New AI technology could help sellers analyse customer’s emotions. But how will people react if their facial expressions and body language are tracked?
The sales rep might not be the only one watching you on your next virtual meeting. Some companies are employing ‘emotion AI’ – a subset of AI that is able to detect human emotions – to monitor individuals on sales calls, provide feedback on their reactions and highlight the most engaging parts of the pitch.
The market for such sales enablement platforms is growing, with consultancy firm Verified Market Research claiming it could be worth $7.3bn globally by 2028.
Zoom is one of the latest entrants to this burgeoning market, launching Zoom IQ for Sales last month. Described as a conversational intelligence software, it claims to offer “meaningful and actionable insights from customer interactions to improve seller performance and enhance customer experiences”.
However, the announcement was met with opposition from a number of human rights groups and privacy campaigners. An open letter to Zoom CEO Eric Yuan – co-signed by the American Civil Liberties Union and the Electronic Privacy Information Center – called for the video communications company to halt its plans to advance the feature. “This move to mine users for emotional data points based on the false idea that AI can track and analyse human emotions is a violation of privacy and human rights.” Zoom declined to be interviewed for the piece.
According to Uniphore, one of the companies behind the technology, its emotion AI assistant ‘Q for Sales’ can “help sellers ‘read the room’, sense emotional cues, and improve engagement”.
Patrick Ehlen is vice-president of AI at the company. He thinks the fears over emotion AI come from a misunderstanding of the technology.
“Personally I’m not a huge fan of the term emotion AI,” he says. “There seems to be a concern that when people talk about emotion AI, they’re talking about a computer system that can actually read your internal emotional state with a high degree of accuracy. The truth of the matter is there’s no AI system that can do that.”
Getting AI to understand context
Psychological research backs up this assertion. The way that people communicate emotion is not universal and facial expressions are much more contextually sensitive than previously thought, according to a 2019 paper from psychology professors at the California Institute of Technology, Northeastern University, University of Glasgow and University of Wisconsin–Madison.
Lisa Barrett, professor of psychology at Northeastern University and one of the paper’s authors, believes the findings “call into question the scientific justification used by some emotion AI technologies”.
Ehlen is therefore reluctant to make any bold claims about the technology’s capabilities. He maintains that “it’s an unreliable technology to be making very big, critical decisions”. However, he thinks it could prove useful on sales calls. In situations where a salesperson may be speaking to multiple people on a video it can be hard to determine who is engaging and who isn’t.
“You’re looking at people in this tiny little window, so it’s not as easy to read the room and see what people’s facial expressions are,” he says. “If you can have a machine gauging their reactions and determining at what point the CFO seems interested, it can be a helpful tool.”
To do this, the visual AI software analyses a number of elements of the conversation, including facial expressions, tone of voice and people’s gestures. “From that information, we’re able to get much closer to having a 360-degree view of what people are doing when they’re in conversation to better understand them,” Ehlen adds.
The platform can then give salespeople real-time feedback on sentiment and engagement to help them adapt their responses and, in theory, improve their sales conversions.
Sybill is another platform that claims to use emotional intelligence to accelerate the sales process. However, co-founder and CEO Gorish Aggarwal is reluctant to say the programme can identify people’s emotional state. “We consider it as a behaviour AI, which is different from emotion because emotions are very subjective,” he says. “You cannot tell whether a person is fearful, angry or contemptuous just by looking at their face.”
The software instead looks to identify an individual’s body language or facial expression to highlight key moments, Aggarwal says: for example, whether someone is nodding along to a conversation or smiling. Salespeople can then revisit the recording to see when people were most engaged and which parts they should follow up on.
Although Aggarwal claims its AI can determine whether someone is nodding their head with up to 95% precision, the technology’s impact in the sales context is yet to be proven. Both Uniphore and Sybill’s platforms launched earlier this year and are both currently conducting valuation studies to determine to what extent their AI programmes can improve sales performance.
Jason Bell is associate professor of marketing at Saïd Business School and works on AI models for computer vision and natural language processing. He believes there is a lot of ‘overclaim’ in the emotion AI market.
“Many webcams and front-facing smartphone cameras are low quality, so even if you have a great predictive model, the signal that it’s getting is not great,” he says. “In principle it’s possible, but I haven’t been convinced by the current technology.”
Relying on this technology in a sales context could lead people down dead ends, according to Bell. His biggest concern centres on the technology’s accuracy. “It dramatically oversimplifies emotional states,” he adds. “If you’re monitoring one or both parties involved in the sales process and categorising their emotions, it can be quite crude … You could create more complications for the sales process without adding a tonne of value.”
Concerns around privacy, as raised in the open letter to Zoom, have also made people wary about implementing the technology. Uniphore and Sybill are both aware of the concern.
“We’re very aware that some people see AI as being somewhat creepy,” Ehlen says. “So, we’ve tried to build in as many protections and safeguards as possible to make people feel comfortable.”
As a result, Q for Sales is an opt-in experience. With Sybill, users are notified that “the call is being recorded and notes are being taken,” although there is no reference to AI unless the user decides to reveal this.
The iterative process of using AI
Aggarwal doesn’t recommend using the programme for internal calls or in situations where there’s a power imbalance. For example, using it as part of an interview process would be “ethically not correct as they wouldn’t be able to refuse”.
There’s also a risk that people change their behaviour when they’re made aware of being recorded. Ehlen admits this could mean “the AI is not going to work as well as it did before” and will have to go through a process of retraining.
“It’s an iterative process,” he says. “As these technologies make their way into the mainstream people will become more comfortable with them and this will become less of an issue.”
Whether these conversational AI programmes become a standard part of the sales process remains to be seen, but convincing customers to be monitored by AI might be the hardest sell.