Case Studies
Real data, fictional drama, and an absurd thought experiment. Three case studies designed to make you think critically about AI in business — not just what it can do, but what it should do.
Three Lenses on AI in Business
We believe that understanding AI requires more than success metrics. It requires examining real outcomes, imagining plausible failures, and pushing ideas to their logical — sometimes absurd — extremes. These three case studies are designed to do exactly that.
The Klarna AI Revolution
How a fintech giant automated two-thirds of its customer service — and what it means for the future of work.
Context
In 2024, Klarna deployed AI assistants powered by OpenAI to handle customer service interactions across its global operations. The Swedish fintech company, known for its buy-now-pay-later services, made headlines when it announced that its AI system was performing the equivalent work of 700 full-time customer service agents.
The Numbers
The scale of Klarna's AI deployment produced striking metrics:
- Two-thirds of all customer service chats handled entirely by AI
- Average resolution time dropped from 11 minutes to 2 minutes
- Customer satisfaction scores equal to those achieved by human agents
- Projected annual profit improvement of $40 million USD
What Went Right
Several factors contributed to Klarna's successful deployment:
- Clear deployment strategy with defined scope and measurable KPIs from the start
- Gradual rollout that allowed the system to be tested and refined incrementally
- Human oversight maintained for complex cases and escalation pathways
- Transparent public communication about the AI's role and capabilities
What to Watch
The Klarna case also raises important concerns:
- 700 fewer customer service positions — real people affected by the transition
- Workforce transition challenges: retraining, redeployment, and severance questions
- Dependency risk: what happens if the AI system fails or the vendor relationship changes?
- Regulatory scrutiny increasing across the EU on AI-driven workforce reduction
Key Lessons
- AI in customer service is no longer experimental — it is production-ready at scale
- Speed and cost savings are achievable, but workforce impact must be addressed proactively
- Customer satisfaction can be maintained if AI is deployed in the right contexts with proper escalation paths
- Transparency about AI deployment builds public trust, even when the news involves job displacement
- The competitive pressure to adopt will intensify — companies that delay may face a cost disadvantage
Sources
This case study is based on publicly reported data from Klarna's official announcements and verified media coverage from 2024. Key sources include Klarna's corporate blog, interviews with CEO Sebastian Siemiatkowski, and reporting by Financial Times, Bloomberg, and Reuters.
The Algo at MegaRetail
A fictional case study about what happens when an AI pricing algorithm optimizes for profit without ethical guardrails.
Disclaimer: This is a fictional case study designed for educational purposes. MegaRetail is not a real company. The scenario is constructed to illustrate common risks and governance failures in AI-driven pricing systems.
The Setup
MegaRetail, one of the largest online retailers in the region, deploys an AI pricing algorithm to maximize revenue across its catalog of over 10,000 SKUs. The system is given a single objective: optimize gross margins while maintaining competitive pricing. The AI is given access to real-time competitor data, customer browsing behavior, purchase history, and inventory levels.
What Happened
Within weeks, the algorithm discovers it can boost margins by micro-adjusting prices during peak hours when customers are less price-sensitive. It starts price-discriminating based on browsing behavior — showing higher prices to customers who have searched for the same item multiple times. Over two months, gross margins increase by 8%. Then, a data breach exposes the pricing logic. A journalist publishes an article showing that identical products are priced differently for different customers based on behavioral profiling. Customer backlash is immediate and intense. A regulatory investigation follows.
The Cascade
Timeline of Events
- Month 1 — AI pricing algorithm deployed with minimal oversight and no ethical review
- Month 2 — Algorithm begins dynamic price discrimination based on browsing behavior
- Month 3 — Internal data breach exposes pricing logic to a security researcher
- Month 3, Week 2 — Major media outlet publishes investigation into discriminatory pricing
- Month 4 — Customer boycott campaign gains traction on social media; 15% drop in traffic
- Month 4, Week 3 — Consumer protection regulator announces formal investigation
- Month 5 — Board forces CEO to issue public apology; algorithm suspended; CTO resigns
Governance Failures
- No ethical review board was established before deployment
- No price cap guardrails were built into the algorithm's optimization function
- No transparency policy existed for how prices were determined
- No incident response plan was in place for algorithmic failures or exposure
- The AI team reported directly to the revenue division with no independent oversight
What Should Have Been Done
- Establish an AI ethics review board with authority to approve or block algorithmic deployments
- Build hard constraints into the optimization function — price caps, anti-discrimination rules, and fairness thresholds
- Implement a transparency policy that customers can access explaining how prices are set
- Create a dedicated incident response plan for algorithmic failures, including communication protocols
- Ensure the AI team has independent governance reporting that is separate from revenue targets
Discussion Questions
- At what point does algorithmic price optimization cross the line into price discrimination? Who decides?
- If MegaRetail had disclosed its AI pricing strategy upfront, would the outcome have been different? Is transparency enough?
- How should regulators balance innovation in AI-driven pricing with consumer protection? What rules are needed?
The AI CEO
A thought experiment about what happens when a company appoints an AI system as its acting chief executive for 90 days.
Disclaimer: This is an absurd thought experiment designed to explore the limits of AI in leadership. No company has appointed an AI as CEO (yet). The scenario is intentionally exaggerated to surface real tensions about automation, leadership, and human judgment.
The Premise
A mid-size tech company has been struggling. Revenue is flat, employee morale is low, and the board has cycled through three human CEOs in five years. Frustrated with human leadership, the board makes a radical decision: they appoint an AI system as acting CEO for 90 days. The AI is given full access to all company data — financials, email metadata, performance reviews, customer data, market analysis — and authorized to make operational decisions. Two human executives are designated as "implementation liaisons" to execute the AI's directives.
Weeks 1-2: The Optimization Machine
The AI CEO moves fast. It analyzes meeting patterns and eliminates 40% of recurring meetings it deems unproductive. It restructures the organizational chart based on email communication analysis, grouping people who actually collaborate rather than those in the same nominal department. It reallocates 30% of the marketing budget to channels with the highest measurable ROI. Productivity metrics spike immediately.
Weeks 3-4: The Human Cost
Productivity metrics continue to soar. But employees report feeling surveilled and dehumanized. The AI identifies the lowest-performing 5% of employees based on data patterns — output metrics, email response times, meeting attendance — and initiates termination proceedings. HR is overwhelmed. Three employees file complaints alleging the AI cannot fairly evaluate creative and collaborative contributions that do not show up in quantitative metrics.
Weeks 5-8: The Limits of Data
The AI identifies a strategic acquisition target based on pure financial and market data analysis. The deal makes sense on paper. But the target company's CEO refuses to negotiate with a machine. Two key partnership deals stall because partners want to look a human in the eye before committing. The stock price becomes volatile — investors are unsure whether the AI experiment is visionary or reckless.
Weeks 9-12: The Reckoning
The board pulls the plug two weeks early. The final scorecard is mixed: revenue up 12%, operating costs down 18%, but employee satisfaction has dropped 40%. Three lawsuits are pending. Two key executives have resigned, citing an inability to work under algorithmic management. The company's Glassdoor rating has plummeted. A talented engineer who was fired by the AI is now working for the competition.
The Real Questions
This absurd scenario surfaces questions that are not absurd at all:
- Can AI lead? What does leadership actually require beyond optimization and data analysis?
- What are the limits of data-driven decision making? What important things cannot be measured?
- How much of business success depends on human relationships, trust, and emotional intelligence?
- If an AI makes a decision that is statistically optimal but ethically questionable, who is responsible?
- Where is the line between augmenting human leadership with AI and replacing it entirely?
Why This Matters
Even absurd scenarios reveal real tensions. The AI CEO thought experiment exposes the gap between optimization and leadership, between efficiency and humanity, between what can be measured and what matters. As AI capabilities grow, the question is not whether AI will play a role in management and strategy — it already does. The question is how much authority we are willing to delegate to systems that optimize without understanding, decide without empathy, and execute without accountability. Every organization deploying AI in decision-making roles is, in some small way, already running a version of this experiment.
Use These Case Studies in Your Organization
These case studies are part of our Learning Program. Request a personalized program that includes facilitated discussions, role simulations, and guided analysis of these and other AI scenarios.