Newforma announced on 21 October that it has acquired Lumi AI, a Toronto company whose analytics engine, in the words of the release, “turns plain-language questions into accurate, explainable answers.” Terms of the transaction were not disclosed.
On that non-disclosure: the release is silent, and we could find no reported price anywhere else. Newforma is privately held — an Ethos Capital company, per its own boilerplate — so unlike a public acquirer it files no quarterly report that might disclose the figure months later, which is how Trimble’s price for Document Crunch eventually surfaced. What the public record does carry is Lumi’s own funding history: a $3.7 million seed round in March 2025 led by AgFunder, with Forum Ventures, ADQ and the Qora71 syndicate participating. The release dates the company’s founding to 2023 in Toronto, which makes this an exit roughly three years in.
What Was Actually Bought
Read Lumi’s own site and the first thing that stands out is how little of it is about construction. The headline is “Auto-analyze ERPs and Databases to Reveal Untapped ROI,” and the product is pitched at supply chain, distribution and customer operations. The release’s own description of Lumi’s prior market says it plainly: “Mid- to-large F100 enterprise clients using Lumi uncovered multi-million-dollar opportunities hidden in inventory, sales, and procurement datasets.”
So Newforma has not bought an AECO point solution. It has bought a horizontal enterprise-analytics engine and a team, and the asset is a method rather than a domain. Per the release, Lumi’s engine “builds a semantic layer over the operational systems it reads, so it works from what the data means rather than only where it lives,” and that semantic and clarification layer was “developed over three years.”
The clarification half is the unusual part. The release says the engine “asks clarifying questions when a prompt is vague, shows the logic behind every answer, and proactively flags data quality problems when it finds them.” Tudor Boiangiu, Lumi’s co-founder and CTO — who joins Newforma as VP of AI Engineering, with the Lumi engineering team continuing to build from Toronto — reduces three years of work to one line in the release:
“We spent three years building an engine that asks instead of guessing. That work is worth nothing without data that already means something, and Newforma has 20 years of it.”
That is the acquisition logic stated by the engineer rather than the acquirer, and it is the more useful version. A question-asking engine is only as good as the schema underneath it, and a schema is the thing a horizontal analytics startup cannot buy.
The Schema Is the Dowry
Newforma’s half of the trade is structure. The release says the engine “will initially be integrated into Newforma platforms, drawing on more than 20 years of AECO project data structure, including issues, RFIs, submittals and communication history,” and then makes the point that the structure is semantic, not merely large: “An RFI is not a generic ticket, a submittal cycle has a shape, and a coordination issue means something specific.”
Anyone who has watched a general-purpose model reason about construction documents will recognise why that sentence is in the release. The outputs the company promises depend entirely on it — per the release, Newforma Konekt “will surface which consultant is slowing RFI turnaround, which project’s issue pattern matches one that went wrong before, and where a submittal cycle is drifting from a firm’s own baseline.”
Each of those is a query that is trivial to express and historically expensive to answer. The release is candid about why: “That data answers a question only when someone stops and builds a report,” and “Not all firms have dedicated business intelligence teams.” For a mid-sized architecture practice, the consultant-turnaround question has always been answerable in principle and unanswered in practice, because answering it meant a week of someone’s time.
Lumi’s layer attaches to Vojo, the assistant Newforma introduced at Newforma World 2026 in May, whose flagship bet was that the highest-return place to put AI in AECO is the administrative sludge — document search, submittal review, filing email. Vojo asks; Lumi, per the release, “adds the analytics layer that turns those questions into verifiable answers in Newforma Konekt’s smart project dashboards, with rollout details to follow.” Construction Dive reports CEO Peter Cannone saying Lumi will not be available to non-Newforma customers and that there will be no layoffs at Lumi.
A Standard, Published, With the Company’s Name on It
The most quotable thing in the release is not about the deal. Newforma has set out three tests it says AI used to inform project decisions should meet: answers must be explainable, so the reasoning is visible; verifiable against the source data, so they can be checked; and accurate enough to act on, because on a live project a decision carries liability. Cannone, in the release:
“A lot of companies are telling this industry their AI is better. We’d rather say out loud what better means and be measured against it. Explainable, verifiable, accurate enough to act on when liability is on the line. Our customers don’t need another AI that sounds confident, they need answers they can verify.”
It is marketing, and it is also a hostage. A vendor that publishes the criteria has given every prospect and every competitor a scorecard with its own name at the top, and “accurate enough to act on” is the sort of phrase a customer will quote back during a renewal. Our read is that this is the more interesting half of the announcement, because the acquisition will take quarters to show up in a product and the standard is testable in a demo.
It also lands in a week where the same instinct showed up everywhere. The same day, Buildots gave its assistant a “glass box” feature that explains the assumptions behind a forecasted delay; earlier the same morning, Document Crunch extended contract-risk review to drawing sets with source-linked citations as its load-bearing feature. Confidence has stopped being a selling point in construction AI. Auditability has become one.
”Single View,” Not “Single Source of Truth”
The release spends real space on positioning, and it is positioning against the rest of the sector rather than against Lumi’s absence. Cannone closes his quote on it:
“Everyone in this industry wants to be your single source of truth. We would rather be the single view into all of them.”
The argument, per the release, is that project information already sits across “design tools, CA trackers, document systems, email, Teams messages and the field,” and that asking a firm to consolidate into one platform asks it “to move its work into one place and leave the rest behind.”
That is a coherent strategy for a company of Newforma’s size, and it is the opposite of what the sector’s major platforms have been buying. The consolidation thread in our archive runs the other way: Trimble embedding Document Crunch’s contract AI into the core of its suite, Procore paying approximately $845 million for DroneDeploy’s reality capture, Autodesk’s $3.6 billion MaintainX deal reaching past design into operations. Those are platforms widening the set of work that happens inside them. Newforma is betting that for the firms it serves — architecture and engineering practices with a long tail of vendors they are not going to abandon — the valuable position is the reading layer rather than the system of record.
Which is also, bluntly, the position available to a company that is not going to win a consolidation race against Trimble and Autodesk. Strategy and constraint point the same way here, and that does not make the strategy wrong.
What Would Show It Working
The release says the first work is “in workflows where an answer can be checked against the project record the firm already knows to be true,” and that Newforma will widen the scope from there. That is a sensible sequence and a deliberately unfalsifiable schedule; rollout details are explicitly left to follow.
Three things would tell a prospect whether the three-part standard is real. Whether the clarifying questions survive contact with AECO vocabulary, where an ambiguous prompt is the normal case rather than the exception. Whether “verifiable against the source data” means a citation into the actual RFI or submittal record a user can open, rather than a summary of what the system looked at. And whether the cross-system reach the positioning depends on extends to the systems a firm actually uses, including the ones Newforma does not sell. Ibrahim Ashqar, Lumi’s co-founder and CEO, frames the ambition in the release as making data interaction “as simple as asking a question.” The hard part was never the asking.
