Previous Submission

Hi, my name is C-3PO. How can I assist you today?

Next Submission

AI is driving the next labor revolution, starting with call centers

Automation reaches the cubicle

Just as automated machines revolutionized factory labor models over the last century, artificial intelligence (AI) has the potential to dramatically alter the labor models of services over the next hundred years. [1] Increasingly over the late 1900s, robots could more reliably execute repetitive (and/or dangerous) manufacturing tasks at lower cost than their human counterparts. With the development of greater compute power – thanks to Moore’s Law – alongside increasing consumer comfort with digital interactions and self-service, AI has emerged promising to automate human tasks far from the shop floor. The call center, with high volumes of similar inbound queries and rote requests, is one of the most promising near-term applications of this digital transformation. [2][3][4][5]

The typical call center today is a sea of cubes filled with agents on headsets sitting in front of screens, talking and typing at all hours of the day, most days a year. While online self-service has never been more accessible, call center volumes remain high, as rising customer expectations can often more than offset the volume decrease from self-service. [6] Companies like LiveOps, which leverages an online platform to flexibly staff a large network of at-home agents, have already begun to chip away at typical call center format, but the real threat to agents comes from AI. [7][8] Typical customer inquiries often concern account status (e.g., checking balances, noting address changes, requesting cancellations), technical help (e.g., “my wifi isn’t working”), purchase advice (e.g., “which of these phone plans best fits my needs”), and complaints and returns (e.g., “my delivery never arrived and I want a refund” or “your employee was rude to me”). [9] The vast majority of these inquiries take the form of predictable (if not always simple) questions with predictable answers, suggesting virtual assistants powered by AI could perform most of these tasks at a comparable or even superior level to human agents.

Virtual assistants in, human agents out

Given their superior value creation at lower cost, virtual assistants are likely to increasingly replace human agents in the call center. Using text and voice to communicate with humans, virtual assistants are powered by algorithms, often machine learning algorithms (including natural language processing), which can classify incoming queries and determine appropriate responses. By analyzing large amounts of data on past interactions as well as some pre-programmed associations and customer records, the virtual assistant can determine which past interactions the incoming request is most like, and respond with the associated response most likely to be relevant and successful for the customer and query at hand. For example, after examining a database of past telecom interactions, a virtual assistant could to suggest to an irate broadband customer a number of potential fixes for a non-functional router, adjusting each subsequent suggestion based on the caller’s input. While diagnosing technical fixes and advising on purchases may seem more complex than checking account balances, the call center has encountered most inquiries hundreds if not thousands of times, making even complex asks in fact repetitive and predictable. Virtual assistants, then, are well positioned to perform such tasks.

While it is still early to call a clear winner in the virtual assistant space, Amazon and Google provide compelling examples of what a winner in this space might look like given their aggressive push to-date into the home assistant space. [10] Indeed, both tech giants have developed call center virtual assistant products (Lex and Contact Center AI, respectively), and we can use their models as comparison points to today’s human agents. [11][12]

Virtual assistants have the potential to create superior value for customers and the companies they frequent across nearly every industry. Virtual assistants can perform tasks (like checking accounts or diagnosing technical issues) faster than human agents can, as the algorithm can process data more quickly than an agent thinking through a problem or typing a search query into a database, reducing the throughput time of each call and therefore taking up less of the customer’s time, improving customer service. Virtual assistants can also reduce customer wait times to essentially zero (besides connection and processing times), as the capacity of the virtual assistant to simultaneously handle inbound volume is limited only by server capacity rather than number of hired bodies currently available (and not on break) in the call center: this further reduces the total throughput time for each customer from the time s/he dials the support number or opens the chat box to issue resolution improving customer service. Gone as well is any need for the dreaded transfer ping pong between agents as the customer need only deal with one virtual assistant with instant access to all relevant data. [13][14]

Virtual assistants enable not only higher quality service, but also more consistent service. Virtual assistants are always available, enabling companies to provide constant coverage over weekends, holidays, and even adverse natural events like snowstorms and hurricanes. Virtual agents are also inherently more consistent than human agents in their customer interactions: level of experience, personal motivation, and current mood are no longer variables affecting the quality of service provided. [15] In particular, virtual assistants never experience emotional fatigue, a common problem among human agents, who are often faced with repeated hostile interactions with dissatisfied customers. Through virtual assistants, best practices are also immediately universally available to all customers, rather than slowly disseminating through meetings or training sessions. [16]

Ultimately, all this improvement in the quality and speed of service for customers has the potential to substantially boost brand value for companies: by lowering customers’ reluctance to connecting to customer support, virtual assistants could increase the volume of inbound queries and thereby provide the company a more accurate portrait of actual, real-time customer needs and pain points while solving latent issues customers may have previously decided were too much trouble to request the company’s help in solving. [6] Further, by constantly analyzing inbound requests and customer responses, virtual assistants facilitate more continuous and consistent data collection than a human agent could, enabling companies to identify and more quickly address those issues that are most common or critical to customer satisfaction (among other possibilities). Machine learning also has the potential to uncover optimal solutions humans had never before considered, providing opportunity for novel and superior problem solving. More effectively addressing more customer needs through virtual assistants has the potential to create substantial brand value.  Human agents, then, risk being outdone by virtual assistants on quality, speed, and availability of service: these bots create superior value directly and indirectly for the companies currently using human-staffed call centers.

The value capture model for virtual assistants is also more attractive to enterprise clients. [17][18] We’ll use Amazon Lex pricing as an example, which charges $0.004 per speech request and $0.00075 per text request processed. [19] For a call center representative overseas, in India, for example, (making ~$3,000 per year) to be more cost effective than a virtual assistant, the agent would have to answer over 75 calls a day, 365 days per year (assuming 4:1 speech to text requests, 30 requests per call, and zero holidays or weekends). [20] For a US call center representative, who makes an average of ~$31K per year, that number is over 800 calls per day, every single day of the year. [21]

Moreover, Amazon’s pay as you go model requires no upfront or minimum fees, making scaling up or down virtually costless for the company [22]; hiring, training, and firing human agents, on the other hand, is quite expensive, as is asking them to work overtime. Even when volumes aren’t highly variable, call centers in the US experience attrition rates of 30-45% (versus a US average of 15%) [17], meaning virtual agents eliminate a costly and logistically challenging element of the current call center model. Moreover, the enterprise no longer needs to pay for managers to supervise and motivate agents and likely needs only retain a smaller team of quality control agents. A third-party solution like Amazon Lex means enterprises don’t even need to invest much long-term in coders or data scientists, outsourcing that development to a tech giant.

Please hold

While the value creation described above largely takes a future state of near to full automation of call center requests for granted, there are certainly tasks that virtual assistant can’t handle today or potentially ever. When a home insurance customer calls to report that her/his house has burned down, for instance, today’s virtual assistant is simply not equipped to respond in an appropriately empathetic way. [23] Call center agents, however, are still ultimately likely to lose for two major reasons. First, even in the near term when virtual assistants can only handle the most routine, predictable tasks, this will inevitably siphon some of the volume from today’s human agents, meaning call centers will need fewer FTEs. As the algorithms take in increasingly quantities of data, it’s reasonable to assume that virtual assistants will progress up the complexity curve. Just look at Google’s Duplex versus the infuriating rigidity of yesterday’s endless interactive voice response (“For your account status, please press 1”). [16] Second, longer term, consumers are likely to become increasingly comfortable with digital interactions (and those interactions may feel increasingly natural as the technology improves), and they may simply no longer expect companies to provide individuals 24/7 for customers to chat with personally. [24][25]


The assistant will become the master

Call centers are likely just the beginning of the automation of service tasks. We may well see these virtual assistants assisting and then increasingly replacing medical professionals, marketers, as well as human personal assistants.




[1] Ryan Vlastelica, “Automation could impact 375 million jobs by 2030, new study suggests,” MarketWatch, December 2017,, accessed March 2, 2019.

[2] Chris Baraniuk, “How talking machines are taking call centre jobs,” BBC, August 2018,, accessed March 2, 2019.

[3] Bobby Hellard, “M&S to replace call centre staff with an AI chatbot,” ITPRO, August 2018,, accessed March 2, 2019.

[4] Dave Michels, “Please Hold for Next Available Bot,” no jitter, November 2018,, accessed March 2, 2019.

[5] Business Insider Intelligence, “80% of businesses want chatbots by 2020,” Business Insider, December 2016,, accessed March 2, 2019.

[6] Maurice Hage Obeid, Kevin Neher, and Greg Phalin, “Why your call center is only getting noisier,” McKinsey & Company, July 2017,, accessed March 2, 2019.

[7] LiveOps company website,, accessed March 2, 2019.

[8] Arthur Sheyn, “LiveOps,” Deloitte Insights, June 2013,, accessed March 2, 2019.

[9] Tarundeep Singh, “What are the most common issues that people call customer service for?” Quora, October 2016,, accessed March 2, 2019.

[10] Jon Walker, “Chatbot Comparison – Facebook, Microsoft, Amazon, and Google,” emerj, February 2019,, accessed March 2, 2019.

[11] AWS, “Amazon Lex,” AWS,, accessed March 2, 2019.

[12] Google Cloud, “Contact Center AI,” Google,, accessed March 2, 2019.

[13] J Arnold & Associates, “AI in the Contact Center,” Cisco, February 2018,, accessed March 2, 2019.

[14] Ratna Puri, “The rise of the chatbot in the contact center,” The Economic Times, September 2017,, accessed March 2, 2019.

[15] Akansha De, “A Look at the Future of Chatbots in Customer Service,” readwrite, December 2018,, accessed March 2, 2019.

[16] Ambit, “The Death of Call Centres,” Chatbots Magazine, November 2017,, accessed March 2, 2019.

[17] Trips Reddy, “How chatbots can help reduce customer service costs by 30%,” IBM, October 2017,, accessed March 2, 2019.

[18] Maruti Techlabs, “Can Chatbots Help Reduce Customer Service Costs by 30%?” Chatbots Magazine, April 2017,, accessed March 2, 2019.

[19] AWS, “Amazon Lex pricing,” AWS,, accessed March 2, 2019.

[20] PayScale, “Average Call Center and Customer Service Executive Salary,” PayScale,, accessed March 2, 2019.

[21] glassdoor, “Call Center Representative Salaries,” glassdoor,,26.htm, accessed March 2, 2019.

[22] Jeff Barr, “LiveOps Cloud – Tapping the Billion Dollar Call-Center Market on AWS,” AWS, April 2016,, accessed March 2, 2019.

[23] Spark Central, “Why chatbots in the contact center won’t replace live agents,” Spark Central, May 2018,, accessed March 2, 2019.

[24] Simon Erickson, “A Conversational AI Leader Is Going Public,” The Motley Fool, February 2019,, accessed March 2, 2019.

[25] BRAIN, “Chatbot Report 2018: Global Trends and Analysis,” Chatbots Magazine, March 2018,, accessed March 2, 2019.

Other sources

Dave Gershgom, “Google is building ‘virtual agents’ to handle call centers’ grunt work,” Quartz, July 2018,, accessed March 2, 2019.

Esther Shein, “CodeObjects taps Watson AI capabilities to develop chatbot offering,” SearchITChannel, February 2019,, accessed March 2, 2019.

GlobeNewswire, “2018 Call Center Industry Report Reveals Call Center Innovation Lagging,” Markets Insider, June 2018,, accessed March 2, 2019.

Kai-Fu Lee, “AI Could Devastate the Developing World,” Bloomberg, September 2018,, accessed March 2, 2019.

Manish Dudharejia, “4 Ways AI is Transforming E-Commerce Businesses,” TotalRetail, February 2019,, accessed March 2, 2019.

Parmy Olson, “Google, Microsoft And Startups Are Going To War On Chatbot Technology,” Forbes, July 2018,, accessed March 2, 2019.

Rufus Grig, “Humans and chatbots collaborating in the contact centre? Good call!” IT ProPortal, May 2018,, accessed March 2, 2019.

Thomas Wieberneit, “Are AI-Powered Chatbots Ready for Mainstream Adoption?” customerTHINK, October 2018,, accessed March 2, 2019., “Call Center Statistics,”,, accessed March 2, 2019.


Article header:

First photo:–7596551_7596552.html

Article video:

3 thoughts on “Hi, my name is C-3PO. How can I assist you today?

  1. Great post! You presented so much information that I hadn’t even considered. The broader discussion around automation’s impact on the workforce will definitely be something to watch in every industry over the next decade. Though it is hard to resist when the automated alternative is so much more cost effective and efficient than the human counterpart.

  2. Thanks for sharing this great post and analysis! I’ve managed call centers before. Employee retention, morale and welfare have always been the big pain points. When I dismissed a 50-person call center team due to its unjustifiable ROI, I even received some form of threatening from the call center manager… Definitely not something I want to experience again. But this brings up a intriguing question of how could we properly address the “losers” as a result of this digital transformation. Can we empower them to some extent that they could still add value to the digitalized world?

    Another amazing thing about this pre-recorded robot call is beyond the inbound call use case in handling customer support issues you mentioned, businesses can also utilize this technology to conduct outbound sales calls. During my internship at a VC fund, I visited a company applying similar technology to record scripts to sell insurance via outbound call. In fact in my trial I couldn’t even detect it’s the robot speaking with me because all the tones, pauses and pronunciations seemed so natural! It had a very sweet voice as well. Once I expressed my interest in hearing more about the product, the call was then handled by a real person. This will save a lot of labor cost because the conversion of people picking up such sales calls to becoming really interested is so low. The insurance company in this case could save tons of cost by only allocating real person to address complicated queries from customers and having the initial screening handled by robot.

  3. Marissa,

    Thank you very much for a deep and insightful post about this nascent industry. This technological transition and its implications for customer service and labor remind me of the transformation that AT&T (as the monopolistic provider of telephone services in the United States at the time) executed when it implemented dial machines in the 1930s.

    Then, as now, the expected growth in demand for communication services was expected to grow at a much faster rate than the available labor pool (at that time, for telephone operators. Additionally, the increased number of customers and possible connections made the operator’s task increasingly complicated, while automated dial machines could complete a connection in a relatively constant time regardless of the number of telephones in the network. Fears that a lack of an operator would result in a degraded customer experience were supplanted by the realization that connections were much faster and reliable. I feel that fears that a chatbot or voice assistant cannot deliver a satisfactory experience will recede as the technology improves.

    I also think that we can draw some lessons for the shifts in labor that this automation created. In the first ten years after implementing dial machines, AT&T lost approximately 70,000 jobs (mostly suffered by the younger, single women that AT&T employed as operators), while the majority of the ~20,000 jobs created for installation and maintenance of the dial machines went to engineers and electrical technicians. I expect similar shifts when virtual assistants start to become more prevalent.


    [1] Gross, Daniel, and William Kerr, “AT&T: Managing Technological Change and the Future of Telephone Operators in the 20th Century,” Harvard Business School Case No. 718-486. Boston: Harvard Business School Publishing.

Leave a comment