Why Agentic and Conversational AI Products Are Not What CX Leaders Think - with Baker Johnson of UJET

Johnson traces how customer experience became siloed as a support function rather than encompassing the entire customer lifecycle. Thirty years of treating CX as a cost center led to ruthless efficiency metrics, legacy technology stacks, and measuring success by customer deflection rather than outcomes. This mindset prevents meaningful AI transformation.

• •CX should span entire customer lifecycle (sales, marketing, support, loyalty) but has been reduced to contact center operations
• •Three decades of cost-center mentality created focus on deflection metrics instead of business outcomes
• •Organizations apply AI to automate legacy processes rather than reimagine workflows
• •Human agents are blamed for failures when they're actually cleaning up messes created by poorly designed AI systems

Johnson challenges the industry's obsession with creating 'delightful' customer experiences. Customers don't want journeys—they want outcomes. The acquisition of conversational analytics company Spiral enables UJET to design organizational blueprints around actual customer concerns rather than manufactured journey maps. The best interaction is the one that never needs to happen.

• •Customers want their time back, not delightful experiences—'just take my money and give me what I want'
• •Don't speed up bad experiences; redesign them around outcomes, not journeys
• •Conversational analytics at scale reveals what customers actually need versus what companies think they need
• •Focus on designing processes around real customer concerns, not idealized journey maps

Johnson identifies the core reasons enterprises see poor ROI from conversational AI: retrofitting AI into broken processes ('paving the cow paths'), measuring success by deflection rather than outcomes, using AI as a gatekeeper to human agents, and applying customer journey design thinking when customers just want immediate resolution. Early adopters who deploy, learn, and iterate are getting ahead.

• •Enterprises are 'paving the cow paths'—making poorly designed processes 10x more efficient instead of transforming them
• •Measuring AI success by deflection rates sets up human agents to fail with the most complex, frustrated customers
• •AI deployed as a gatekeeper creates adversarial customer experiences rather than helpful ones
• •Trial and error approach with contained mistakes is the path forward for early adopters

Johnson debunks two major misconceptions: that AI will replace all human agents, and that it will never replace any. The real issue is applying AI to the wrong jobs with wrong metrics. The 'out of the box' myth persists—people treat generative AI like search engines. Enterprise deployment requires persona assignment, task definition, context, and most critically, reconciling systems of record with real-time interaction data to give AI agents proper agency.

• •Binary thinking about AI replacement misses the point—it's about assigning the right jobs to AI versus humans
• •Small language models and fine-tuning are replacing generic large language models for enterprise use cases
• •Data governance and knowledge management are prerequisites—AI needs reconciled data from systems of record and real-time interactions
• •Too much latency between systems prevents AI from being accurate, effective, informed, and accountable
• •Consumer AI (agent-to-agent interactions) will require new protocols like MCP for data management

Johnson advocates for exhausting manual processes (pen and paper, spreadsheets) before deploying automation tools. This builds the ability to spot anomalies and understand what's actually happening. The human-in-the-loop approach requires developing intuition through manual work first, then applying technology. Leading with tools instead of thought process is the number one reason enterprises fail with AI.

• •Never invest in automation tools until manual resources are exhausted—builds anomaly detection skills
• •Progression: pen and paper → spreadsheet → system ensures humans learn to spot when outputs look wrong
• •Full dependency on tool outputs without manual experience creates blind trust in potentially flawed models
• •Human nature, process design, and intentionality are the lynchpins for AI success, not the technology itself

Johnson's core advice: the status quo is accumulated debt from years of bad thinking and bad technology. Leaders must start with a clean sheet of paper, embrace AI to 10x themselves, and abandon all sacred cows. Companies achieving $100M revenue with 3-5 people aren't automating enterprise processes—they're using AI as a coworker to co-design and co-model entirely new business models.

• •Status quo wasn't thoughtfully designed—it's technical and process debt accumulated over decades
• •All career experience from the last 20 years risks becoming obsolete; leaders must personally reinvent
• •Start with pen and paper to define ideal processes with zero commitment to past thinking
• •Silicon Valley startups reaching $100M+ with tiny teams use AI as a coworker, not an automation layer
• •The transformation isn't about technology companies sell, but technology on which they operate

Matthew DeMello summarizes three key takeaways: redesign processes before deploying AI (automating legacy workflows accelerates problems), reconcile data across systems of record and real-time channels (without contextual data, models fail), and design intentional workflows that treat AI as collaborator not gatekeeper (zero-sum assumptions miss the real ROI of transformation).

• •Clean sheet process design must precede AI deployment to avoid automating broken workflows
• •Data reconciliation between systems of record and real-time interactions is non-negotiable for AI effectiveness
• •Human-AI balance requires intentional design treating AI as collaborator, not cost-saving replacement
• •Real ROI comes from transformation, not one-to-one human-to-AI replacement ratios