Authenticx CTO offers AI governance guidance
A patient calls a health-care provider to discuss what happened in their most recent therapy session. The quality assurance session is automated by artificial intelligence (AI), which helps identify areas of friction and opportunities for improvement.
As Michael Armstrong knows, though, success with those AI-powered conversations depends on spotting and mitigating what he calls adverse events. These could include situations where an AI tool doesn’t understand what the patient says or makes some kind of mistake. This is where AI governance in customer experience (CX) design becomes integral to preserving customer trust.
Armstrong is the CTO for Authenticx, a technology provider based in Indianapolis that specializes in call center analytics software built for the health-care industry. In March, Authenticx released AI for Contact Center Quality Management, a platform designed to automate evaluations, provide human agents with critical context and deliver personalized coaching.
Armstrong said the need for effective AI governance in CX has only become more important as the technology is deployed in customer-facing situations. He said Authenticx’s approach involves developing a rubic that breaks down examples of adverse events in AI conversations and policies that can help avoid them.
“The rubric itself is not AI, but whatever we define in this rubric will eventually make its way into our AI,” Armstrong told 360 Magazine. “If you build a system where you’re highly confident that your AI is going to give you the right answer, then you’re in really good shape as it relates to governance.”
Keeping humans (with expertise) in the loop
Authentcix has also chosen to keep a human in the loop by deliberately hiring employees with a background in health care, such as social workers and nurses, who have the domain expertise necessary to label data and review how its AI tools are working.
While some organizations are setting up teams and committees internally to establish AI governance controls, Armstrong said many are looking to vendors like Authenticx as a trusted source of expertise.
“Most of our clients are really asking for help (in this area),” he said. “They’re really sort of struggling, and that’s another reason why we feel it’s really important to go in with a with a point of view on how to handle this.”
When the AI bias concerns don’t add up
AI governance in CX is challenging in part because organizations are trying to covert the technology from probabilistic to deterministic outcomes which are explainable. Armstrong said his firm gets lots of questions and concerns about the potential for AI bias, for instance, without realizing where the bias originates.
“There’s no bias in math,” he said. “Where the bias comes from is actually in the training data, and what the labels are in the training data. How you structure your training data – for example, do you have a minority class that, like, is not included in your training set?”
Good AI governance also recognizes that the technology might never be perfect, but provides a feedback loop that allows for learning and continuous improvement, Armstrong said.
Making CX teams part of AI governance conversations
Organizations that want to improve AI governance may need to be more consultative and inclusive with their CX teams.
Armstrong said he couldn’t recall a customer situation where a CX leader was directly involved. It’s an area that is seen as more of a cybersecurity leader’s jurisdiction. But the best governance strategies don’t merely avoid risks: they make for better experiences.
“Every day there seems to be a new definition of what this really means,” he said. “I call it intentional AI. We just try to look at it and say, ‘I want this (process) to be predictable. I want it to be explainable. I want it to be as deterministic as possible.”
