaining an entry point in the European AI market.
Similarly, Amazon recently added support for Mistral’s foundational models on its Bedrock AI service platform, underscoring its strategic commitment to providing diverse solutions from mostly open-source AI models.
Speaking of open-sourced models — Meta’s next open-source language model, Llama 3, is set to be released in July and could be twice the size of the current Llama model, with more than 140 billion parameters. This will theoretically allow it to be more responsive to the user, providing context for difficult topics rather than blocking tricky questions.
Overall, the current trend indicates a tightening race in AI, and Google’s recent missteps have undoubtedly given its competitors a decent chance to catch up. While it is certainly too early to count Google out, the search giant does need to reassess its go-to-market strategy and potential company culture issues if it wishes to reclaim its leading position in the AI race.
The Ouroboros of AI
For a while, a key aspect of Google’s perceived lead in AI, especially in terms of LLMs, lies in the vast amounts of consumer data that Google sits on via its various services. However, recent developments in AI data licensing deals indicate a shift in the sourcing of AI training data.
A notable example of this shift is the deal between social media giant Reddit and Google, worth reportedly $60 million annually. This agreement enables Google to use Reddit’s content for training its AI models, highlighting the value of social media data in refining AI capabilities.
Similarly, Automattic, the parent company of Tumblr and WordPress, is finalizing deals with AI companies OpenAI and Midjourney to sell user data for AI training purposes. Automattic’s commitment to sharing only public content from non-opting-out sites underscores the delicate balance between leveraging user data for AI advancements and respecting privacy concerns.
These developments are indicative of a broader shift towards collaborative and transactional approaches to data sourcing for AI. As a result, AI data training deals have become a lucrative opportunity for media companies treading water in today’s online publishing hellscape. For example, OpenAI rolled out a partnership program last year to collect datasets from third parties to help train its AI models. The intensifying competitions among tech giants reflect the increasing importance of diverse, real-world data in developing more sophisticated and effective AI models.
Google, on the other hand, has been hit with a 2.1-billion-euro ($2.3 billion) lawsuit by 32 mostly European media groups including Axel Springer and Schibsted, alleging that they had suffered losses due to the company’s practices in digital advertising.
To Google’s credit, the company has been paying independent publishers to test its unreleased generative AI models. The program, first reported by Adweek, allows a handful of publishers access to its latest AI platform in exchange for receiving analytics and feedback. A Google representative told Adweek they’re in the early stages of providing small publishers with AI tools “to help journalists with their work.”
The burgeoning open-source AI movement, fueled by Big Tech’s contributions, is driving innovation at an unprecedented pace. However, as MIT Technology Review suggests, this golden era of openness and collaboration faces potential threats from increasing competition and regulatory scrutiny. There is a distinct possibility that the leading AI stakeholders like Meta and Microsoft might retract their open-source contributions in response to these mounting external pressures to keep its AI models under lock and key.
Yet, given that an increasing amount of online content is becoming AI-generated, the potential prospect of AI being trained on licensed social and publisher content may result in a paradoxical ouroboros-like scenario, where AI models are trained on AI-generated content and regurgitate itself in a closed loop. This obviously poses significant challenges to the development of AI, and would require a concerted effort to ensure that AI models are trained on diverse, authentic, and human-generated information, so as to preserve the integrity and richness of the information ecosystem.
In conclusion, the AI competitive landscape is rapidly evolving into a complex ecosystem of strategic partnerships and legal challenges. The “Everyone vs. Google” narrative not only captures the current competitive dynamics but also highlights the broader implications for the future of AI development and its impact on society.
