The Future of AI-Enabled Service:

Executive Summary

Helen, a PhD researcher at the Cambridge Service Alliance, presents research on how generative AI is transforming service design—a human-centered, systemic approach to shaping service experiences and delivery.

Key Context

Service design traditionally follows four stages—discover, define, develop, deliver—and focuses on orchestrating interactions between users and organizations. While generative AI has significantly boosted productivity, most implementations are failing to deliver value because most pilots fail, show no measurable outcomes, and do not achieve meaningful returns

The core issue is not the technology itself, but how organizations design workflows around it. Many firms simply embed AI into existing processes rather than rethinking services end-to-end.

Role of Generative AI in Service Design

GenAI is being applied across all design stages:

• Discover: Data summarization, insight extraction
• Define: Persona generation, problem framing
• Develop: Ideation, prototyping
• Deliver: Content creation, scenario simulation

Compared to traditional methods, AI enables:

• Large-scale data analysis and pattern detection
• Faster synthesis of complex inputs
• Rapid prototyping and early testing
• Scalable, data-driven validation

Research Insights

Based on interviews with approximately 40 practitioners across industries, the research identifies key design principles for effective AI integration. Two highlighted principles:

1. Ground AI in Real-World Contexts

AI performance often breaks down in real usage conditions.

• Example: A hotel chatbot failed to distinguish between similar user intents
• Insight: Testing must involve real users, real data, and real scenarios, not just simulated inputs 

2. Use AI as a Collaborative Partner

Organizations gain more value when treating AI as a co-creator rather than a tool:

• Enhances ideation using real customer data
• Accelerates concept development
• Supports technical feasibility testing without delays

Strategic Implications

Effective AI-enabled service design requires two complementary approaches, these approaches were adapted from Kimbell, 2011. 

• Design for services: Establishing organizational structures, governance, and strategy to support AI adoption
• Design of services: Building and refining actual AI-enabled service experiences

Key Takeaway

Success with generative AI in service design depends less on the technology itself and more on rethinking workflows, grounding solutions in reality, and leveraging AI as a collaborative partner.

Link to Presentation Below

Executive Summary

Robert Christmas from AstraZeneca outlined how the company is applying generative AI in enterprise service and process design, building on research from the Cambridge Service Alliance. The focus is on scaling AI across a large, complex organization undergoing transformation.

Key Context

AstraZeneca is executing a major transformation program (“Axial”) to redesign ~1,500 business processes across global operations (manufacturing, finance, and commercial functions).

Key objectives:

• Drive standardization, scalability, and selective differentiation
• Support long-term growth and operational scale
• Enable transformation through a ~$2B, multi-year initiative

Generative AI has been actively integrated into this program over the past year.

Traditional vs. AI-Enabled Design Approach

Historically, service and process design relied on:

• Subject matter experts (SMEs) across functions and geographies
• Best-practice (“fit-to-standard”) frameworks
• Heavy use of external consultants
• Extensive governance, documentation, and compliance requirements

With AI, AstraZeneca is shifting toward:

• Treating AI as a “digital co-worker” that augments SMEs
• Accelerating design processes rather than fully automating them
• Exploring future potential for greater autonomy

Key Challenges Identified

Three foundational barriers to effective AI adoption:

1. Data Quality
   • AI effectiveness depends on structured, high-quality, enterprise-specific data
   • Significant investment required in data models and semantic layers
2. Adoption & Change Management
   • Productivity gains depend on user trust and usability
   • Resistance arises when outputs require heavy correction
3. Trust & Governance
   • Questions around AI decision-making, accountability, and compliance
   • Particularly critical in regulated industries like pharmaceuticals

Practical Applications & Learnings

1. AI for Quality Assurance at Scale (High Success)

• AI evaluated 5,000+ user stories (design requirements)
• Identified inconsistencies, duplication, and gaps
• Provided structured improvement recommendations

Impact:

• Significant efficiency gains vs. manual review
• Improved consistency and compliance
• Successful large-scale deployment

2. AI for Complex Documentation Generation (Mixed Results)

• AI generated outputs such as SOPs, test scripts, and training materials
• Combined multiple data sources (process flows, standards, etc.)

Findings:

• Outputs were well-structured but prone to:
  • Errors and hallucinations
  • Misinterpretation of complex processes
• Required significant human correction initially

Outcome:

• Iterative improvement over time
• Highlighted importance of high-quality input data and feedback loops

3. AI in End-to-End Design (Experimental)

• Leadership exercises used AI to rapidly generate product concepts

Insights:

• AI dramatically accelerates ideation
• However, it changes human behaviour:
  • Reduced sense of ownership over outputs
  • Tendency to over-trust AI results without sufficient critique

Key Strategic Insights

• AI is most effective as an augmentation tool, not a replacement (today)
• Scaling AI across the enterprise is far harder than running pilots
• Data readiness is the primary constraint, not technology
• Iterative adoption is necessary—early outputs may be poor but improve with feedback
• Human-AI collaboration requires new mindsets, especially around ownership and critical evaluation

Organisational Implications

To successfully embed generative AI in service design, organizations must:

• Invest heavily in data foundations and governance
• Redesign workflows, not just insert AI into existing ones
• Build AI literacy and trust across teams
• Establish clear accountability frameworks
• Collaborate externally (e.g., academia, startups) for cutting-edge capabilities

Key Takeaway

AstraZeneca’s experience reinforces that enterprise-scale AI transformation is primarily an organisational and data challenge, not a technical one. Real value emerges when AI is integrated thoughtfully into workflows, supported by strong data foundations, and used as a collaborative partner rather than a standalone solution.

“These insights align closely with findings from ongoing research with both AstraZeneca and the CSA into how organizations are integrating Generative AI into service design. We  found that data quality, cross-functional alignment, and governance are consistently the conditions that determine whether GenAI delivers meaningful value in service contexts. Our research further strengthens these findings by confirming that the challenges and opportunities highlighted within AstraZeneca are not isolated, but reflect broader patterns emerging across the field.”

Link to Presentation Below

Executive Summary

Vinod Tete from HCLTech presented insights on how to design and implement generative AI projects successfully, shifting focus from using AI in service design to designing AI deployments themselves. The session draws on joint research with the Cambridge Service Alliance, analysing patterns from real-world enterprise implementations.

Key Context: Why AI Projects Fail

Despite strong initial enthusiasm, many generative AI initiatives have struggled:

• Up to 95% of pilots and proofs of concept fail
• Growing scepticism emerged around 2025 regarding AI’s real-world value

Key reasons for failure:

• Ad hoc implementation across disconnected business units
• Lack of organizational readiness and governance
• Weak or unclear business cases
• High costs (e.g., token usage, specialized talent)
• Poor data quality and fragmentation (“junk in, junk out”)

Research Approach

The findings are based on:

• 40 large enterprises across industries
• Quantitative surveys + 12 weeks of qualitative interviews
• Input from senior leaders (CIOs, AI heads, strategy leaders)

The goal: identify common success patterns in scalable AI deployments.

Real-World Use Cases of Successful AI Deployment

A. Industrial Manufacturing (Truck Maintenance)

• AI integrated data from multiple systems (service history, manuals, parts, etc.)
• Generated field repair instructions for engineers

Impact:

• Reduced repair time from ~7 days to ~3 days
• Improved first-time fix rates and operational efficiency

B. Maritime Contract Analysis

• AI analysed thousands of complex contracts
• Identified loopholes and risks

Impact:

• Reduced penalties and financial losses
• Improved future contract design

C. Media & Broadcasting

• Unified fragmented subscriber data across 100+ platforms
• Enabled data monetization and targeted advertising

Impact:

• Created new revenue streams from previously underutilized data

D. Entertainment & IP Protection

• AI detected unauthorized use of intellectual property
• Automated legal enforcement actions

Impact:

• Increased royalty revenue
• Reduced IP misuse at scale

Key Success Factors for AI Implementation

1. Structured, Centralized Approach

• Successful organizations used a central AI foundation team
• Business units built use cases on top of shared infrastructure
• Avoided fragmented, siloed experimentation

2. Strong Data Foundation

• Clean, integrated, and accessible data is critical
• Data readiness directly determines AI effectiveness

3. Clear Business Value

• Early focus on cost savings, but shift toward revenue-generating use cases
• Successful projects tied directly to measurable outcomes

4. Scalable Design from the Start

• Avoid isolated pilots with no path to scale
• Design systems and architecture for enterprise-wide adoption

5. Standardized Technology Stack

Emerging Trends

• Significant shift from pilot/POC stage to production deployments (~43%)
• Increasing maturity in how organizations structure AI initiatives
• Growing emphasis on designing AI systems holistically, not just experimenting

Key Takeaway

Successful generative AI adoption depends less on isolated innovation and more on structured design, strong data foundations, and enterprise-wide scalability. Organizations that treat AI as a strategic capability, rather than a series of experiments, are the ones realizing tangible value.

Link to Presentation Below

Executive Summary

Matthäus, a collaborator with the Cambridge Service Alliance, presents research on how conversational AI can be effectively integrated into physical environments e.g., retail, healthcare, workplaces), moving beyond traditional one-to-one digital interactions in private settings.

Key Context

A case study in a public retail setting shows that AI agents deliver the most value not as a simple information tool, but as a “sparing partner”—helping users explore options, challenge assumptions, and make better decisions. This leads to deeper engagement, an enhanced service experience and improved outcomes (e.g., personlized service, more informed purchases or better tailored advice).

To succeed, organizations must rethink design for physical contexts. Key requirements include:

• Clear visibility and value communication to encourage adoption
• Context-aware responses leveraging local data (e.g., inventory, location)
• Design for social interactions, as multiple users often engage simultaneously
• Seamless hand-offs between digital advice and physical action (or human staff)
• Continuous learning loops using interaction and contextual data

When implemented well, AI sparing partners

• Act as a scalable complement to human staff, especially during peak demand
• Enable judgment-free interactions, increasing user comfort and engagement
• Unlock new data from physical interactions, improving service design and personalization
• Create a defensible competitive advantage through context-specific, continuously improving systems

Also meeting rooms and workspaces are physical spaces: Hence, beyond customer use, AI sparing partners also support employees as advisors, compliance assistants, and change facilitators.

Key Takeaway

Embedding AI into physical spaces transforms it from a passive tool into an interactive, context-aware collaborator, but success depends on thoughtful design that accounts for human behaviour, environment, and real-world workflows.

Link to Presentation Below