Why 74% of Enterprise CX AI Programs Fail — And How to Make Them Work

By
Ricardo Saltz Gulko
-

Share on LinkedIn Share on LinkedIn

The original article was initially posted on Eglobalis.com here for you.

Introduction

Don’t rush into AI without strategy or data start by an evaluation. In the race to adopt AI for customer experience (CX), many enterprises are charging ahead without laying a proper foundation. As of 2023, roughly 79% of organizations report using some form of AI in their CX toolset. Yet enthusiasm often outpaces readiness: global surveys show only about 13% of companies are fully prepared to capture AI’s potential in CX. The result? A startling 85% of AI projects fail, largely due to poor data quality. Without clean, unified customer data and a solid strategy, AI simply amplifies existing flaws instead of solving them. Executives are learning that AI is no magic fix – it won’t rescue a broken CX or disjointed processes on its own. In fact, the majority of companies have yet to see tangible value from their AI investments: by late 2024, only 26% had moved beyond pilots to achieve real gains, while 74% struggled to generate any measurable value.

The following sections examine real-world cases – both failures and partial successes – of enterprise AI adoption in B2B customer experience across fintech, call centers, CRM/CDP integration and customer-facing tools. Each illustrates the expected benefits versus actual outcomes, with quantitative CX impacts where available. We’ll see high-profile missteps where rushed automation backfired, alongside success stories where a strategic, data-driven approach paid off. Crucially, each case concludes with three actionable recommendations to help businesses do better. The key lesson: Don’t leap on the AI bandwagon without first building a strong strategy and data foundation. AI can transform customer experience, but only if deployed thoughtfully – with clean data, clear objectives, and humans in the loop. Proceed with caution and purpose, or risk expensive setbacks on the road to CX excellence.

20 practical recommendations To avoid failure:

Before even starting this article, here are 20 practical recommendations that can dramatically increase your chances of success when adopting AI or Agentic AI in B2B and B2C environments. These suggestions are tailored for executive decision-makers and delivery leaders seeking sustainable, customer-focused outcomes.20 Practical AI Success Guidelines for B2B Leaders based in my work with Samsung, SAP and some other groups:

  1. Define Strategic Intent Early – Clarify the business problem AI is solving. Tie it to specific CX outcomes like retention, onboarding speed, or SLA adherence.
  2. Treat Data as Infrastructure – AI is only as strong as the data behind it. Build unified, real-time access to clean customer data before you touch a model.
  3. Avoid Proof-of-Concept Paralysis – Don’t stop at pilots. Plan for full rollout paths at the start, including change management, training, and KPIs.
  4. Map to B2B Lifecycle Touchpoints – Apply AI to key value stages: onboarding, renewals, escalation, and usage support—not just front-line chatbots.
  5. Use Agentic AI for Role-Specific Augmentation – Focus on Agent AI that supports sellers, CS managers, and delivery leaders—context-aware copilots outperform generic automation.
  6. Assign an AI Product Owner – Treat every AI use case like a product: with accountability, iterations, feedback loops, and embedded business sponsors.
  7. Start with Measurable Wins – Prioritize use cases with clear metrics (e.g., case deflection %, onboarding time, customer lifetime value). Build credibility before scaling.
  8. Include Legal and Risk Early – In regulated B2B industries, AI can’t succeed unless compliance is engaged from design—not just at deployment.
  9. Design Human-AI Handoff Paths – Never trap users in AI-only loops. High-value clients should always have an escape hatch to a skilled human when needed.
  10. Train Your People, Not Just Your Models – Employees must understand how AI decisions are made, when to trust it, and how to correct it.
  11. Deploy AI That Explains Itself – Choose or build models with traceable logic. B2B clients need justification for recommendations, forecasts, and risk scoring.
  12. Reinforce Customer Empathy in Design – Agentic AI shouldn’t just be functional—it must reflect the tone, understanding, and patience of trusted partners.
  13. Avoid Premature Over-Automation – Many failures stem from assuming AI can fully replace humans. Begin with augmentation and graduate based on proven outcomes.
  14. Monitor CX Metrics Closely – Use qualitative and quantitative CX data (like CES, CSAT, adoption rate, resolution rate) to evaluate if AI is actually helping customers.
  15. Create an AI Governance Council – Ensure cross-functional oversight of ethical use, bias mitigation, and ongoing model performance.
  16. Test or Experiment Under Real Load – Simulate real-world complexity, volume, and escalation scenarios before rollout. B2B CX is nonlinear—AI must be resilient.
  17. Engage Customers in AI Feedback Loops – Let B2B clients flag when AI responses fail. Reward that feedback with faster support or transparency.
  18. Align AI to Account-Level Strategy – Tailor AI outputs to client tiers, segmentation, and relationship stage—enterprise customers expect customization.
  19. Scale Only After Success Signals – Set defined criteria (e.g., 95%+ resolution on Tier 1 cases, or 20% time savings in renewals) before scaling to other regions.
  20. Don’t Confuse Trend with Readiness – Just because the market says “do AI now” doesn’t mean your company is prepared. Readiness beats urgency.

1. Fintech’s Automation Pitfall – Klarna’s Cost-Cutting Backfire

Klarna, the Swedish fintech giant, sought to cut costs and streamline customer service by replacing nearly 700 employees with a generative AI chatbot. The expected outcome was increased efficiency and round-the-clock availability. For a brief period, Klarna claimed the chatbot handled two-thirds of all support inquiries. However, this automation-first strategy quickly backfired.

Customer satisfaction dropped sharply due to the bot’s inability to resolve complex or sensitive issues—like fraud claims, payment disputes, or delivery errors. Complaints surged, and users reported feeling frustrated and dehumanized. Klarna’s leadership acknowledged that the company had gone too far, too fast. By mid-2024, the firm began rehiring human support agents and shifted to a hybrid AI-human model.

Three Practical Recommendations:

  1. Don’t fully replace humans—augment them. Use AI to handle repetitive Tier 1 questions but retain skilled humans for nuanced or emotional customer needs.
  2. Pilot AI with guardrails and fallback paths. Always provide customers a way to escalate to a human when the AI struggles.
  3. Measure customer effort and sentiment. Track not just resolution rates, but how customers feel after interacting with AI. Satisfaction drops may signal where automation hurts CX.

2. Data-Driven Personalization Win – NAB’s “Customer Brain” Platform

National Australia Bank (NAB), a major financial institution, implemented a customer intelligence platform known as “Customer Brain” that analyzes more than 2,000 behavioral and transactional data points to create highly personalized service interactions.

The platform leverages over 800 AI models to determine optimal engagement strategies in real-time. Customers are routed to the most relevant offers, services, or human agents depending on their needs, improving the bank’s ability to cross-sell and resolve issues.

The result? A 40% increase in digital engagement and a 20% drop in follow-up service requests—driven by more accurate first-interaction resolution. Customers receive proactive notifications and personalized experiences based on their history, lifecycle stage, and financial behaviour.

Three Practical Recommendations:

  1. Invest in CDPs and data integration. A high-functioning personalization engine depends on structured, unified customer data—build that first.
  2. Deploy AI with measurable KPIs. Don’t just monitor engagement—tie personalization efforts to reduced churn, increased CLV, and fewer support tickets.
  3. Automate intelligently, escalate strategically. Use AI to push personalized options, but route uncertain or sensitive cases to skilled agents for best-in-class CX.

3. Customer-Facing AI Misstep – Air Canada’s Chatbot Legal Snafu

Air Canada implemented a generative AI chatbot to assist customers with common queries and reduce support center pressure. The bot was intended to enhance self-service and streamline communication. However, in late 2023, the chatbot misinformed a customer about the airline’s bereavement fare policy, falsely stating they could claim a partial refund after travel.

When the passenger followed this advice and requested reimbursement, Air Canada denied the request. The issue escalated legally, and the tribunal ruled that Air Canada was responsible for the misinformation given by its chatbot. The company was ordered to honor the refund, and its legal argument—that the AI was a separate entity—was rejected.

This case revealed serious flaws in governance and training of AI systems. The bot was eventually pulled down to avoid further reputational and financial damage. Customers reported trust erosion, and analysts warned other enterprises about legal risks tied to unsupervised AI in customer-facing roles.

Three Practical Recommendations:

  1. Audit AI training data and logic. Ensure AI bots only reference and quote vetted, official policy documents.
  2. Add legal disclaimers and human fallback. Clearly notify users of AI limitations and allow seamless escalation to a real agent.
  3. Monitor and test frequently. Treat chatbots as evolving tools—regularly test them with edge cases to prevent misinformation disasters.

4. Augmenting Call Centers – Telstra’s AI-Enhanced Support Success

Telstra, Australia’s largest telecom provider, introduced AI assistant tools within its contact centers to enhance—not replace—its human workforce. Two generative AI systems were implemented: one to summarize customer history in real time and another to retrieve accurate answers from a secure internal knowledge base.

The result was a sharp increase in efficiency. Over 90% of customer service agents reported improved speed and accuracy, while first contact resolution improved by 20%. Unlike some peers, Telstra didn’t lay off support staff—instead, they trained them to collaborate with AI, empowering agents and boosting morale.

This hybrid AI-human model allowed Telstra to reduce repeat calls, shorten handle times, and provide better customer outcomes. The AI acted as a co-pilot, not a replacement—agents could stay focused on listening and resolving, not toggling screens or searching for info.

Three Practical Recommendations:

  1. Deploy AI as a real-time assistant. Focus on tools that help agents during calls—summaries, suggestions, and instant answers improve both speed and quality.
  2. Train and involve your workforce. Involve agents in AI rollout to encourage adoption. Make AI feel like a tool to help, not a threat.
  3. Measure first-contact resolution. Use this key CX metric to determine if AI tools are improving the overall experience—not just cutting costs.

5. Pushing Digital-Only Support – Frontier Airlines’ Botched Experiment

Frontier Airlines attempted to reduce costs by removing all customer service phone numbers and pushing passengers toward a chatbot and text-based channels. While the strategy promised scalability and 24/7 availability, its execution fell flat.

The chatbot frequently failed to resolve common travel concerns such as cancellations, refunds, or rebookings. Customers were met with generic answers and no escalation option, sparking outrage and social media backlash. Without voice support or live agent handoff, Frontier left customers stranded during critical travel moments.

This cost-cutting decision damaged the airline’s reputation, with media dubbing the move an abandonment of basic customer care. Eventually, Frontier was forced to enhance human-assisted chat support, but trust had already eroded.

Three Practical Recommendations:

  1. Don’t eliminate human support options. Offer digital channels, but always provide a fallback to real agents, especially in high-stakes industries like travel.
  2. Test bot effectiveness before scale. Run small pilots to ensure AI can handle real scenarios with accuracy and empathy before removing other channels.
  3. Listen to customer sentiment. Use VOC data to catch rising frustration early—high abandonment or repeat contact rates are red flags your automation isn’t working.

6. Self-Service at Scale – Zoom’s AI-Powered Support Deflection

Zoom implemented AI-powered virtual agents and knowledge base enhancements to handle the overwhelming surge of support queries during its rapid global expansion. The expected benefit was increased self-service efficiency and reduced dependency on live human agents.

After deploying AI systems, Zoom saw its self-service rate increase significantly. The AI tools could resolve a growing number of inquiries without human intervention, particularly for common account, access, and troubleshooting issues. This allowed human agents to focus on high-priority, complex cases, and improved the speed and accuracy of responses overall.

A third-party study found that support agent productivity increased by 14% following the rollout of generative AI tools. Additionally, Zoom’s AI was able to deflect a higher proportion of tickets and provided faster answers to users through real-time recommendations based on customer intent.

Three Practical Recommendations:

  1. Track deflection vs. resolution rates. Ensure that AI isn’t just deflecting tickets—it should actually resolve them, reducing repeat contacts.
  2. Optimize your knowledge base. AI only performs as well as the content it can reference. Keep documentation clean, current, and search-friendly.
  3. Segment your user base. Use AI to personalize content for different customer types (e.g., enterprise vs. individual users) for greater precision and relevance.

7. AI Chatbot Redeployment – IKEA Frees Up 8,500 Staff for Higher-Value Service

IKEA deployed an AI chatbot named “Billie” to handle repetitive customer questions like delivery status, product availability, and assembly support. As digital traffic increased, Billie became a central self-service tool on IKEA’s website and app.

The impact was substantial. Billie successfully deflected a significant portion of Tier 1 inquiries, improving resolution speed for common requests. As a result, IKEA was able to redeploy over 8,500 support employees into value-added roles such as personalized interior design advisors and consultative sales agents.

This strategic use of AI not only improved operational efficiency but also elevated the customer experience. Customers received instant answers for simple tasks and richer human-led service for complex needs.

Three Practical Recommendations:

  1. Use AI to elevate—not eliminate—human work. Automate routine tasks to free staff for strategic, experience-enhancing roles.
  2. Train AI on your most frequent questions. Analyze historical chat data to determine where automation adds the most value.
  3. Integrate AI across platforms. Make sure your chatbot works consistently on web, mobile, and in-app interfaces to maximize reach and adoption.

8. CRM Integration for Peak Demand – Wiley’s 40% Efficiency Gain

Wiley, the global publishing company, faced recurring customer support spikes during academic enrollment seasons. To manage demand, Wiley integrated AI agents into its Salesforce Service Cloud CRM to resolve common technical and account inquiries without human involvement.

The result was a 40% increase in support efficiency. Routine tasks—like password resets, subscription renewals, or access to digital content—were successfully handled by the AI agent. This helped Wiley flatten seasonal volume spikes and reduce response times, improving satisfaction for students and institutions alike.

AI responses were embedded directly into CRM workflows, meaning that human agents had full visibility when handoff was required. This seamless transition ensured customers never had to repeat themselves or restart conversations.

Three Practical Recommendations:

  1. Automate known spikes. Use AI to handle recurring seasonal issues and prevent burnout during peak periods.
  2. Integrate with your CRM. Ensure AI and human agents share the same platform and context for seamless customer transitions.
  3. Apply AI where volumes are high and answers are standard. Use AI to focus on high-volume, low-complexity issues and route advanced cases to specialists.

9. Public-Sector AI Fail – NYC’s Business Chatbot Gives Illegal Advice

In 2024, New York City launched the “MyCity” AI chatbot to assist entrepreneurs in navigating local regulations. The tool was supposed to help businesses with permits, labor laws, and compliance. But the AI soon made headlines for giving dangerously incorrect—and sometimes illegal—advice.

Examples included telling users they could fire employees for reporting harassment or keep customer tips for the business, both of which violate city and federal law. The city quickly added disclaimers, but the damage to trust was done.

This public failure showed the risks of deploying generative AI tools trained on uncontrolled or insufficiently vetted data in complex legal environments. It also emphasized the importance of involving subject matter experts in model tuning and oversight.

Three Practical Recommendations:

  1. Train AI on vetted, authoritative sources. Avoid scraping general web content. Use official documentation only.
  2. Launch with disclaimers and oversight. Clearly label AI as informational and allow human validation for regulated decisions.
  3. Start in closed beta. Before public rollout, test with experts and end users to catch risks and hallucinations in real-world contexts.

10. Talent and Culture – The Human Factor Holding AI Back

Many AI deployments fail not because of technology limitations, but because of people. Enterprises frequently underestimate the cultural barriers to adoption, especially when employees fear being replaced or don’t trust the tools.

In several telecom and financial services firms, agents and analysts resisted AI recommendations or underused new tools. Why? Poor communication, lack of training, and no involvement in the design process. One bank’s AI-enabled advisor dashboard was ignored because advisors didn’t understand how suggestions were generated—or how they’d be evaluated if they didn’t follow them.

These problems stall adoption and limit ROI. Conversely, companies that frame AI as augmentation—not automation—and include frontline teams in the process are seeing better outcomes and faster rollouts.

Three Practical Recommendations:

  1. Create a human-first AI narrative. Communicate how AI supports, not replaces, employees—and back it with action (like training or new roles).
  2. Include users in AI design. Co-create tools with the people who’ll use them. Their input leads to better UX and higher adoption.
  3. Track adoption metrics. Monitor usage, satisfaction, and productivity by user segment to quickly address barriers or misunderstandings.

11. The Ethics and Compliance Tightrope

Enterprise AI, especially in B2B contexts, often deals with sensitive data—financials, customer records, HR information. Missteps in handling this data can create serious legal, regulatory, and reputational consequences.

In sectors like healthcare and finance, compliance teams are often blockers not because they resist innovation, but because AI projects don’t meet required privacy or audit standards. One multinational logistics firm had to pause its AI-driven pricing engine after GDPR auditors flagged insufficient transparency in how rates were decided.

Additionally, untested models can introduce bias, discrimination, or misinformation into critical decisions. AI scoring tools used in hiring or loan approval must be explainable and fair, or they risk being challenged by regulators or advocacy groups.

Three Practical Recommendations:

  1. Integrate compliance from the start. Include legal and risk officers early in the AI lifecycle—not just before go-live.
  2. Make AI explainable and auditable. Ensure models can justify outcomes. This is essential in regulated industries.
  3. Develop ethical AI policies. Formalize your stance on bias, consent, and human oversight to avoid project-by-project firefighting.

12. B2B AI Reality Check – Mixed Results and Strategic Lessons

While many AI experiments have focused on B2C environments, B2B companies are also grappling with adoption challenges—and the stakes are often higher due to complex sales cycles, high-value accounts, and mission-critical services.

Take Salesforce, for example. After launching Einstein GPT for its enterprise clients, some users reported that while the features were promising, initial results fell short due to data silos and inconsistencies in implementation across departments. Similarly, Oracle’s B2B clients experienced delays in realizing promised AI benefits in customer success workflows due to gaps between AI predictions and CRM integrations.

Meanwhile, ServiceNow reported strong results using GenAI internally and externally to streamline enterprise service management, but only after significant investment in employee enablement and controlled pilot programs.

These examples highlight why B2B adoption requires not just advanced tools, but end-to-end orchestration: strategic alignment, AI-trained staff, consistent data quality, and agile rollout.

Three Practical Recommendations:

  1. Map AI to revenue-generating B2B workflows. Focus on areas like renewal prediction, onboarding, SLA resolution, and escalation prevention.
  2. Involve customer success and delivery teams. These frontline B2B teams are best positioned to validate and shape AI use cases.
  3. Invest in enablement. Equip your B2B account managers, consultants, and partners with training to co-own AI outcomes with confidence. The early wave of enterprise AI in B2B customer experience has delivered a mixed picture. From Klarna’s all-in chatbot collapse to NAB’s precision-personalization success, the key differentiator has been strategy, not technology. Companies that began with clean, connected data, mapped realistic goals, and preserved human empathy saw measurable gains in loyalty, resolution rates, and operational efficiency.

By contrast, those who rushed in without data readiness, governance, or employee involvement faced expensive failures, legal consequences, or damaged CX. Generative AI has promise—but only if treated not as a shortcut, but as a precision instrument requiring calibration, control, and human complement.

The examples covered across fintech, telecom, travel, public services, and education show a clear pattern: AI works best when deployed with clarity, constraint, and care. It is not a replacement for customer experience strategy—it’s a tool that must be built into it.

Organizations serious about AI must now shift from experimentation to execution. That means:

Those who do will not just avoid failure. They’ll become the models of future-ready, AI-augmented customer experience leadership.

Conclusion

The early wave of enterprise AI in B2B and B2C customer experience has delivered a complex mix of outcomes. Companies like Klarna, NAB, IKEA, Wiley, and ServiceNow have demonstrated that the difference between failure and long-term success isn’t the technology—it’s the clarity of purpose, data quality, internal alignment, and the will to test, adapt, and improve over time.

We’ve seen AI tools dramatically boost personalization, self-service, and operational productivity—but also witnessed real-world disasters from over-automation, legal missteps, poor design, and cultural resistance. These aren’t theoretical issues; they’re lived failures and hard-won lessons that show AI must be rolled out deliberately, transparently, and in sync with customer needs and human teams.

AI is neither hype nor savior—it is a tool. A powerful one. But in the wrong context, or deployed without empathy and data readiness, it can damage the very experience it was meant to improve.

Leaders must move forward with both urgency and restraint. They must measure AI’s success not only by cost savings or resolution speed, but by customer outcomes, team trust, and long-term loyalty. Those who get it right are not just improving CX—they are redefining what it means to serve and connect in the age of intelligent systems.

Data Sources

Share on LinkedIn Share on LinkedIn

Ricardo Saltz Gulko
Eglobalis and European Customer Experience Organization ECXO.ORG
Ricardo Saltz Gulko is the founder of Eglobalis and the European Customer Experience Organization (ECXO). He is a global B2B strategist working with large enterprises on Customer Experience, Professional Services, design-led innovation, and data-driven service models. His work turns customer signals into measurable business outcomes, helping organizations unlock new revenue, strengthen competitiveness, and scale adoption. Eglobalis serves Fortune 100 companies including Samsung, Oracle, SAP, and HP.
View All Posts »

ADD YOUR COMMENT

Please use comments to add value to the discussion. Maximum one link to an educational blog post or article. We will NOT PUBLISH brief comments like "good post," comments that mainly promote links, or comments with links to companies, products, or services.

Please enter your comment!
Please enter your name here