From Hype to Help: Evaluating and Applying AI Where It Matters in CX

The Reality Check

The numbers tell a sobering story about AI implementation in customer experience:

• 81% of service organizations are using AI
• Less than 30% have scaled AI to meaningful impact
• 39% struggle with identifying the right use cases
• 85% of pilots are underdelivering on expected outcomes

Why? AI initiatives fail without purpose. When you reach for vague goals, you realize misaligned outcomes.

"Far too often we're relying on vendors or their sellers to tell us where we need to apply the tools. Those are typically their product strengths—they're not necessarily aligned with the needs of your organization."

The Promise: When AI Is Done Right

When you focus on needs rather than features, AI can deliver transformative results:

• CSAT rise: 30-50%
• Average handle time reduction: 20-60%
• Agent capacity growth: Significant increases

These improvements don't just reduce costs—they improve customer retention, increase revenue, and create more rewarding agent jobs that reduce turnover.

The Framework: Plan → Select → Deploy → Evolve

Phase 1: Plan (Outcomes First)

Stop leading with features and capabilities. That's the "fire then aim" philosophy—and it's wrong.

Start with outcomes, then identify use cases that support those business goals clearly.

The Four Use Case Categories

1. Agent Assistance

• Automated wrap-up and call summaries
• Real-time agent assistance
• Automatic CRM/ITSM data insertion
• Goal: Reduce time agents spend on non-value work

2. Performance Analytics

• Predictive analytics for trend identification
• Automated quality management
• Performance coaching opportunities
• Goal: Enable and empower agents, not police them

3. Automation

• FAQ handling (lowest barrier to entry)
• Scheduling and payments
• Self-service empowerment
• Goal: Help customers help themselves

4. CX Optimization

• Personalized service experiences
• Proactive customer engagement
• Reaching out before customers need to ask
• Goal: Anticipate and exceed expectations

How to Spot Good Use Cases

Look for:

• Low risk, high containment: Safe opportunities with big returns
• High volume, high frequency: 40-80% of your total interactions
• Agent productivity boosters: Reduce cognitive load
• Measurable customer impact: CSAT, wait times, loyalty
• Low cognitive load, high time drain: Mind-numbingly simple but repetitive tasks
• Current performance gaps: Where quality information matters
• Scalable with data: Lots of history to train on

The Data Myth

"Garbage in, garbage out frustrates me because when you look at conversational AI, often it's not that much data. Usually you're looking at two to three pieces of information per call."

Focus on conversations rather than massive data models. The garbage in/garbage out mentality doesn't apply to conversational AI the way people think.

Phase 2: Select (Match to Your Needs)

Insider tip: They're all the same.

All conversational AI solutions are built on the same platforms—Open AI, the same models under the hood.

"When they try to tell you they're better or different or using proprietary models, alarm bells should go off. That means they're the same as the rest."

What actually matters:

1. How You Pay

• Per interaction
• Tiered bundles
• Total usage
• Platform licensing/named users

This affects how you scale and forecast costs. Evaluate against YOUR needs.

2. Development & Integration Flexibility

Approach	Best For
Full DevOps	Strong internal engineering teams
Co-build	Most organizations
Low-code/No-code	Teams needing simplicity
Fully managed	Hands-off approach

3. Integration Capabilities

Order of preference:

1. Out-of-box integration
2. Ecosystem partner with off-the-shelf connector
3. Custom development (last resort)

Phase 3: Deploy (High-Impact Pilots)

The 67% Skills Gap: CX leaders report lack of in-house skills as the primary challenge.

Build cross-functional teams with:

• Operations
• IT
• CX owners
• Executive sponsor

The Pilot Formula

1. One well-defined use case

Example: A telco used post-call summary ONLY for escalated calls. Result: 38% reduction in agent wrap-up time in three weeks.

2. Measurable success metrics

Focus on:

• Reducing average handle time
• Saving human effort
• Improving CSAT
• Revenue increase (if applicable)

3. Time-box it: 4-6 weeks

Pilots demonstrating ROI in 6-8 weeks are 5x more likely to receive scaling investment.

Benefits of time-boxing:

• Forces scope limitation
• Clarifies quantifiable outcomes
• Reduces friction in getting started
• People often just leave it on because it's better than before

4. Don't build in a vacuum

Engage users, consumers, even customers. Get feedback. Build champions.

Real Example

A healthcare insurer launched a bot for just ONE use case: ID card reissuance (12% of inquiries). They deflected 70% of those calls in the first month—8.5% of ALL calls contained by one simple use case.

Phase 4: Evolve (Scale What Works)

You're not just proving AI works—you're proving your organization is ready to operationalize it.

Scaling principles:

1. Create modular playbooks: Aim for 80/20 rule—80% framework reuse, 20% customization
2. Don't scale everything: Focus on ROI
3. Standardize success metrics: Efficiency AND experience (customer + agent)
4. Build internal champions: Especially at executive level
5. Tackle integration early: The #1 barrier
6. Upskill continuously: Most important of all
7. Refine your operating model: Become clearer on who does what
8. Invest in feedback loops: Continuous improvement engineering

The Three Skill Categories You Need

Technical Skills (Easiest to acquire)

• ML engineering
• Prompt engineering
• Integration/developer capabilities
• Training: Free, on-demand, in chunks

CX Operations (Learning by exposure)

• Use case scoping
• Translating business pains to AI solutions
• Interpreting AI outputs for business impact
• Training: Webinars, reading, experience

Change Management (Most critical)

• Agent training
• Building organizational trust
• Ethical awareness and governance
• Training: Consider outside consultants or dedicated hires

The Challenge

1. Use the AI readiness checklist (not to score 5 on everything, but to be honest about where you stand)
2. Identify up to three pain points, then scrap two
3. Run ONE pilot with clear KPIs
4. Measure the value
5. Evangelize loudly—don't be humble

"Get the loudest microphone platform you can to get the message out there. People are gonna start wanting to jump on the bandwagon."

---

This session was part of the AI for CX Virtual Summit, presented by AmplifAI.