A recent interview with Brian Schiff, circulating via SaaS Interviews, puts a striking claim on the table: AI-driven automation in customer support can reduce inbound call volumes by up to 90%. The supporting data point is hard to ignore — 300 million calls already handled by AI systems. For operations leaders, this is either the most exciting headline of the quarter or the most dangerous one, depending on how you read it.

Let's read it carefully.

What's Actually Happening in the Market

The claim isn't pure hype. Conversational AI has matured significantly over the past 18 months. Large language models now handle intent recognition, multi-turn dialogue, and resolution logic well enough to close out a meaningful share of tier-1 support interactions — password resets, order status updates, policy clarifications, basic troubleshooting. At scale, across high-volume contact centres, that does add up to dramatic deflection rates.

Pricing models are also evolving, as Schiff notes. The shift away from per-seat licensing toward outcome-based or consumption-based pricing makes AI automation more accessible to mid-market operations — not just enterprise players with eight-figure tech budgets. That changes the competitive landscape faster than most CX leaders have planned for.

So yes, 90% deflection is technically achievable in narrow, well-defined use cases with high query repetition. The number is real. The context around it, however, matters enormously.

Where the 10% Actually Lives

Here is the part that tends to get left out of the headline: the calls that AI cannot handle are disproportionately the ones that matter most to your customers and your brand. Complaints escalating to churn risk. Vulnerable customers in distress. Complex, multi-policy queries that require judgment. Situations where tone, empathy, and cultural nuance determine whether a customer stays or leaves.

That remaining 10% is not a residual rounding error. It is the highest-stakes, highest-emotion, highest-consequence slice of your entire contact volume. Routing it poorly — to an undertrained agent, to a bot that loops, or into a queue with no escalation path — is where organisations actually lose customers at scale.

The 90% reduction story is only operationally sound if the 10% is handled with more skill and care than before, not less.

The Operational Model That Actually Works

This is precisely why the hybrid intelligence model exists — and why it is gaining ground among serious CX operators rather than fading as a transitional compromise. Automation handles volume. Humans handle complexity and consequence. The two layers are not in competition; they are structurally dependent on each other.

What this means in practice for customer operations leaders: your AI layer needs to be trained tightly on your actual query taxonomy, with clear escalation logic and handoff protocols that do not create friction. Your human layer needs to be staffed with agents who are genuinely equipped for complex, emotionally charged interactions — not agents who spend 80% of their time on routine tasks that a bot could handle, and are therefore under-skilled and under-motivated when the hard cases arrive.

It also means that multilingual capability matters more, not less, as AI takes on more volume. When a frustrated customer finally reaches a human agent after an automated interaction, that agent needs to meet them in their language and their register — not default to a lingua franca that creates distance at exactly the wrong moment.

What Smart Operations Leaders Should Do Now

Treat the 90% figure as a design target for your automation layer, not a finished outcome. Audit which query types genuinely qualify for full automation versus assisted resolution. Invest in the human tier as your differentiation layer — hire for language capability, emotional intelligence, and domain knowledge. And build handoff architecture that is invisible to the customer: the transition from bot to human should feel like continuity, not a reset.

The AI revolution in customer support is real. The organisations that will lead it are the ones that understand automation and human expertise as a single, integrated capability — not as a replacement story with a tidy percentage attached.