Contact centers were never a walk in the park for employees, but they became much harsher work environments during the pandemic. According to one survey, only 10% of contact center agents reach proficiency in fewer than two months. Meanwhile, the volume of difficult calls to contact centers is increasing, while turnover remains at at a sky-high rate between 30% and 45%.
It’s against this backdrop that automation products are gaining interest from call center operators — and investors. On the more sophisticated end of the spectrum, call center automation promises to resolve customer service issues to free up agents for more complex work. Replicant, one of the more prominent vendors in the call center automation space, today announced that it raised $78 million in Series B funding led by Stripes with participation from Salesforce Ventures, Omega Venture Partners, IronGrey, Norwest and Atomic. Sources tell TechCrunch that the post-money valuation stands at $550 million.
“[With the new capital,] we plan to ramp up investment in our customer success team to onboard new customers,” co-founder and CEO Gadi Shamia told TechCrunch via email. “We also plan to double our R&D team this year to make our conversations even more efficient and launch new automated channels. We will increase our sales and marketing investment to capture the significant demand we see. And finally, we will invest in our employees by launching additional professional development programs.”
Shamia co-founded Replicant in 2017 alongside Andrew Abraham, Benjamin Gleitzman and Jack Abraham. Shamia was previously GM of product at SAP’s small business solutions group before becoming the acting COO at EchoSign after it was acquired by Adobe. He also helped to launch Magneto, a calendaring system, and was COO at Talkdesk for nearly four years.
Prior to Replicant, Abraham — who joined eBay in 2011 via the company’s acquisition of Milo.com — did stints as a software engineer at Atomic and smart device company Leeo. Gleitzman was a senior software engineer at Hunch and eBay before co-founding several startups including a “virtual reality therapy platform” called Mona.
“Through [my] work, I realized that the best way to increase agent efficiency and reduce customer and agent frustration is by automating many common tasks and letting agents focus on more complex and nuanced calls,” Shamia said. “Gleitzman was one of eBay’s AI pioneers and worked with Abraham and the Atomic team to build a machine that could have an entire phone conversation with a human.”
Replicant aims to automate call flows by integrating with existing systems including customer relationship management software to recognize customers by drawing on their order histories (if applicable) and past calls. The product can capture, transcribe and make searchable customer conversations, and — as do some rival service automation systems — Replicant can engage with customers through SMS and the web in addition to voice.
Replicant provides agents with call summaries and measures trends like overall customer satisfaction, average handle time, competitor mentions, defective products and upsell opportunities. Customers can draw on a library of prebuilt components to design conversation flows using a visual editor. In recent months, Replicant added support for new languages and conversational capabilities that Shamia calls “powers,” like holding on the line, repeating information “conversationally” and matching a customer’s response against a database.
“A core competitive advantage we have at Replicant is the rich and varied data we’ve amassed from tackling more than 30 million customer service calls across industries and use cases. Our [product has] tackled everything from hardware troubleshooting for small business owners, to relaying food orders to restaurant employees, to handling subscription issues for elderly callers, to high-urgency scenarios where callers need roadside assistance,” Shamia said. “[W]e turn scenarios that are commonly frustrating — think of every time you’ve had to go back and forth spelling out your name or reading off a 15-digit policy number to an agent on the phone — to a task that can be completed efficiently in seconds with a purpose-built model.”
When asked about how Replicant handles, stores and retains customer data, Shamia said that the company provides enterprise customers with the ability to choose a data retention window that “works for them,” usually ranging from six months to two years. For use cases involving payment or electronic protected health information, Replicant offers a service called highly confidential turn, which the company says redacts sensitive data in the turn of conversation from Replicant’s database and logs.
Replicant also engages in sentiment analysis, a controversial process that involves the use of algorithms to determine if a chunk of audio or transcribed text is positive, negative or neutral in tone. Sentiment analysis systems — both academic and commercial — have been shown to exhibit bias along race, age, cultural, ethnic and gender lines. Some algorithms associate Black people with more negative emotions like anger, fear and sadness. Others discriminate against non-native English speakers, who tend to use cognates — i.e., English words that look similar to the words of their native language — more often than native speakers.
Replicant claims that it only measures customer satisfaction by asking specific questions (e.g., “How satisfied are you?”) and takes steps to mitigate bias in its systems — including its sentiment analysis systems — as well as the data used to develop these algorithms. Unfortunately, without independent audits or studies to go on, it’s the company’s word against broad-based academic findings. This reporter hopes to see greater transparency from Replicant going forward.
“Our models are thus trained on a variety of accents, emotions and industry-specific jargon, allowing us to achieve [high] inference accuracy even on the most complex service use cases,” Shamia said. “We see an 85% call success rate (as measured by expected business outcome) across customers and use cases.”
Automating customer interactions
There’s anecdotal evidence to suggest that customer service organizations are embracing automation. A 2020 study from the Harris Poll, commissioned by AI vendor Interactions, estimates that 46% of customer interactions are automated — a percentage the co-authors predict will rise to 59% over the next two to three years. Early adopters surveyed for the study cite “soft benefits” like reduced wait times, faster customer complaint resolution, and technical support and personalization.
In response to the growing interest from industry, countless call center automation products have come to market in recent years — both from startups such as Replicant and incumbents including Google, Amazon and Salesforce. Replicant competes with RedRoute, Skit and Voximplant in addition to Ultimate.ai, a customer service tool designed to automatically field simple service requests.
Expert Market Research predicts that the global call center AI market will grow from $967 million in size in 2020 to $3.54 billion by 2026.
“During the last two years, customer service has been under constant pressure as ‘The Great Resignation’ has created persistent agent shortages. And changes in consumer behavior due to [the pandemic] and supply chain issues have driven massive spikes in call volume,” Shamia said. “Executives now understand that the problem can’t be ignored or outsourced, as customers are unwilling to wait hours on hold.”
But do customers appreciate — or even like — automated call centers? After all, automation lacks a human touch — it can’t necessarily de-escalate a frustrated caller. Worse, automation can deter customers from engaging with a brand in a way that might could trust. A poll by PointSource found that 80% of customers would prefer to talk to a human when resolving problems. Adding fuel to the fire, 59% of consumers in a recent PwC survey felt that companies have lost touch with the human element of customer experience.
And what about call center workers? Metrics could be held against them, and simple customer problems — while arguably not the best use of their time — can be satisfying to solve. Then there’s the fear that automation will one day take away their jobs.
Shamia acknowledges that some forms of automation, like poorly designed conversational bots, can act as a roadblock for customers and agents rather a solution. But he asserts that Replicant has learned from the mistakes of the past, allowing companies to automate call flows while enabling agents to focus on more challenging problems.
“The pandemic has accelerated a trend — automation in contact centers — that had already started and exacerbated many of the existing challenges in the customer service space,” Shamia added. “Automation is now part of the
strategic plans of more and more companies — something that will not change post-pandemic.”
Toward that end, 100-employee Replicant says it has “dozens” of enterprise customers who’ve used its tools to service over 8 million customers. Customer deal sizes range from the hundreds of thousands to millions in annual recurring revenue.
“In most of our deals, we are competing against the disbelief that technology can actually achieve the resolution rates our customers are seeing. However, we are also part of replacement cycles for older technologies,” Shamia added. “We also see DIY solutions … in some deals or legacy players like IPSoft’s Amelia.”
To date, Replicant has raised $110 million in venture capital. The San Francisco, California-based company plans to expand its workforce to about 200 people by the end of 2022.