# AI Deployment Layer — Map of Content This map covers a class of company: a vendor that puts AI agents into the daily work of large organisations. It is generic and uses public sources only. The idea stack, in order: a thesis, a forcing function, a wedge, a control layer, a moat test and a diligence plan. Start anywhere and follow the wikilinks. > [!abstract] How to read this map > **Thesis:** the constraint moved from AI access to AI deployment ([[The Deployment Gap]]). **Forcing function:** most pilots stall and Gulf data rules add cost ([[Pilot-to-Production Failure Rates]], [[Data Residency in the GCC]]). **Wedge:** one measured workflow, then expansion ([[Workflow-First Land and Expand]]). **Control layer:** connectors, permissions and audit ([[Connector Layer and MCP]], [[Governance and Audit as Product]]). Then comes the test of whether it is a company or a feature ([[Defensibility Test]]). > [!info] At a glance > - **What** — a software layer between company systems and AI models > - **Core idea** — deployment, not the model, is the scarce item > - **Wedge** — one high-value workflow with a measured result > - **Market** — large, but the size figures are secondary and disputed > - **Stage** — unknown from public data; see [[Open Questions and Numbers to Reconcile]] --- ## 1 · The scarce item is deployment, not access Strong models are cheap to buy. Connecting them to systems, data, rules and people is the hard step. > [!tip] Claim > The vendor that owns the deployment step can use any model as a supply input. → [[The Deployment Gap]] → [[AI Eats Services Not Software]] → [[AI Agents Stack]] ## 2 · Most pilots stall, so buyers have a pain Public studies disagree on the exact rate but agree on the causes: data, integration, change and ownership. > [!warning] Caution > A high failure rate proves pain. It does not prove that buyers will pay for the fix. → [[Pilot-to-Production Failure Rates]] → [[Evidence Quality and Vendor Claims]] ## 3 · Enter with one workflow and expand The first workflow pays for connectors and policy. Later workflows re-use both. > [!tip] Test > Ask for revenue by cohort and by workflow count per customer. → [[Workflow-First Land and Expand]] → [[Workflow Prioritisation Engine]] → [[Land-and-Expand in Enterprise AI]] → [[Consultancy-to-Platform Transition]] ## 4 · Control is the product Open standards make connection cheap. Permissions, audit and human approval make it safe. > [!important] Shift > If connection is cheap, value moves to control of the connection. → [[Connector Layer and MCP]] → [[Governance and Audit as Product]] → [[Human-in-the-Loop Systems]] → [[Model Neutrality and Routing]] ## 5 · Sovereignty is a buying reason with a cost The Gulf has six rule sets. Local hosting, keys and logs are real needs. They raise cost. → [[Data Residency in the GCC]] → [[Sovereign AI Positioning]] → [[Data Embassies]] ## 6 · Price must follow measured value → [[Value Measurement and Pricing]] → [[Outcome-Based Pricing]] ## 7 · Company or feature? Four groups compete for this layer. Suite bundling is the main threat. → [[Competition and Bundling]] → [[Incumbent Bundling Risk]] → [[Defensibility Test]] → [[AI era Defensibility]] → [[Technical Moat Assessment Framework]] ## 8 · How to check it → [[Diligence Question Set]] → [[Technical DD Framework]] → [[AI Agent Vertical SaaS DD MOC]] → [[Emerging Market SaaS DD]] → [[Open Questions and Numbers to Reconcile]] --- ## The company, compressed > [!abstract] Class of company > A vendor that deploys governed AI agents into large organisations. It connects to ERP, CRM, email and documents, sends tasks to the right model, keeps an audit log, and reports business results. It offers regional hosting for Gulf buyers. > > **Reality check:** public data shows a clear story and vendor case studies. It does not show revenue, retention, margin, security audits or independent customer results. Treat the thesis as unproven until grade A or B evidence exists ([[Evidence Quality and Vendor Claims]]). ## Concept map ```mermaid graph TD A[Deployment gap] --> B[Pilot failure] A --> C[Connectors and MCP] C --> D[Governance and audit] B --> E[One workflow first] E --> F[Measured value] F --> G[Expansion] D --> H[Regional hosting] D --> I[Defensibility] G --> I J[Suite bundling] --> I class A,B,C,D,E,F,G,H,I,J internal-link; ``` ## The whole thing in one breath > [!tip] Summary > Models are cheap, so the scarce item is [[The Deployment Gap|deployment]]. Most pilots stall ([[Pilot-to-Production Failure Rates]]), which gives a vendor room to [[Workflow-First Land and Expand|land with one workflow]]. Open standards make connection cheap, so value moves to [[Governance and Audit as Product|control and audit]]. The vendor survives only if the combination of [[Defensibility Test|moats]] beats [[Incumbent Bundling Risk|suite bundling]]. ## Related - [[Context Layers MOC]] - [[Switching Cost Design]] - [[AI Agent Vertical SaaS DD MOC]]