Funding 9 min read

Two PlanGrid Co-Founders Backed Stoa's €3.5M Pre-Seed. This Product Reads the Drawings Rather Than Delivering Them.

Announced 16 November from Helsinki: a €3.5m (approx. $3.9m) pre-seed led by Lifeline Ventures, with Illusian and Wave Ventures joining and angels including former Nokia CEO Pekka Lundmark and PlanGrid co-founders Tracy Young and Ralph Gootee. The release says the platform runs on tens of Nordic sites and saves project engineers and site managers an average of 2-8 hours a week — and it footnotes the two industry statistics it leans on with named sources, dates and a sample size.

A worker in a white hard hat, seen from behind against a clear blue sky, holding up a folded paper construction drawing marked up in red.

Stoa, a Helsinki company, announced a €3.5 million pre-seed round on 16 November led by Lifeline Ventures. The company’s own release, datelined Helsinki and dated 16 November, puts the figure in dollars itself — “€3.5 million [approx. $3.9 million]” — and describes the money as going toward bringing AI “to data centers and other complex construction sites.”

Lifeline Ventures, which describes itself on its own site as “a sector agnostic early-stage (pre-seed and seed) investor,” was joined by Illusian — the investment company of Supercell co-founder and CEO Ilkka Paananen and Wolt co-founder Miki Kuusi — along with Wave Ventures and what the release calls “several international investors.” The named angels are former Nokia CEO Pekka Lundmark; Alexander Ehrnrooth, CEO of the Finnish industrial investment company Virala; Tuomas Kokkila, a board member of the Finnish construction group SRV; and, from outside Finland, Tracy Young and Ralph Gootee, co-founders of PlanGrid.

That last pair is the most interesting line in the document, and it is worth sitting with. PlanGrid’s contribution to construction was getting the drawing set onto a tablet in the field — access, revision control, the end of the paper roll in the truck. Autodesk bought the company for $875 million in 2018, a figure the release states itself. Stoa’s pitch is the next move on the same board: not delivering the drawing to the person, but answering the person’s question out of the drawing. Whether that is the natural successor to PlanGrid or merely adjacent to it is a question the product has to settle. Two of the five people who built the first one — Young and Gootee; the company also counts Ryan Sutton-Gee, Antoine Hersen and Kenneth Stone among its founders — have put money on the second.

What the Product Does

Stoa’s own site describes a system that “reads the drawings, specifications, contracts, schedule and site records your project already holds, and answers from them, with the document and page behind every sentence.” Three design choices in that description are worth naming, because each is a decision other vendors in this category have made differently.

The first is read-only ingestion. The site says Stoa runs a “[r]ead-only sync from M-Files, SharePoint, Dalux, Congrid and the rest, keeping the permissions each one already set.” Permission inheritance rather than a second access model is, at this point, the settled answer across this publication’s coverage rather than a differentiator — CMiC made it the headline feature of the MCP server under its ERP, and ProjectTeam’s MCP server ships the same answer in its security list — and Stoa adopting it is unremarkable in the best sense.

The second is citation. Every sentence is attached to its source document and page, and the site describes clicking a wall type or door tag in a drawing to open “the detail or clause that governs it” beside the plan. This is the same discipline qapture published as a hard limit for its qentin agent — “What is not in the data, qentin does not answer.” Stoa’s version is positive rather than prohibitive, but the mechanism is the same: the answer carries its evidence.

The third is the “skill.” The site says a finding can be exported as a report, assigned as a task, or saved as “the way you handled it,” reusable later. That is the step from a question-answering tool to something closer to a workflow product, and it is the claim in Stoa’s material that would be hardest for an outsider to evaluate from the outside.

Stoa also publishes commitments on handling: setup “[l]ive in two working days,” and data “[s]tored and processed in EU regions, never used to train any model, deleted within 30 days of your request.” For a company whose stated wedge is data-centre construction for global technology firms, those are the terms the buyer’s procurement team will read first.

The Footnote Is the Unusual Part

Funding releases lean on industry statistics as a matter of routine, and the statistics usually arrive unsourced. Stoa’s does not. It cites two: an IDC survey finding that “three out of four construction project owners go over budget, and just as many deliver projects late,” and a McKinsey finding that “construction productivity grew by only 10 percent between 2000 and 2022, compared with 90 percent in manufacturing.” Then it publishes a Sources block naming the McKinsey article and its August 2024 date, and the IDC report with its year, the fact that the survey was commissioned by Procore, and the sample size — “n = 505 project owners in the US and Canada.”

This publication is reporting those two figures as the release’s citations rather than as independently confirmed numbers, and the distinction is worth stating plainly. The Procore page the release points to resolves and confirms the collaboration — “Procore worked with IDC on a survey to see how high-performing and lagging construction owners were using technology” — but that page itself publishes neither the over-budget figure nor the sample size. The McKinsey page did not respond to automated requests at all during the checks for this article, which is ordinary behaviour for that domain and says nothing about the page’s contents. So: the sourcing is there, it is checkable by a human with a few minutes, and we have not closed the loop on either number ourselves.

The same habit shows up on the company’s own metrics. The homepage publishes three figures — “2-8 h a week back per project engineer, most of it time spent finding where something was written down,” “40% shorter from problem raised to decision made,” and “€4B+ of project value running on Stoa, including some of the largest data-centre sites in the Nordics” — and then footnotes all three: “From design-partner interviews, July 2026, and Stoa’s current portfolio.” A vendor metric with a stated method and a date on it is still a vendor metric. It is a considerably more useful one than a vendor metric without them, and the 2-8 hour figure also appears in the release in the same form, attached to “[p]roject engineers and site managers.”

The Data-Centre Wedge

The release’s framing is explicitly national and explicitly about data centres: “[t]he billions of euros in data center investment flowing into Finland are creating demand not only for land and power, but also for homegrown technology.” The operational argument underneath it is a schedule argument. Billion-euro facilities, the release says, “are typically built in 18 to 24 months, so information has to move across the site in real time.”

That is the right place to point a document-comprehension product, and the reason has little to do with AI. Data-centre and semiconductor work is where construction’s schedule risk has concentrated, which is why Buildots’ $130m round was argued on its customer list rather than its technology — Intel, Digital Realty, Turner, JE Dunn, HOCHTIEF and Bouygues are the firms building those facilities. An eighteen-to-twenty-four-month programme on a billion-euro building compresses the penalty for not knowing which revision governs from an annoyance into a cost.

Stoa’s commercial argument from there is about distribution rather than product: “[b]ecause the technology companies building data centers operate globally, a single customer relationship can open several markets for Stoa.” Its stated targets are the Nordics, the UK and the US. That logic is sound and also well understood by everyone else selling into the same buyer, so it is a reason the market is attractive rather than a reason Stoa wins it.

The company’s About section calls Stoa “a Nordic AI company building an operating system for construction projects.” That phrase is now doing a great deal of work across this category — Shackleton raised $5m on 4 November and reached for the same framing — and it tells a reader less each time it appears. The specific sentences in Stoa’s material are more informative than the category label it has chosen for itself.

What the Record Shows, and What It Doesn’t

On traction the documents are consistent but loosely bounded. The release says the platform “is already in use on tens of sites across the Nordics, including data center projects”; its About section says “more than 20 sites”; the homepage says “more than twenty Nordic sites, from data halls to rail.” Some secondary coverage has written it as more than 25. Twenty-plus is what the company’s own pages support, and that is what this article reports.

Named customers are thinner. The release says customers “include several of Finland’s largest construction and real estate companies, as well as some of the largest data center projects in the Nordics,” without naming any of them. The one named customer voice in the document is Pat Nicholson, whom the release identifies as Nordics Director at GagaMuller — an Irish project and cost management consultancy that works mainly for data centre owners and which opened its Helsinki office in 2025. His quote is about adoption friction rather than output: “[b]ringing something new into a live site always causes some apprehension, but as Stoa works through the portals we already use, it quickly becomes second nature.” That is a usefully specific thing for a customer to say, and it is also not a productivity number.

Neither the release nor the secondary coverage checked for this article gives a valuation. The release states that “[t]he round is one of the largest pre-seed rounds ever raised in Finland” — a company claim, reported here as one; it is not a figure this publication has ranked. Tech Funding News reports that CEO Johannes Korpela says the round closed in under a week and that the team is close to ten full-time people; both are that outlet’s reporting rather than the release’s.

The founding team is named in full: Korpela as CEO, with Camilla Komulainen, Taavi Paananen, Eino Miettinen and Pyry Pohjanoksa. Paananen is co-founder and chief product officer, and the release says he was named to the Forbes 30 Under 30 list this year. The release also states that team members have previously sold an AI company, raised a venture fund of around €10 million, and helped found Supercell’s AI innovation lab, with backgrounds at Supercell, AWS, SRV and Peikko Group. It does not say which member did which, and this article does not either.

What to Watch

The number that would settle the most is a named Nordic data-centre operator or main contractor confirming the 2-8 hour figure in its own voice, on its own project. Stoa has done the harder half of that already by publishing a method and a date for the figure, which is more than funding releases of this size usually do. The remaining half is the half that only a customer can supply.

The second thing to watch is whether read-only holds. A product that answers from the record and cites its page is a low-risk purchase precisely because it cannot change anything; the commercial pressure in this category runs steadily toward writing back — raising the RFI, moving the date, closing the inspection. Stoa’s “skills” feature is the first step along that path. Every company in this publication’s coverage that has taken it has, so far as we have found, landed on the same answer: a human in front of anything consequential. The interesting question for a company this young is whether it gets to choose its own timing.