# Delivery Economics and the Scaling Ceiling ## Deployment scale limits the company, and software scale does not The figures of Foreman show the growth curve. The time to put each proof of concept into production is **months, not days**. The labor cost for this work is a low-five-figure sum. A deployment engineer supports only a small number of customers. Integration is the line with the lowest margin, because the cost is in proportion to the hours. > [!important] The ceiling as a formula > The company multiplies the number of deployment engineers by the number of customers for each engineer. The result is approximately the number of customers that the company serves at one time. > > The company has fewer than ten persons. This fact limits the business to a scale much below the scale that a platform narrative shows. The limit changes only if the company uses much more of its work again. ## Why the reported margin can look better than the model The company reports the cost of goods sold as **zero**, because the company does not put the deployment staff in this cost. If the company puts the deployment labor in this cost, the gross margin looks like the margin of a services business. This is a point of presentation in the accounts. It does not necessarily show that the company did something incorrect. But it is important for the multiple. ## The test for linear scale | If the test shows | Then | | --- | --- | | The engineer-days for each deployment decrease with each customer | The company is a platform, and the margins increase. | | The engineer-days are constant or increase | The company is a services company. The headcount increases in proportion to the revenue. | | The deployments use less than half of the codebase again | The company is a services company (see the ratio in [[Industrial AI Unit Economics]]). | > [!warning] On-premise deployments increase the work > Each single-tenant install is an isolated member of the fleet. The company must upgrade each member. Without upgrade tools, the versions become different. The maintenance cost increases with the number of customers. > > The notes [[Model Maintenance at Scale]] and [[Industrial MLOps]] give more information. ## Why it matters This is the quantitative core of the diligence. The test is cheap. The test uses three recent deployments. For each deployment, the test measures the elapsed time, the engineer-days, the shared commits and the custom commits. The result shows which valuation frame applies. ## Related - [[Foreman MOC]] - [[Platform or Solutions - The Central Question]] - [[Deployment Velocity]] - [[Bespoke Engineering in Industrial AI]] - [[Consultancy-to-Platform Transition]]