AI Thinking Process
Pivot attempted before kill decision: Attempted pivot to room-level occupancy for smart-thermostat integration. Pre-empted by 4+ large incumbents (Google Nest, Apple/Ecobee, Amazon) already shipping presence detection as native features.
KILLED — kill_type=accuracy_cliff. WiFi CSI-based appliance-level electricity disaggregation achieves only 60-75% accuracy in single-room conditions, dropping below 40% in multi-room environments — below what consumers accept when comparing against Sense's CT-clamp product. The accuracy limitation is a physics constraint at the receiver level (multi-path interference), not a software problem. The pivot to occupancy-sensing was pre-empted by Google Nest, Apple Ecobee, and Amazon, all of which ship presence-detection natively.
Lesson: WiFi CSI ISAC is a real capability but scale-shifted into the home hits two walls simultaneously: (a) accuracy-cliff for anything measurement-critical (fails vs CT-clamp products), and (b) incumbent-native-feature absorption for anything convenience-oriented. Consumer WiFi-sensing plays need a use case that is neither measurement-critical nor already covered by Google/Apple/Amazon — which leaves very little space.
Kill Reason
WiFi CSI-based appliance-level electricity disaggregation achieves only 60-75% accuracy in single-room conditions, dropping below 40% in multi-room environments — below what consumers accept when comparing against Sense's CT-clamp product. The accuracy limitation is a physics constraint at the receiver level (multi-path interference), not a software problem. The pivot to occupancy-sensing was pre-empted by Google Nest, Apple Ecobee, and Amazon, all of which ship presence-detection natively.
Risk Analysis
Risk analysis available for latest engine ideas.
Loading...
Related ideas you can explore free:
killed: Distribution is the weakest of any idea in this session — no killer channel exists. Book Creator dominates the 'create not consume' positioning in iPad education and could add AI text-suggestion and tap-trigger interactivity in a single sprint. Roblox Studio with Roblox Assistant (AI generation, 2024) is extending into kid-creation formats. The HN Decker signal is a nostalgia data point, not a market-demand signal — kids are not the HN audience. Conviction fell below the 35% minimum viability threshold at deeper analysis.
killed: The receiver-side family — whose bank-account aggregation consent is required for the product to function — often prefers not to have their spending itemized to the sending worker. This is not a 'market doesn't want it' structural adoption barrier but a 'the party you need to install the app does not want it installed' structural barrier. Without receiver consent to bank aggregation, the product cannot deliver the core reconciliation function. The structural adoption barrier collapses the required dual-install regardless of sender demand.
killed: Incumbents Allflex (Merck), Nedap, Afimilk have built regional dealer networks in SEA/LatAm/Kenya/Turkey over 10-20 years. A hardware startup entering these markets would need 3+ years to build equivalent distribution before the farming co-ops will risk adopting an unproven brand. The technical wedge (herd-scale CSI sensing without per-animal RFID) is real, but distribution moat vs incumbents is the binding constraint in EM ag-tech.