What is the link between AI agents and ERP? AI agents fail in project-based companies when they work on fragmented data, without business context and without the authority to act inside a governed system. An integrated ERP gives an agent one data set, a cost structure it can reason over, and roles and approvals that control what it may do. DANAOS ProjectVIEW AI is built on that principle: its agents run on ProjectVIEW ERP, the integrated ERP built on the BoQ ↔ WBS ↔ Cost Codes framework. This page is for directors and C-level teams in construction, shipbuilding and repair, marine and offshore, mining and quarrying, and project-based manufacturing who are deciding where AI agents should sit in their systems landscape. Why do AI agents fail in project-based companies? Most failed agent pilots do not fail because the model is weak. They fail because of what the model is connected to. Four causes come up again and again. Fragmented data. Estimates sit in spreadsheets, commitments in a purchasing tool, progress in the planning software, invoices in accounting. Each system holds part of the truth. An agent that reads one of them gives a partial answer with full confidence. No business context. A purchase order line is just a number unless the agent knows which BoQ item, which WBS element and which cost code it belongs to. Without that link, the agent cannot tell whether a quotation is above the budget rate or whether a cost is committed against the right activity. No transactional authority. An agent outside the system of record can suggest, but it cannot create, approve or post anything in a controlled way. Either it stops at advice, or it acts through workarounds that bypass permissions and leave no audit trail. It amplifies broken processes. AI will not fix broken processes. It will amplify them. If approvals are informal, cost codes are inconsistent and data is re-keyed between systems, an agent automates the inconsistency at higher speed. We cover the process side in more detail in the insight AI agents will not fix a broken ERP. What does an integrated ERP give an AI agent? An integrated ERP is not a data source an agent happens to read. It is the frame that makes the agent’s work reliable and controllable. A shared cost structure. In ProjectVIEW ERP every estimate, budget, commitment, progress claim and cost is tied to the BoQ, the WBS and the cost codes. An agent can follow a number from the tender to the final account without guessing how the pieces connect. One data set. Estimating, procurement, subcontractors, materials, machinery, HR and accounting work on the same records. ERP records what happened. AI explains why it happened and what will happen next. Roles and permissions. The agent works through the governance chain Identity → Role → Permission → Data → AI → Action. It sees and does only what the user’s role allows. Approvals and audit. Consequential actions go through the same approval workflows as human actions, and every step is logged. The combination is deterministic ERP plus probabilistic AI: the ERP enforces the rules, the AI finds patterns, explains and recommends. How does an agent work on top of ProjectVIEW ERP? ProjectVIEW AI agents follow a fixed cycle: Detect → Investigate → Explain → Recommend → Request Approval → Execute → Audit. Example: a procurement agent detects a supplier quotation above the budget rate for a BoQ item. It investigates the item’s cost recipe, previous purchases and current commitments, explains the gap, and recommends an action, for example re-quoting or approving with a variance note. The buyer or project manager approves or rejects. Only then is the action executed in the ERP, and the full sequence is recorded for audit. Agents monitor and analyse continuously, but consequential actions always need human approval. That is a design choice, not a limitation. How does AI on fragmented data compare with AI on an integrated ERP? Dimension AI on fragmented data AI on an integrated ERP Data Copies and extracts from several systems; often out of date One live data set across modules Business context Agent must infer how items relate Every record tied to BoQ ↔ WBS ↔ Cost Codes Answers Plausible but hard to trace to source Traceable to the transactions behind them Permissions Separate access model, often broader than the user’s Inherits the user’s ERP role and permissions Actions Advice only, or workarounds outside controls Executes in the ERP after approval Audit Partial or none Each step logged with the approver Process quality Amplifies existing gaps Works within defined workflows Maintenance Pipelines and mappings to keep in sync No duplicate data layer to maintain When is AI on fragmented data acceptable? Not every use of AI needs an integrated ERP. Be honest about the use case. Drafting and summarising. Writing a letter, summarising a contract or translating a method statement works well with a general assistant and a document. One-off analysis. A single export analysed once, with a person checking the result, carries limited risk. Exploration. Testing what AI can do before a systems decision is a reasonable first step. The line is crossed when the agent’s output feeds a commitment, a payment, a certification or a forecast that management relies on. At that point the agent needs the context and controls of the system of record. What if we already run another ERP? ProjectVIEW AI can also work on top of existing systems. It connects to third-party ERP such as SAP and Oracle, Primavera P6, MS Project, BIM models and document systems, and maps data to the BoQ, WBS and cost code structure. There are three ways to buy: ProjectVIEW ERP only, ProjectVIEW AI on top of existing systems, or both together as ProjectVIEW OS. The trade-off is clear: the more fragmented the underlying landscape, the more effort goes into mapping and reconciling data before agents can be trusted. See third-party ERP integration and ProjectVIEW vs generic ERP systems. How should a company prepare for AI agents? Agree one cost structure (BoQ, WBS, cost codes) and use it in every module. Formalise approvals: who approves what, at which value, on which project. Remove re-keying between systems where possible. Start agents on one process with clear rules, for example quotation checks against budget rates. Keep a human approver on every consequential action and review the audit log. For the AI side, see ProjectVIEW AI agents and security and governance. For the cost structure, see what is project cost control. Frequently asked questions Can AI agents work without an ERP? They can draft, summarise and analyse documents. For decisions that affect cost, commitments or payments, they need the data, context and approvals that an integrated ERP provides. Without them, outputs are hard to trace and harder to control. Why does the BoQ ↔ WBS ↔ Cost Codes framework matter for AI? It gives every transaction a place in the project’s cost structure. An agent can compare a quotation to the budget rate, a commitment to the WBS budget, or progress to cost, because the links already exist in the data. Do ProjectVIEW AI agents act on their own? Agents detect, investigate, explain and recommend. Consequential actions require human approval before they are executed in the ERP, and each step is logged for audit. Do we need to replace our current ERP to use ProjectVIEW AI? No. ProjectVIEW AI can run on top of existing systems such as SAP or Oracle, or together with ProjectVIEW ERP as ProjectVIEW OS. The cleaner the underlying data, the faster agents become useful. How long does a ProjectVIEW ERP implementation take? Implementation time is 5 to 7 months. Pricing depends on the modules, the number of users and the deployment option. Want to see it on your own data? See how ProjectVIEW AI agents work on a single, governed ERP data set built on BoQ ↔ WBS ↔ Cost Codes. Book a Demo