The 4 eras of customer support
Firefighter era
First customer support professionals joined in the 1960s and 70s as companies set up centralized offices to handle inquiries via telephone. This was the Firefighter era when customer support followed a predictable, albeit inefficient, pattern: a customer encountered a problem, reached out for help, and waited for a solution. The support team was in-house or outsourced to a call center somewhere where hiring costs were lower.
Support was inherently reactive. The organization was always one step behind the customer. For decades, support professionals relied on basic databases or even physical manuals to retrieve information and provide solutions.
Success was measured by time of resolutions — yes, the goal was to get the customer off the phone as quickly as possible, often at the expense of true resolution quality.
Scaling services was the main issue at this time. If you had 10% more customers, you had to hire 10% more people. Support costs eventually ate growth margins and became a financial nightmare for many companies.
Process era
The 1990s and early 2000s ushered in the digital revolution, bringing email, live chat, and Customer Relationship Management (CRM) software into the mainstream. Thankfully, I joined the industry at this point!
Support was no longer just a phone call; it was a digital thread that could be tracked, assigned, and measured. Websites began providing FAQs and basic search functions, allowing users to find some answers without calling in or texting. Workflows allowed us to route questions to the correct department or queue which brought a significant reduction in simple support tickets.
The problem of this era was the deflection trap. We got so good at hiding from customers behind FAQs and IVRs that we forgot to actually solve their problems. We lost intimacy and turned customers into ticket numbers.
Understanding era
In the 2010s, the introduction of Neural Language Models (NLM) and Natural Language Processing (NLP) fundamentally changed how machines interacted with human text and language. Support moved from matching keywords to understanding intent. Unlike the rigid bots, NLM-powered systems interpret complex queries. If a user says "I'm having trouble getting into my account," the system doesn’t just look for the word "account"; it understands the intent was "login issue".
Systems handle variations in language, regional slang, and typos to detect customer intent or when a customer is frustrated or angry, allowing for smarter routing and even resolution of simple queries.
The problem is that understanding requires intent libraries. We had to manually program every way a person could say "I lost my password" for our workflows and automations to work.
Agentic era
We have now entered the fourth epoch, defined by Large Language Models(LLMs) and agentic AI. This is the era of contextual understanding and predictive strategies where the AI doesn't just suggest a fix—it executes it.
Instead of waiting for a ticket, AI monitors signals—like billing anomalies or product telemetry—to intervene before the customer notices a problem. The reason an AI agent can resolve a ticket is its ability to call tools. Through secure API integrations, the agent can update a CRM record, process a refund in a billing system, or trigger a diagnostic test on a user’s account. This turns the support interface into a concierge service rather than a place to get homework assignments on how to fix things yourself.
What is next for us, support professionals?
The narrative that AI will replace human support completely is rubbish. The role is undergoing a flight to quality. As routine work (password resets, order tracking, etc.) is handled by agents, humans are moving into high-value strategic roles.
Yes, there will be less support jobs, but they will be more demanding. Support professionals will:
— Focus on the deep technical troubleshooting of unique, multi-layered issues that fall outside documented
scenarios.
— Manage the freshness and structure of the content the AI relies on, ensuring the systems’ brain is always
accurate.
— Tune prompts, set escalation rules, and audit the AI's performance.
— Treat the AI agents like a new employee that needs coaching.
— Use data-driven insights from AI interactions to suggest product changes that reduce the need for support in
the first place.
The best support professionals will be the ones who can look at the data and say, “The AI is failing on 15%
of billing queries because Article X is outdated—I'm fixing the docs now”.