How Trek Funded an Edge AI Bike Prototype in 18 Months
Trek's Advanced Tech director compressed 35 AI opportunities down to 3, won a fresh AI budget during a company-wide freeze, and shipped a CES 2025 on-device AI handlebar — all inside 18 months.
18 months
Physical AI prototype at CES
Gregor Mittersinker
CEO & Founder
Loft Design
Connect ↓
The Challenge
Trek's Director of Advanced Technologies wanted to bring AI into the organization but faced a sprawling value chain, frozen budgets, hiring freezes, and post-COVID travel restrictions. With too many potential AI entry points and no clear prioritization framework, the team needed a fast, credible path to executive approval — one that could survive scrutiny from Trek's president without requiring a multi-year, high-risk investment.
What They Built
Loft Design deployed their Growth Edge sprint methodology to audit Trek's full value chain and identify 35 distinct AI opportunity areas. These were filtered to three finalists using agreed-upon metrics balancing prototyping feasibility with ROI potential. Loft then delivered a compelling executive business case that secured a large new AI budget during a period of financial austerity. The unlocked budget enabled a second phase: designing a physical AI-embedded handlebar system for CES 2025, housing a small language model running locally on a microcontroller that functions as an on-device AI cycling companion — answering real-time questions about terrain, battery life, and trail conditions without internet connectivity.
Loft Design deployed their Growth Edge sprint methodology to audit Trek's full value chain systematically, surfacing 35 distinct AI opportunity areas across the business. These were evaluated against agreed metrics balancing prototyping feasibility with ROI potential, narrowing to three finalists. Loft then constructed a compelling executive business case for the highest-priority opportunity — one designed to survive scrutiny from Trek's president in a period of hiring freezes, travel restrictions, and post-COVID financial austerity. The business case secured a significant new AI budget. Phase two used that budget to design and build an AI-embedded handlebar system for CES 2025: a physical prototype housing a small language model running locally on a microcontroller, with no internet connectivity required. The on-device SLM functions as a real-time AI cycling companion — answering questions about terrain, battery life, and trail conditions from data available on the device. Running a capable language model on a resource-constrained microcontroller without cloud dependency required significant edge AI engineering work, combining embedded systems expertise with hardware-software integration across a twelve-plus-month timeline.
AI Role
In the second phase, a small language model runs locally on a microcontroller embedded in a handlebar system, providing an on-device AI cycling companion that answers real-time questions about terrain, battery life, and trail conditions without requiring internet connectivity. In the first phase, AI-assisted analysis tools supported the evaluation and scoring of 35 identified opportunity areas against agreed-upon feasibility and ROI metrics.
Infrastructure
Small language model (SLM, on-device) • Microcontroller hardware (embedded systems) • Generative AI design tools • Edge AI optimization stack
Integration Points
On-device sensor data → SLM inference engine (microcontroller) • SLM → real-time handlebar interface (terrain, battery, trail queries) • No external API or internet connectivity — fully edge-deployed
Impact
AI Budget Approved During Company-Wide Freeze
Loft's structured executive narrative enabled Trek's Advanced Tech director to secure a significant new AI budget during a period of hiring freezes, travel restrictions, and post-COVID financial austerity.
35 Opportunities → 3 Funded Priorities
Growth Edge sprints mapped Trek's entire value chain, identified 35 AI opportunity areas, and narrowed the field to three high-ROI priorities — giving the director a crisp, defensible recommendation for the president.
Physical AI Prototype at CES in 18 Months
Within approximately 18 months of initial engagement, Loft had a working physical prototype at CES 2025 — the world's first on-device AI agent for cyclists, running a small language model on local hardware with no internet connection required.
Implementation Complexity
Running a language model locally on a microcontroller in embedded hardware — without internet connectivity — is a technically demanding engineering challenge that requires specialised skills in embedded systems, edge AI optimisation, and hardware-software integration. The physical product design and CES prototype delivery add further complexity beyond a standard software AI implementation.
Best Fit For
Best for organizations with a dedicated innovation function committed to shipping real physical-digital products — in categories like mobility, consumer hardware, medtech, or life sciences — that need a structured process to quickly prioritize AI opportunities and build an executive-approved investment case.
