# Consultancy-to-Platform Transition The most common failure mode for deep tech startups: building something genuinely valuable but delivering it through services instead of software. Revenue grows linearly with headcount. Margins stay thin. Investors see a consultancy. ![[Screenshot 2026-02-05 at 12.42.59.png]] The tell: ask how much of the codebase is reusable across customers. If the answer is "60-70% reusable" but deployments still take 320 hours, someone is wrong. Either the reusable percentage is inflated, or the bespoke 30% is where all the time goes. ![[Screenshot 2026-02-05 at 12.43.05.png]] Where bespoke engineering hides in industrial AI: - Data pipeline construction (every plant has different historian systems) - Domain-specific event labelling (requires process engineering judgment) - Threshold calibration and alarm tuning (plant-specific, needs operator buy-in) - Model validation against plant data (most time-consuming step) - Change management and operator trust (high-touch, doesn't compress) ![[Screenshot 2026-02-05 at 12.56.39.png]] The transition path: productize each bespoke step one at a time. Start with whatever step consumes the most hours. Build tooling that reduces it by 4x. Then move to the next bottleneck. This is exactly what a [[Venture Building Manifesto]] studio model is designed to do: bridge the gap between domain expertise and operational scale. The venture studio fit is strongest when the founding team has genuine domain depth but lacks the product/platform engineering to abstract their expertise into repeatable software. Related: [[Deployment Velocity]], [[Industrial AI Unit Economics]], [[Bespoke Engineering in Industrial AI]], [[Industrial AI MOC]] --- Tags: #deeptech #systems #kp