emiliosbestinsights.rivetgarden.com
←Back

Gartner says 91% of CX leaders feel pressure to use AI — what should I do first?

Gartner recently reported that 91% of customer experience (CX) leaders feel pressured to integrate AI into their support operations. The promise is compelling: faster resolutions, 24/7 availability, and personalized interactions at scale. But the path from enthusiasm to effective execution isn’t straightforward — especially in regulated or high-volume environments where customer trust and compliance can’t be compromised.

If you’re looking to develop a CX AI strategy and move from pilot to production, this guide will help you focus on what matters first. We’ll explore essential considerations for vendor selection, the critical distinction between AI that assists versus AI that resolves, and the importance of automation levels, deterministic workflows, and enterprise-grade observability.

AI in CX: Assist vs. Resolve

One of the most common misconceptions vendors pitch is the idea of fully autonomous AI agents that “resolve” customer issues end to end. In reality, the technology is still maturing, and many organizations benefit most from AI that assists human agents rather than replacing them immediately.

  • AI that assists: AI supports agents by suggesting next best actions, providing quick access to relevant knowledge, or routing conversations intelligently. This improves efficiency while maintaining human oversight.
  • AI that resolves: AI independently handles certain interactions without human intervention. This requires highly robust, deterministic workflows and is best applied to narrow, well-understood issues.

Companies like Zowie offer pricing models based on per-conversation pricing, which allows you to start small by letting AI handle routine questions and escalate complex queries to humans. This can balance cost efficiency with risk management.

The Three Levels of Automation: Knowledge, Process, and Orchestration

When developing your CX AI strategy, think beyond just chatbot interfaces. Automation in CX typically occurs at three progressive levels:

  1. Knowledge Automation: Use large language models (LLMs) and retrieval-augmented generation (RAG) techniques to automatically surface relevant answers from your knowledge base. Tools like Intercom’s Fin leverage AI to enhance knowledge searches, reducing time-to-respond.
  2. Process Automation: Automate repetitive tasks and decision rules, such as ticket categorization, priority setting, or customer data retrieval. This reduces manual effort and error.
  3. Orchestration: Coordinate multiple systems and workflows end-to-end, including handoffs between AI and humans, multi-channel engagement, and compliance checks. Platforms like Zendesk’s Advanced AI are evolving toward this level.

Example: From Knowledge Automation to Orchestration

Let’s say a customer asks, “How do I reset my password?” At the knowledge level, an LLM enhanced with RAG fetches the official help document immediately. At the process level, the AI automatically creates a ticket with the correct tags and priority. At orchestration, the AI could escalate to a human if multiple failed reset attempts are detected, log all interactions in the CRM, and trigger compliance alerts if needed.

Deterministic Workflow Execution to Reduce Hallucination Risk

One frequent risk with LLM-powered agents is hallucination—when AI confidently provides incorrect or fabricated information. To mitigate this risk, deterministic workflow execution is essential. This means your AI should follow explicit, predefined logic paths, verifying outputs at each step rather than generating freeform responses unbound by rules.

Deterministic systems balance AI creativity with business logic, compliance requirements, and error handling. For regulated industries like finance or healthcare, this approach is a must-have. When evaluating vendors, ask to review their workflow engines and see how they prevent or log hallucinations.

Enterprise Observability, Audit Trails, and Compliance

AI in CX isn’t just about answering questions faster—it’s about trust and accountability. Enterprise observability and audit trails provide transparency into AI decisions, actions, and customer interactions. This is critical for:

  • Regulatory compliance: Ensuring every automated action can be traced and justified.
  • Quality assurance: Monitoring AI accuracy and detecting bottlenecks or failure points.
  • Risk management: Quickly identifying and addressing unexpected AI behavior or customer dissatisfaction.

Platforms like Zendesk Advanced AI have started embedding audit log features so your teams can review conversation transcripts, AI suggestions, and human overrides at any time. When selecting vendors, insist on seeing their observability dashboards and audit trail capabilities before committing.

Making Vendor Selection Decisions: What to Look For

In the rush to adopt AI, vendors often lead with flashy feature lists but fall short on real-world operational needs. Here’s what I ask and check—your “due diligence checklist” for vendor selection:

Criteria Why It Matters Example Vendors Pricing Model Look for transparent, scalable pricing that rewards better performance, not just headcount. Per-conversation pricing (like Zowie’s) can align costs to impact. Zowie (per-conversation pricing) LLM & RAG Integration Advanced language models enhanced with retrieval-augmented generation improve accuracy by grounding AI on your data. Intercom Fin, Zendesk Advanced AI Workflow Determinism Predefined, rule-based workflows reduce hallucinations and support compliance. Zendesk Advanced AI Audit Trails & Observability Essential for compliance, risk mitigation, and continuous improvement. Zendesk Advanced AI Support for Pilot to Production Flexible deployment options and strong vendor support help scale successful pilots into enterprise rollout. Zowie, Intercom, Zendesk

Pilot to Production: Start Small, Then Scale

The safest way to harness AI’s potential without causing headaches at 2 a.m. is to start with a manageable pilot. For getzowie example:

  1. Identify a high-volume, low-risk use case — such as password resets or FAQs.
  2. Deploy AI-assisted workflows with human-in-the-loop review to build confidence.
  3. Collect audit logs and analyze performance, focusing on KPIs like Cost Per Resolution and Customer Satisfaction.
  4. Address gaps in workflow determinism and compliance based on pilot learnings.
  5. Expand automation scope gradually, adding orchestration capabilities and multi-channel support.

Platforms with transparent pricing models and strong observability (like Zowie’s per-conversation pricing paired with audit trails in Zendesk) enable iterative improvement without unexpected budget overruns.

Conclusion: Focus on Outcomes, Not Just Features

Integrating AI into your CX operation is less about adopting the latest flashy tool and more about aligning technology with business outcomes. Prioritize solutions that:

  • Deliver measurable cost savings or resolution speed improvements
  • Offer deterministic, auditable workflows to minimize risk and comply with regulations
  • Start with AI as an assistant and evolve towards resolver capabilities at scale
  • Support your journey from pilot projects to robust production environments

If you keep those principles top of mind and evaluate vendors grounded in real-world operational needs — not just marketing hype — you’ll feel confident addressing Gartner’s pressure to innovate responsibly with AI.

Remember: any automation that broke at 2 a.m. can haunt you for months. Build with trust and visibility first, then pursue full autonomy later.