AI in customer support has moved past the hype cycle into practical utility. But not all AI implementations are equal. Some genuinely reduce ticket volume and help agents. Others are rebranded keyword matching with a chatbot skin.
AI that helps agents (not replaces them)
The most practical AI in support today drafts responses for agents to review and edit, summarizes long ticket threads so agents get context fast, and suggests relevant knowledge base articles. Tools like FyneDesk AI Assist and Intercom Fin take this approach, keeping humans in the loop while reducing the mechanical work.
Customer-facing AI portals
The newer approach is AI built into the customer portal itself. Instead of making customers search your knowledge base, they ask a question and get an AI-generated answer pulled from your docs. FyneDesk and Intercom both offer this. When it works well, it deflects 20-40% of tickets before they are created.
The cost question
Intercom charges $0.99 per AI resolution. At 500 resolutions per month, that is $495 on top of your seat licenses. FyneDesk includes AI features in its free plan. Zendesk requires their most expensive tier for AI. The value calculation depends heavily on your ticket volume.
What AI cannot do yet
Handle emotionally complex situations. Make judgment calls about refunds or exceptions. Understand context that requires business knowledge not in your docs. Agents are still essential. AI just handles the repetitive pattern-matching that burns them out.