Artificial intelligence (AI) can perform at a high level on specialized academic tests and some tests of creative thinking.
Some AI systems can even be mistaken for humans in conversational tests. But convincing output is not the same as reliable output. AI can still fabricate information and make serious errors. By September 2026, a global database maintained by legal researcher Damien Charlotin had identified more than 2,000 legal cases addressing confirmed or suspected AI hallucinations, with outcomes in some cases including financial sanctions and disciplinary referrals.
In construction, those errors can carry substantial consequences. Procurement documents define what vendors are asked to price and deliver, so an inaccurate specification can become a commercial problem rather than just an incorrect answer.
McKinsey’s recent assessment of AI in architecture, engineering, and construction (AEC) argues that the opportunity lies in redesigning end-to-end workflows, not simply adding isolated AI tools.
But putting AI to work safely remains a challenge.
Rob Massoudi had been observing weaknesses in construction procurement since 2008. In 2022, he began formal research into a better approach. During that work, an early AI drafting tool specified a roofing membrane, complete with a model number, that did not exist.
Had the error gone unnoticed, the fabricated specification could have passed from the scope of work into a bid and signed contract, leaving the project team with a requirement no supplier could fulfill. Correcting it at that stage could delay the work or trigger disagreements over substitutions, costs, and responsibility. For Massoudi, the experience made the risk clear: an AI hallucination in construction is not just “a quirky bug.” It can become a change order, a claim, or a lawsuit.
That experience helped shape plnd (pronounced “planned”), the construction procurement and capital-planning platform he founded for property owners and operators.
plnd’s core procurement workflow governs the work that happens before construction begins, from defining the scope and issuing bid requests to comparing proposals. The same project record carries the resulting information forward into future projects and capital planning.
Rather than treat each project as an isolated document-writing task, plnd connects a structured property record with a classification of construction work, version-controlled construction knowledge, and a policy engine that determines which requirements apply. Scope requirements, bid line items, supporting documents, approvals, and revisions remain linked to the same project record. AI performs defined tasks within this architecture rather than deciding the procurement process itself.
The workflow is deterministic: the same approved inputs and versioned requirements trigger the same controls for package assembly, review, approval, and release. In plnd, construction knowledge is treated as governed data rather than prose buried inside templates.
A Governed Approach to AI-Generated Procurement
plnd is built for property owners and operators who need greater consistency and control over construction procurement, whether they manage a single property or a large portfolio.
When procurement documents are assembled from individual templates and project managers’ own records, similar jobs can reach the market with different requirements and bid structures. Useful information from past projects may remain scattered across files and spreadsheets.
The problem is not a lack of construction expertise. It is the lack of a consistent way to capture and apply that expertise from one project to the next.
Using a property’s records, project requirements, and the owner’s standards, plnd generates a coordinated scope of work, bill of quantities, and request for proposals in about five minutes. Alongside those documents, the project manager receives a project-specific list of risks and potential change-order triggers. This gives the project manager an opportunity to investigate uncertainties and address potential issues before vendors price the work. Readiness checks flag missing information and requirements, and a human reviewer must approve the package before it is issued to vendors.
Speed is only part of the value. The coordinated package structures both what vendors are asked to price and how their proposals are compared.
Once approved, the bid request is issued to vendors through access-controlled links to the released package. Proposals and bid breakdowns return directly to plnd, where the platform analyzes the submissions against the issued bid structure and organizes them into a line-by-line tabulation for the project manager’s review. Bidders’ questions and clarifications stay attached to the same package, keeping the scope, proposals, and review record together.
“The work you put into one project should make the next one better,” Massoudi said. “plnd builds on the property information, standards, and corrections you provide, so it becomes better informed about your properties and requirements as you use it. You’re not starting over with each project. You’re building knowledge that stays with your business and informs the next scope, bid review, or capital plan.”
Each customer’s project records and knowledge base are kept separate from those of other customers and protected by access controls. The knowledge accumulated from that customer’s inputs serves their own business; their data is not used to train or improve AI models for other customers or third parties.
plnd also preserves the exact document version issued to vendors. The platform records a cryptographic hash, or digital fingerprint, for each released file, so a later copy can be checked against the issued file to detect changes. Separate distribution and access records show which revision was issued, to whom, and when it was viewed or downloaded. Together, these records help the project team trace what was issued and determine whether a disputed copy matches it.
Field teams use SiteOptix, plnd’s companion mobile app, to capture conditions, photographs, and inspection findings. Those records feed document generation and add to the factual record of each property. That richer property record, together with project history, informs multi-year capital plans.
Completed projects remain searchable, giving owners and project managers access to what was scoped, bid, and spent. The experience gained from past work stays with the business, even when the people who managed it move on.
plnd is betting that in construction, the next phase of AI will be defined not simply by what models can generate, but by whether companies can control, verify, and stand behind what those models produce.
