Customer service AI is often framed around deployment, but questions around which use case to select and how quickly the first agent can go live do not cover the full lifecycle.
Csaba Tamas, Chief Product Officer at Parloa, told CX Today that customer service AI needs a post-launch operating model that gives business teams the ability to build, monitor, improve and scale AI agents after deployment.
“We believe in a two-stage deployment,” Tamas said. “Stage one is that we deploy the platform in the enterprise, and that's an IT job.”
The first stage creates the technical foundation, covering installation, backend connectivity, customer relationship management (CRM) and (ERP) connections, system integrations and user access.
“This is to install the platform, to make the connections to the backend systems, to the customer relationship management, to the enterprise resource planning, to all the backend systems which the agent needs to interact with, configure the user access,” Tamas said.
Stage two moves responsibility closer to the business.
“Then, in stage two, the subject matter experts and the line of business leaders should take over the control,” Tamas said. “And they should start building their agents on top of the platform.”
Business Teams Need Practical Control
AI agents operate inside business processes, which change over time as processes change when policies are updated, product definitions move, new service conditions appear and customer demand changes.
If business teams cannot adjust the AI agent, the agent can fall behind the operation it is meant to support.
Tamas said subject matter experts and line-of-business leaders should be able to manage agents directly.
“They should be able to single-handedly create, test, deploy and monitor multiple agents.”
This is a different model from traditional project delivery, which gives the business more control over the agent’s lifecycle, while IT remains responsible for the platform foundation and core integrations.
Enterprise buyers should consider what happens after an AI agent is deployed; who can improve it after deployment, how changes are governed and how quickly the business can respond when customer conversations expose new gaps.
Treat AI Agents Like Employees
Tamas uses a workforce analogy to describe how enterprises should approach AI agent management.
“You need to be able to think of it as you think about your own employees. When you hire someone, you don't stop the interaction with that employee at that very moment.”
An employee needs coaching, feedback, performance reviews and development. Tamas said AI agents need a similar management rhythm.
“You need to have check-ins. You have to do coaching. You have to do performance review. You have to provide feedback and growth opportunities to your employee. The same is true for AI agents.”
Rather than suggest that AI agents should be treated as people, the point is that live service performance needs ongoing attention.
“You need to be able to observe how the AI agent is performing and always give little nudges, little reconfigurations to bring the agent to the right direction,” Tamas said.
A post-launch operating model gives teams a structure for those reviews and adjustments.
AI Agent Building Is Moving Through Maturity Levels
Parloa sees a maturity curve emerging in AI agent creation and management. Tamas said the company has borrowed a concept from autonomous driving, where capabilities are described in levels.
In AI agent building, at Level 1, specialists still do most of the work.
