Skip to content

Latest commit

 

History

History
40 lines (31 loc) · 3.83 KB

File metadata and controls

40 lines (31 loc) · 3.83 KB

📖 Case Studies: Real AI Products, Real Numbers, Real Sources

TL;DR: Interviews reward candidates who argue from evidence: named products, dated facts, honest numbers. Every stat in this section was verified against primary sources (papers, SEC filings, company announcements, tribunal records) in July 2026, and the ones we couldn't verify are flagged so you don't repeat myths.


The deep dives

Case The one-line story Best used for
GitHub Copilot Capability-curve timing + distribution + freemium ladder → category king Strategy, pricing, developer products, the 55% RCT
Netflix Recommendations The canonical "invisible ML = $1B of retention" story, and how to cite it honestly Classic ML value, metrics, personalization
Duolingo A decade of quiet ML → GPT-4 first-mover → "AI-first" backlash Data flywheels, premium AI pricing, comms risk
Klarna AI Assistant The efficiency triumph that publicly reversed Metrics guardrails, agent products, exec storytelling
Failures Hall of Lessons 10 sourced disasters, one lesson each Risk questions, "tell me about an AI failure," design reviews

Quick-hit cases (facts verified July 2026)

Product Verified facts PM takeaway
ChatGPT Launched Nov 30, 2022; est. 100M monthly users in ~2 months (UBS/Similarweb analyst estimate. Note: estimate, not disclosure); 900M weekly actives + 50M paying subscribers (OpenAI, Feb 2026) The reference consumer-AI growth curve, and a lesson in citing estimates as estimates
Notion AI Nov 2022 alpha waitlist hit 2M+ signups; GA Feb 2023 as $10/member add-on; May 2025: add-on killed, AI bundled into Business tier ($20/user/mo) Add-on pricing while AI is novel → bundling when it's table stakes
Zoom AI Companion Launched Sept 2023 free with paid plans; paid-tier AI MAU +184% YoY (May 2026 earnings) "Free with subscription" as retention moat, monetize later
Adobe Firefly Launched Mar 2023 trained on licensed/Adobe Stock content; 1B images within ~4 months; 24B+ generations by 2025 "Commercially safe" as differentiation: legal posture as product strategy
Spotify Discover Weekly / AI DJ DW launched July 2015, with 100B+ tracks streamed from it in 10 years (Spotify, 2025); AI DJ launched Feb 2023 Personalization as brand; a decade-long compounding feature
Grammarly → Superhuman 40M+ DAU reported; acquired Coda (2024) + Superhuman (2025); rebranded company to Superhuman (Oct 2025) Single-feature AI companies must expand into suites as models commoditize their core

How to use case studies in interviews

flowchart LR
    A["Claim you want to make<br/>'efficiency metrics need quality guardrails'"] --> B["Anchor: name + date + number<br/>'Klarna, Feb 2024: AI handled 2/3 of chats...'"]
    B --> C["Turn: the insight<br/>'...but by May 2025 the CEO publicly reversed:<br/>quality had degraded'"]
    C --> D["Apply to the question<br/>'so for YOUR support AI, I'd pair<br/>cost-per-ticket with CSAT as co-equal metrics'"]
Loading

Three rules:

  1. Date every number. "Netflix says recommendations drive 80% of viewing" is a 2015 self-reported figure. Saying so makes you more credible, not less.
  2. Prefer pairs: a success + a failure on the same theme beats two successes (Copilot + Watson Health on "timing the capability curve" is a great pairing).
  3. Never repeat unverified myths. Common ones we could NOT verify (don't use): "Copilot writes 46% of code" (undated talking point), "Klarna fired 700 people and replaced them with AI" (false framing: the 700 was a workload-equivalence claim; the layoffs were separate and earlier), Spotify AI DJ usage percentages, "ChatGPT has 1B users."