Contact Centers Move Agents Up The Complexity Curve As AI Takes Tier One

Every wave of contact center automation removes the simplest work from the queue. An IVR routes the call, a chatbot closes the password reset, and a customer who reaches a live agent has usually already tried to solve the problem alone. The calls that survive that filter are exceptions, escalations, and conversations where money or emotion is already in play. Whoever picks up needs context and the authority to act on it, a different job from the one contact centers typically build their training around.

Andrés Baldrich is the former Director of Customer Experience at Scotiabank, where he led a Canada-facing service operation of more than 1,000 employees. Baldrich spent more than a decade with the bank across Colombia, the Dominican Republic, and Canadian-facing businesses, scaling the Dominican operation from roughly 40 to 600 staff. He also worked operational workstreams on two banking integrations: Scotiabank's acquisition of a majority stake in BBVA Chile and Scotiabank Colpatria's purchase of Citibank's Colombian consumer business.

"People are calling to speak to a human when they can't solve problems for themselves. Contact center leaders have to move their people into more specialized subjects. AI is going to be your co-pilot, not autopilot," says Baldrich. An agent handling those calls can also spot when a customer is sitting in the wrong product, and a service call becomes a revenue conversation. Getting people to that level is the part he says leaders keep underestimating.

Complexity over headcount

Baldrich counts several automated decisions between a customer dialing and a person answering. Segment, product mix, and the stated reason for the call decide where the interaction lands, and software makes each of those judgments. He puts weight on how that sequence reads from the customer's side. "Whenever you are a customer, voice AI, you instantly know that it's voice AI," Baldrich says. "You are not calling because something is working perfectly fine."

Executive appetite for automation points either at headcount or at complexity, and Baldrich watches organizations choose. Aiming it at complexity means lifting the ceiling on what a single agent can resolve. He argues in favor of that route, with a caveat about volume. "Of course there's going to be some reductions, because you don't need the same amount of people to do the same amount of work," he notes. "But we have to evolve into a more complex situation on how we can solve better."

Training for the harder call

Contact center training was built around memorization, with an agent expected to hold processes, edge cases, and product rules in their head. Empathy was treated as something a hire already brought with them. Baldrich argues that ordering has now inverted, with the memorized material the easiest part to replace. "AI is solving one of the main issues, and it's repository, where to find the information," he explains. "Ten years ago, trainings were based basically on knowing the product."

Baldrich wants the hours that memorization used to consume spent on scenario work. Simulation and sandbox exercises put an agent through a full interaction before they meet a customer, including the cultural variation that changes what a good response sounds like. A caller's expectations shift by market, and the same reassurance reads differently in each one. "We need to keep on focusing on how to train our people to serve humans, to understand situations, to retain," Baldrich says. "Every company has AI now."

An investment question and a password reset aren't the same conversation, and Baldrich thinks many training programs treat them as if they were. A new hire arriving straight from high school has rarely managed a credit card, a line of credit, or a retirement plan. He would put that ahead of any product module in a new agent's first weeks. "Most people don't know basic personal finance," Baldrich adds. "If you don't have those basic first steps, then how can you advise a customer about their mortgage and what account suits them better?"

Measuring the whole picture

Real-time assistance is where Baldrich sees the biggest operational payoff. A model listening to a conversation can surface relevant processes while the agent stays on the customer. The knowledge problem then moves to the background, and the agent keeps attention on the part automation handles poorly. "The human can solely focus on how that person is feeling and how they can help to fix the issue," Baldrich explains.

Efficiency dashboards are where Baldrich thinks leaders tend to get it wrong. A model handling lookups shortens calls, and a scorecard weighted toward speed will register that as an improvement. He suggests churn, satisfaction, and repeat contact read alongside the efficiency numbers before anyone calls the implementation a success. "To measure the correct implementation of an AI environment, you need to take a look at the whole picture," Baldrich concludes. "Not just removing heads or reducing the time that we spend with the customers, but how we are impacting the long term relationships."