Case studies

Case studies built around real business questions.

A focused library of business case studies. Start with the featured work, then explore the deeper analyses across growth, operations, risk, finance, marketplaces, workforce analytics, and applied AI.

Flagship work

Five studies put the business question, the available evidence, and the limits of each analysis in view.

ComplaintFlow: A Reliable AI Triage Service for Consumer Support

A production-oriented AI service that routes financial complaints, retrieves approved support playbooks, escalates uncertainty, and records enough evidence to evaluate every decision.

Business questionSupport teams need to route complaints quickly, but an unreliable AI response can send a customer to the wrong queue or invent guidance.

Decision signalComplaintFlow combines a transparent baseline, approved-playbook retrieval, provider retries, PII redaction, schema checks, human escalation, and an auditable SQLite decision log.

Online Retail: Finding the Customers Behind Repeat Revenue

A transaction analysis showing where repeat revenue comes from and which markets deserve closer growth attention.

Business questionThe retailer had many transactions, but it was not clear which customers and markets created repeat revenue.

Decision signalAfter cleaning 541,909 transactions, the analysis found 4,338 identifiable customers and a strong concentration of revenue among the highest-value customers.

NYC Taxi Operations: Matching Capacity to the Right Trips

A large-scale trip analysis showing when demand peaks and why airport-linked trips need a different operating plan.

Business questionTaxi demand was not evenly distributed across time or location, making a single operating plan inefficient.

Decision signalAfter filtering 3.48 million raw records, the analysis showed that evening demand peaked in Manhattan while airport-linked areas produced longer, higher-value trips.

NYC Restaurant Inspections: Where Quality Risk Needs Attention

A messy inspection-data analysis showing how a restaurant group could prioritize quality support without treating every location as the same risk.

Business questionRestaurant inspection records were detailed but repeated across violations, making it difficult to compare quality risk fairly.

Decision signalAfter rolling 228,315 recent violation rows into 73,478 inspection records, the analysis found meaningful differences in critical-violation rates across boroughs.

sentence-transformers

Detecting Fake Job Posts Before They Reach Job Seekers

A fraud-detection study showing why the language in a job post is more useful than surface-level profile details.

Business questionFake job posts were rare, so a system could appear accurate while missing every scam.

Decision signalThe best balance caught 75% of fake posts and was correct 98% of the time when it raised an alert.

Open case study

Supporting library

Additional analyses demonstrate breadth across product, marketing, pricing, finance, marketplaces, workforce, and risk.

Credit Default Risk: Prioritizing Support Before Losses Grow

A transparent risk-prioritization study that combines repayment history and exposure while keeping fairness, drift, and support outcomes visible.

Business questionA lender needs to focus attention where repayment risk and potential exposure are higher without turning a historical score into an automatic lending decision.

Decision signalThe validated UCI archive contains 30,000 clients and 24 fields covering credit limits, repayment status, bills, payments, and the observed default outcome.

Workforce Planning: Separating Hiring Pressure From Turnover

A workforce decision framework that distinguishes open roles, completed hiring, voluntary quits, and total separations.

Business questionLeaders can misread staffing pressure when openings, hires, and exits are treated as one workforce signal.

Decision signalIn the BLS December 2025 aggregate series, openings were 6.55 million, hires were 3.3%, quits were 2.0%, and total separations were 5.203 million.

New York Short-Term Rentals: Finding Marketplace Concentration

A supply-side marketplace analysis that separates listing growth from marketplace health and identifies where concentration creates risk.

Business questionA large listing count does not prove healthy demand, fair choice, or resilient supply.

Decision signalThe June 2026 snapshot has 30,555 listings; 24.7% of listings belong to the top 1% of observed hosts, and the filtered median listed price is $171.

Read case study

financial performance

financial analysis

revenue

profitability

Company Performance: Explaining Growth Before Calling It Healthy

A public-filing analysis that separates reported growth from profitability and turns divergence into management questions.

Business questionRevenue growth can hide margin pressure, while profit growth can be distorted by mix, accounting, or one-off items.

Decision signalFor Microsoft, SEC Company Facts reports FY2023 revenue of $211.9B and net income of $72.4B, rising to $281.7B and $101.8B in FY2025; reported net margin moved from about 34.1% to 36.1%.

Airfare Pricing: Making Revenue Decisions With the Right Benchmark

A commercial decision framework that separates nominal fare movement from inflation and shows what additional evidence is needed before changing route prices.

Business questionA higher average fare can reflect real pricing power, inflation, route mix, or passenger mix; acting on the wrong explanation can reduce demand or margin.

Decision signalThe official BTS series provides annual and quarterly domestic fares in current and inflation-adjusted dollars, making it possible to separate market context from route-level pricing action.

Read case study

marketing performance

campaigns

customer segments

response rate

Bank Marketing: Targeting Response Without Confusing It With Profit

A campaign decision study that identifies the strongest response signals, quantifies the addressable audience, and defines the missing economics needed for budget allocation.

Business questionThe bank made 45,211 contacts, but broad response rates hid the difference between audience quality, channel coverage, and prior campaign history.

Decision signalThe overall positive-response rate is 11.7%; cellular contacts respond at 14.9%, unknown contact types at 4.1%, and previously successful contacts at 64.7%.

Read case study