Insights

AI agents: customer service moves to agentic interfaces

Not today's chatbots, but agentic AI that resolves cases end-to-end. By 2040 the contact centre shrinks 70-80%.

Discuss Your Challenge

What is agentic AI vs current chatbots

Today’s chatbots: scripted decision trees, fallback to a human when “not understood”. Customer experience often frustrating.

Agentic AI by 2030-2040: models that can:

  • Understand customer context (profile, recent activity, history).
  • Talk to several internal systems at once (billing, network, CRM, partners).
  • Perform actions (refund, plan change, ticket creation, partner contact).
  • Reason through multi-step problems.
  • Escalate to a human only when genuine judgement is required.

By 2040 this is standard tier 1 + a meaningful share of tier 2 service.

What changes in operations

Contact centre shrinks. Tier 1 → 80% by AI, tier 2 → 50%, tier 3 → 20% (where judgement essential).

Skill shift. Operators are not “answering questions” but “training, supervising, escalation handling”.

Operating cost per case drops by an order of magnitude. Cost per AI resolution in cents, per human — tens of dollars.

Quality measurement changes. NPS, FCR, repeat-rate per AI agent vs per human.

Where it gets hard

Trust. Customer must trust an AI agent with financial decisions. Build slowly.

Liability. Agent made a mistake — who is liable? Insurance, regulatory framework needed.

Edge cases. AI is good on 80-90% of cases, the rest needs human judgement. Risk of under-escalation.

Cultural fit. Does the AI speak the dialect? Customer perception varies.

What operators must do now

Pilot agentic capabilities in narrow scope (e.g. balance check, simple plan change).

Quality benchmarks against human baseline.

Customer consent — explicit choice “AI agent or human”.

Talent retraining for supervising / escalation roles.

Liability framework engagement with regulator and insurance.

Where the cracks are

Hyperscalers (OpenAI, Anthropic, Google) offer voice + agent capabilities. Operator becomes integrator, not AI builder.

Customer fatigue. If the AI agent is already at the bank, retailer, operator — customer expects consistent quality.

Regulator concerns. Automated decisions especially in financial / billing — heavily regulated in the EU (GDPR Art 22), soon in UZ.

← Back

Ready to discuss your challenge?

Tell me what's not working or what needs to be built. First conversation — no obligations.

Usually respond within a few hours

Discuss a challenge
Choose a convenient way to connect
Telegram
Fast reply
Fast
WhatsApp
Voice and documents
📞
Call
+998 99 838-11-88