ASAPP CEO on Scaling Enterprise AI

Katie Stein, CEO of ASAPP, is a seasoned technology and business executive with more than two decades of experience spanning corporate strategy, operations, digital transformation, and enterprise growth. Before joining ASAPP as CEO in August 2026, she served as CEO of Atain and spent eight years at Genpact, where she was Chief Strategy Officer and Global Business Leader for Enterprise Services and Analytics, overseeing corporate strategy, major service portfolios, client experience, and M&A initiatives. Earlier in her career, Stein held senior leadership roles at Mercer, including Global Chief Operating Officer for its $2.5 billion Retirement, Health and Benefits business, and worked as a Project Leader at Boston Consulting Group. Her career has largely centered on helping large organizations improve operational performance, deploy technology more effectively, and execute complex business transformations.

ASAPP is an AI company focused on transforming enterprise customer service through its AI-native Customer Experience Platform (CXP). Founded in 2015, the company develops technology that brings together AI agents, human expertise, customer data, and enterprise systems across voice and digital channels. At the center of the platform is GenerativeAgent, an AI customer service agent designed to understand conversations, reason through customer requests, take actions through connected enterprise systems, and involve human agents when additional oversight or judgment is required. ASAPP also provides tools for testing, monitoring, governance, and observability, allowing large organizations to deploy customer-facing AI while maintaining greater control over how automated agents behave.

Your career has taken you from strategy and transformation roles at BCG, Mercer and Genpact to leading Atain, and now to becoming CEO of ASAPP. What attracted you to ASAPP at this point in your career, and after your first 30 days, what have you identified as the biggest opportunities and priorities for the company?

ASAPP first came onto my radar during Winter Storm Fern earlier this year. At Atain, we were working every angle to create enough capacity for one of the largest U.S. airlines as thousands of disrupted travelers tried to get home. ASAPP didn’t just answer questions; its GenerativeAgent rebooked more than 3,800 passengers in six days, including complex itinerary changes. That showed me what enterprise AI looks like under real pressure. My first 30 days have reinforced that view. Our opportunity is to focus and scale: concentrate where ASAPP has the clearest right to win, help customers move from experimentation to production impact, and execute with greater speed and consistency.

ASAPP’s approach brings autonomous AI agents, deterministic workflows and human experts together within a single orchestration layer. How do you determine which decisions should be left to an AI agent, which should follow predefined rules, and which should involve human judgment?

We don’t assign an entire interaction to AI, a predefined workflow, or a person. We look at each decision based on its complexity, risk, and need for accountability. AI is well suited to understanding intent, interpreting context, and adapting to the customer. Workflows provide consistency when required steps or policies must be followed. Humans add judgment when a situation is ambiguous, exceptional, or consequential. Orchestration brings all three together within the same interaction. The objective isn’t maximum autonomy. It’s resolving more customer needs while applying the right level of control at each decision point.

ASAPP treats human involvement as an ongoing part of the AI system rather than simply an escalation path when automation fails. As AI agents become more capable, where do you believe humans will continue to add the most value in customer-service interactions?

Human ingenuity! A few months ago, I spoke with a senior CX leader who observed that a relatively small number of calls account for a disproportionate amount of handle time. These are often edge cases with no clear FAQ, knowledge article, or standard operating procedure. They are difficult because the organization hasn’t encountered – or codified – the answer before. Humans will continue to add the most value in those moments: interpreting ambiguity, balancing competing considerations, and applying judgment where there is no established playbook. But involving a person shouldn’t require abandoning automation and transferring the entire interaction. The AI gathers what’s needed and presents a person with a specific request and context; the human provides the judgment, and then AI continues the interaction. That expertise resolves the immediate issue, and helps expand what AI can resolve next.

Many enterprises have successfully demonstrated generative AI in pilots but struggle when they try to deploy it across millions of real customer interactions. What do companies most often underestimate when making the transition from an AI pilot to a production-scale agentic system?

Companies most often underestimate the long tail. Pilots usually test defined use cases under controlled conditions. At production scale, rare situations become everyday occurrences: customers change direction, provide incomplete information, or combine needs in ways no one anticipated. That’s why scaling agentic AI isn’t just a model challenge. It requires integration with enterprise systems, governance, monitoring, and a way to involve people when judgment is needed. The pilot proves that the technology works. Production takes the architecture and operating discipline to make it reliable at scale.

As enterprises move from a single AI assistant toward multiple specialized agents working together, what new technical or governance challenges emerge around orchestration, observability and accountability?

The new challenge is coordination. More agents create more handoffs, decision points, and opportunities for context or accountability to break down. Enterprises need orchestration that determines which agent should act, what information and permissions it receives, and how their work contributes to the customer’s resolution. Observability needs to extend across the full interaction so organizations can understand what happened and why. That becomes the foundation for governance. The advantage won’t come from deploying the most agents. It will come from coordinating all that specialized intelligence into one system that is reliable, traceable, and accountable.

ASAPP recently introduced Continuous Red Teaming, which automatically tests its AI systems against more than 50 classes of vulnerabilities as models evolve. How should enterprises rethink security when AI agents are no longer simply generating responses but are increasingly able to take actions on behalf of customers?

The stakes change when AI moves from answering questions to taking action. An unauthorized transaction or exposure of customer data is far more consequential than an inaccurate response. Enterprises need clear boundaries around what an agent can access, what it can do, and when human approval is required. Security also can’t be a test performed once before deployment. As AI becomes more capable, bad actors are innovating too. That is why ASAPP continuously tests against thousands of adversarial scenarios spanning more than 50 vulnerability classes. Security has to evolve as quickly as technology, so enterprises can expand what AI can do without compromising customer trust.

Generative AI is inherently probabilistic, yet customer-service processes often involve transactions, compliance requirements and other tasks where outcomes must be predictable. How can enterprises preserve the flexibility of generative AI while introducing enough deterministic structure to make these systems dependable?

The key is to separate the conversation from the business process. We think Generative AI should do what it does best: understand the customer, navigate ambiguity, and respond naturally. But critical processes – like verification, transactions, and compliance – should operate within step-based workflows that enforce the company’s rules and required sequence. That architecture gives enterprises the best of both: a flexible experience for the customer and predictable execution for the business. It also makes processes easier to test, audit, and improve as models evolve. Enterprises don’t need to make every customer conversation rigid to make AI dependable. They need to be very clear about where the AI can adapt and where the process absolutely cannot.

Contact centers have traditionally measured automation through metrics such as containment and cost reduction. As agentic AI becomes more sophisticated, what metrics should enterprises use to determine whether AI is actually improving the customer experience rather than simply reducing human involvement?

We are moving away from metrics like deflection and containment and toward what I call “verified resolution.” I want us to be accountable for understanding a customer’s intent and actually resolving it. As a customer, I want the experience to be pleasant, but more than anything, I want the reason I reached out to be resolved. When we measure that outcome, rather than simply whether AI kept someone from reaching a human agent, we align the technology with the job the customer actually needs done.

If one human expert can increasingly supervise or guide multiple AI-driven conversations, how do you expect the contact-center workforce to evolve? Do you see new roles emerging around AI supervision, training, governance and continuous improvement?

This is what gets me really excited about the future contact center operating model. Today, human agents are on the front line, often helping customers who are disappointed, confused, or angry, while spending much of their own time navigating complex systems to resolve those issues. It’s demanding work, and burnout is high. As agentic AI takes on more of that work across voice and digital channels, I believe we’ll see a new class of CX professional emerge: the human agent as analyst. Their time will increasingly be spent on situations that require judgment, aren’t black and white, or carry regulatory sensitivity. That’s a fundamentally different role, and one that AI itself is helping create.

After your first 30 days leading ASAPP, what has surprised you most about where enterprises currently are in their adoption of agentic AI, and what do you believe will define the next stage of AI-powered customer service?

I’ve been impressed by how willing the clients I’ve spoken with are to make agentic AI a core pillar of their CX strategy rather than something adjacent to it. The challenge is that Digital, IT, Security, and Operations are often moving at different speeds and measuring success differently. That fragmentation makes it difficult to see the enterprise business case clearly. The next stage will require organizations to align around what actually matters at the enterprise level and understand where agentic AI in CX can move those outcomes. Today, that impact can get diluted across existing scorecards and siloed metrics.

Thank you for the great interview, readers who wish to learn more should visit ASAPP.