Skip to main content
Perspective:

Serving the hybrid customer

Designing a hybrid-to-hybrid telco CX for humans and their AI agents

The next customer is hybrid – sometimes human, sometimes their AI agent – so telcos must be hybrid too. But to make this work at scale, customer experience (CX) journeys must be machine-readable, policy clear, consented, and observable, and execution must run through multi-agent orchestration with risk-aligned human-in-the loop oversight.

As artificial intelligence (AI) moves from experimentation to enterprise-grade capability, telcos are approaching an inflection point: the CX interface is becoming hybrid, with people and their AI agents engaging on both the customer and telco sides.

The good news for telcos is that they already hold an unfair advantage. They have the data – behaviours, interests, context, intent, real-time location, mobility, and network telemetry. They also hold the trust – identity, transactions, consent, and customer control. Few industries operate closer to the moments that matter.

The challenge now is converting these raw advantages into an operating system for hybrid-to-hybrid CX orchestration: making products, policies, and journeys machine-readable; exposing the logic behind offers and decisions; embedding consent and delegation so agents operate within authorised boundaries; and enabling autonomous completion with human-in-the-loop oversight aligned to outcome criticality.

Four hybrid-to-hybrid CX interaction archetypes

Hybrid-to-hybrid CX is not one experience; it is four interaction archetypes. They differ in terms of who leads the work and where humans intervene.

Four imperatives to realise the full value of hybrid-to-hybrid CX

In a hybrid, agent-mediated world, the question is no longer what a user will click, but what an agent can reliably complete – and what should remain human-led.  Hybrid-to-hybrid CX should adapt to customer intent – specifically, whether they are assurance-seeking (human touch), autonomy-seeking (self-serve) or delegators (agent-first).

Build an always-on signal layer that makes products, policies, entitlements, quality of experience (QoE), and usage legible to machines in real time.

Possible outcomes: Micro-upsells at moment of relevance; dynamic plan optimisation; better customer retention through effective risk prediction and proactive remediation; a foundation for new AI-enabled services built on unique telco data.

Expose stable, deterministic APIs for high-value intents and express rules as policy-as-code with clear inputs, side-effects, and outcomes.

Possible outcomes: Upsells execute immediately (no ‘next billing cycle’, no decision fatigue); autonomous fixes that close loops without escalation; execution rails make agent-native bundles operable for new sources of monetisation (personal assistants, context-aware travel/safety, home/device or orchestration.)

Design agent journeys to be configurable by customer preference and context.

Possible outcomes: Higher conversion and attach by matching experience to intent and interaction preferences; improved retention by protecting high-touch moments and compressing handling time; reduced revenue leakage through thresholds and triggers for agent-to-human escalation.

Provide granular identity, consent and delegation so customers – and their agents – can opt into value on their terms.

Possible outcomes: Consent, delegation, and auditability unlock external agent access to context signals; predictability, transparency, and policy clarity increase win-rate when agents compare offers.

Alas, a final word of caution: telcos must not repeat the mistake of the over-the-top (OTT) era, when value shifted to digital entrants who understood the interface better. In the age of hybrid-to-hybrid CX, the interface is not the screen, but the shared space between human and agent. Whoever designs for both human and agent wins the customer.

Ultimately, the future of AI is human: agents deliver speed, but humans anchor trust through clear delegation, explainability, and recourse.

Did you find this useful?

Thanks for your feedback