Newforma bought Lumi AI to answer questions from your RFI and submittal history. Here's how a PM should test it

Newforma, a project information management vendor, bought AI analytics startup Lumi AI this week so its customers can ask plain-language questions of their RFI, submittal and email history. If your project records sit in Newforma, ask for a trial on a closed job. If they sit in several tools, note that Lumi will not be sold outside Newforma's customer base.

What did Newforma actually announce?

Newforma acquired Toronto-based Lumi AI; terms were not disclosed, per Construction Dive. Lumi co-founder and CTO Tudor Boiangiu joined Newforma as vice president of AI engineering, and the Lumi team keeps building from Toronto.

Newforma's own pitch, as reported by PropTech Connect, is that AI used for project decisions should be explainable and verifiable against source data. It says Lumi's engine asks follow-up questions when a prompt is vague, shows the reasoning behind each answer, and flags data quality problems it finds. Those are company claims. No accuracy figures or benchmarks were published.

Why should a GC or PM care?

The questions PMs actually need answered live in old records: which sub's submittals bounced twice, which RFIs sat past their response date, what was said in the email before a change order. Today that is a person scrolling a log. A tool that answers those questions across jobs would change who can do that research, and how fast.

It also shows where the market is going. Construction Dive notes this follows Procore's purchase of Datagrid, Autodesk's of Rhumbix and Trimble's agreement to buy Document Crunch. The platforms are buying AI teams and keeping them inside their own walls.

What's the catch?

• Closed ecosystem. Newforma CEO Peter Cannone said Lumi won't be available to non-Newforma customers. An answer is only as complete as the data the tool can see.
• Gaps between tools. Cannone himself said the data problem sits in the connections between tools. If half your correspondence is in a different system, a confident answer can still be incomplete.
• No published accuracy. "Shows its reasoning" is not the same as "is right."
• Not live yet. We found no launch date or pricing for the combined product.

How should a project team test it?

1. Pick a closed job where you know the history cold.
2. Ask ten questions with known answers, such as "Which submittals were rejected more than once?" or "Which RFIs went past their due date?"
3. Check each answer against the log. Count wrong and missing answers separately; missing ones are the dangerous kind.
4. Ask one vague question. See whether it asks a follow-up or guesses.
5. Check permissions. Confirm it only shows records a given user could already open. Our recent piece on NetSuite's construction agent covers the same closed-job test for finance data.

What still needs a human?

Deciding what an old RFI answer means for today's condition, whether a rejected submittal was a real problem or a paperwork one, and what to put in a notice letter. The tool finds the record. The PM owns the call.

Takeaway: before any vendor demo, write down your ten hardest "where did we say that" questions from your last finished job. That list is your test, whichever platform you evaluate.