Capital project management is a discipline defined by accountability. Owners answer to boards, regulators, funding agencies, and the communities that will use what gets built. Construction managers answer to the owners who hired them. That accountability does not transfer to a software vendor, and it does not transfer to a model.
AI is arriving in capital programs faster than the governance around it. The pressure to adopt is real, and so is the risk of adopting badly. The six questions below are not a product evaluation checklist. They are the questions that decide whether the intelligence you add to your program strengthens your accountability or quietly erodes it.
1. Can it show the source record behind every number?
An AI that produces a cost forecast, a schedule risk score or a contract summary is producing something you may act on and will certainly be asked to explain. If the answer cannot be traced to a specific record, the change order, the pay application, the schedule revision, then it is not evidence. It is a suggestion in the costume of a fact.
What good looks like: every AI-generated figure links back to the record it came from, and a reviewer can open that record in one click.
2. Does it respect the permissions already in place?
Most owner organizations have spent years getting access control right. Contract values are visible to some people and not others. Claim positions are restricted. Vendor pricing is confidential. An AI layer that reads across your data without inheriting those rules will eventually surface a restricted figure inside a helpful summary, to someone who was never cleared to see it. That is a permissions breach with a friendly interface.
What good looks like: the AI sees exactly what the person asking sees and no more, governed by the permission model you already administer rather than a second one maintained alongside it.
3. Is it reading your data, or training on it?
This is a procurement question, not an IT footnote. Ask where the data goes, whether it leaves your tenant, how long it is retained, and whether it becomes training material for a model other organizations will use. Get the answer in the contract, not the sales deck.
What good looks like: a written statement of data residency and retention, plus an explicit commitment that your program data is not used to train shared models.
4. Is there an audit trail a dispute would survive?
Capital programs get audited, and they get litigated. Two years from now someone may ask why contingency was drawn down in March. "The system suggested it" is not a record. You need to know what was asked, what the system returned, who reviewed it and what changed as a result.
What good looks like: AI interactions are logged like any other system activity, with user, timestamp, query, response and the downstream action.
5. Who signs?
Automation has a way of becoming authority without anyone deciding it should. When a forecast, a risk flag or a document summary is machine-generated, the workflow still needs a named human approval step. This is not distrust of the technology. It is that accountability has to land on a person, because that is what your board, your auditor and your contract all assume.
What good looks like: AI output enters the same review and approval workflow as any other submission, and the approver is a person with a name.
6. Does it work on your actual records?
There is a wide gap between an assistant that answers questions about documents you upload into a chat window and one that works across the structured data of a live program: commitments, change orders, RFIs, submittals, cost and schedule. The first is a convenience. The second changes how the program runs. Ask which one you are being shown.
What good looks like: a demonstration on real project structures and live records, not a curated document set.
How PMWEB answers these
PMWEB built the Intelligence Control Framework to answer these questions structurally rather than case by case. It is the governance architecture that defines how AI capabilities are designed, deployed and governed inside the platform: what AI can access, how it stays bound by existing permissions, and where human judgment remains in the loop. The guiding principle is short. AI advises, the operator decides.
"The potential of AI in capital project management is enormous, but so is the responsibility. The data that flows through a capital program is sensitive, the stakes of every decision are high, and the organizations accountable for outcomes cannot afford to hand their judgment to a system they do not control."
Vijai Viswanathan, Chief Product and Technology Officer, PMWEB
Learn more about PMWEB AI and the Intelligence Control Framework at pmweb.com/ai.
