# Consultancy-to-Platform Transition The most common failure mode in deep tech: 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 is simple. Ask how much of the codebase is reusable across customers. They'll say 60-70%. Then ask how long a deployment takes. If the answer is 320 hours, that reusable number is doing heavy lifting in the pitch deck and very little in the field. The bespoke 30% is eating all the time. [[Bespoke Engineering in Industrial AI]] breaks down exactly where those hours hide: wiring up data pipelines to whatever historian system the plant runs, getting process engineers to label events correctly, calibrating thresholds and alarms, validating models against live plant data, and the really stubborn one, convincing operators to trust the system. ![[Screenshot 2026-02-05 at 12.43.05.png]] ![[Screenshot 2026-02-05 at 12.56.39.png]] The first thing that fools people is pilot revenue. Enterprises have R&D budgets specifically for testing new technology. A $46-60k pilot with a 3-month timeline and an exit clause is a rounding error for the buyer and a vanity metric for the seller. Those blue-chip logos look great on a deck, but they don't mean anyone expanded past the test. [[Team Growth x Product Market Fit]] is clear on this: only long-term contracts indicate real value. If all revenue is new logos and none of it is expansion, the [[Land-and-Expand in Enterprise AI]] motion is unproven. Break revenue into new customers vs. expansion of existing. Radically different scenarios for the same number. Here's the nuance though. That bespoke work early on is correct. [[3 Hard Truths of Deep Tech Commercialization]] lays it out: you need custom, application-specific products first before you can generalise into a platform. Platforms only make money under very specific conditions: engaged communities, mature technology, and real market pull creating [[Network Laws]]. The trap is staying in bespoke mode and calling it a platform anyway. Jumping to "platform" language before doing the ugly, vertical-specific work is how teams get stuck. The other trap is teams who dress up slow deployment as "enterprise-grade thoroughness." R&D becomes a cloak for unproductized engineering. Benchmarking battles replace customer feedback. Being "enterprise ready" becomes the excuse for never compressing deployment hours. [[The Deep Tech Growth Cycle is different]] calls this out as corporate tax: veterans from large organizations importing hierarchical decision-making, longer timelines, and larger budgets that break down when you need speed. [[Commercialising deep tech early is uncomfortable though essential]] because early market exposure is where you actually learn. The [[Productivity Paradox]] applies too. Deep tech communities over-index on benchmarking that doesn't translate to what the customer cares about. The way out is boring and sequential. Find the step that burns the most hours. Build tooling that cuts it by 4x. Move to the next bottleneck. [[Wright's Law]] applies: every doubling of cumulative deployments should yield a measurable drop in [[Deployment Velocity]]. If hours per deployment aren't falling, there's no learning curve. No learning curve means linear scaling. Keep compressing until deployment feels like configuration, not a custom engineering project. This is what a [[Venture Building Manifesto]] studio model is built for: bridging the gap between [[domain specific sense-making]] and operational scale. The fit is strongest when the founding team has genuine domain depth but lacks the product engineering to abstract their expertise into repeatable software. Related: [[Deployment Velocity]], [[Industrial AI Unit Economics]], [[Bespoke Engineering in Industrial AI]], [[Industrial AI MOC]], [[Team Growth x Product Market Fit]], [[Land-and-Expand in Enterprise AI]], [[3 Hard Truths of Deep Tech Commercialization]], [[The Deep Tech Growth Cycle is different]], [[Incumbent Bundling Risk]], [[Execution x Evolution x Disruption]], [[Technical Moat Assessment Framework]] --- Tags: #deeptech #systems #kp #firstprinciple