CES 2026: AI is Rewiring Product Development - And Crypto Can Use the Same Playbook
At CES 2026, McKinsey showed how to compress six to nine months of product work into roughly two weeks. The demo, hosted at the Fontainebleau, combined AI-driven insights, digital testing, and simulated customers into a tight loop that makes slow teams look obsolete.
"The key to great product is fast iteration. Try this. Did it work? Okay, it's okay, but these three things are still a problem." That mindset, backed by AI and scale, is what turns ideas into shippable decisions in days.
What McKinsey's workflow actually does
- Ingest signal at scale: Pull 100,000+ unprompted comments from TikTok, app reviews, and social threads. Cluster them into specific attributes engineers can act on. It's faster - and often truer - than surveys.
- Turn insights into concepts fast: Generate visual concepts in about an hour. Frame clear value props and claims you can test. Keep everything measurable.
- Test with AI personas and real people: Run flows and copy against modeled "personas" (e.g., suburban mom, 45-year-old football dad) and large samples of real users. For crypto, think Bitcoin maxi, DeFi yield chaser, mobile-only retail user. Stress-test comprehension, risk perception, and conversion friction before mainnet.
- Decide with data in days: Get statistically meaningful reads from thousands, not 20 people behind glass. Kill weak ideas. Double down on winners. Feed learnings back into the backlog.
- Loop every two weeks: Tiny build, relentless testing, rapid iteration. That's the cycle.
Why generic AI won't save your roadmap
You can throw the right questions at a general chatbot and still get weak answers. The difference is training and data. McKinsey's edge comes from two decades of product cases and outcomes, then tuning AI on top of that library.
For product leaders, the takeaway is simple: your proprietary feedback, support tickets, experiments, and postmortems are the asset. Build your own corpus, wire it into RAG or fine-tuning, and evaluate outputs like you evaluate features - with benchmarks, not vibes.
A 30-day plan to copy the model
- Week 1 - Build your signal engine: Aggregate TikTok, Reddit, X, app-store reviews, support logs. Strip PII. Normalize into a single schema. Stand up embeddings + a vector store.
- Week 2 - Cluster and prioritize: Auto-cluster themes. Manually label the top 25 attributes tied to activation, retention, or revenue. Translate each into testable hypotheses and acceptance criteria.
- Week 3 - Define and calibrate personas: Draft 6-10 personas with clear jobs, constraints, and risk profiles. Calibrate with a small panel so your agents mirror real behavior.
- Week 4 - Run the two-week loop: Produce 3-5 concept variants. Test with 1,000+ people and your persona agents. Pick the winner. Ship a small slice. Repeat.
Metrics that matter
- Time-to-signal (TTS): Hours from idea to first statistically valid read.
- Iteration velocity: Tested changes per sprint (aim for 3+).
- Cost per insight: Dollars per statistically sound decision.
- Activation/retention lift: Especially on critical paths.
- Abandonment rate on key steps: Onboarding, funding, checkout, or swap.
- NPS or CSAT delta per change: Tied back to specific attributes.
- Risk flags: Misunderstood fees, security concerns, compliance friction.
For crypto builders: apply it to wallets, exchanges, and DeFi
- Onboarding: Compare seed phrase flows vs. passkeys/email. Measure drop-offs, not opinions.
- Token design: Use agents to poke holes in emissions, vesting, fees, and rewards. Test "what if" scenarios before they cost you public trust.
- Liquidity and risk: Simulate bank-run moments, oracle hiccups, and gas spikes. Check user comprehension of slippage and fee transparency.
- Compliance UX: Make KYC/AML steps clearer with fewer surprises. Run comprehension tests on disclosures.
- Wallet connect and funding: Measure connection success rate, time-to-first-fund, and the impact of fee previews on conversion.
Guardrails you'll need
- Privacy: Strip PII and respect platform TOS when ingesting public data.
- Bias: Balance your panels and test for fairness across segments.
- Quality: Use golden datasets and adversarial prompts to catch hallucinations and weak reasoning.
- IP and claims: Lock prompts, datasets, and outputs behind policy and review.
- Human-in-the-loop: PMs and researchers review deltas before production.
Consumer brands are already getting big, trustworthy reads in days. Crypto teams don't get to blame "market conditions" for moving slow. The blueprint is on the table. The only question is who uses it first.
Want to see a reference model for the organizational side of this change? Browse McKinsey's Rewired approach and the CES program at CES.
If your team needs structured upskilling to run this loop, explore practical AI courses for product roles.
Your membership also unlocks:
