Volve has added an orchestration layer to its tendering platform, automating tasks that span calculations, estimating and the interpretation of drawings and BIM models. AEC Magazine reported the change on 6 October, describing a system that accepts an instruction in natural language, works out which steps are needed, runs the checks and returns findings with source attribution.
The example in the reporting is concrete enough to be worth quoting as the whole product thesis: a user can ask the system to “check that the quantities in the specification match the drawings and the IFC model.” What the orchestration layer contributes is that nobody has to specify how.
The Difference Between Reading and Reconciling
Construction AI has spent the past few years getting good at reading one document. Extract the obligations from a contract, find the deviations in a specification, take off quantities from a drawing set — these are now crowded product categories with real customers.
Reconciling several documents against each other is a harder and more valuable problem, and it is the one this release claims to address. A specification, a drawing set and an IFC model are three descriptions of the same building, produced by different people at different times, and construction’s most expensive mistakes live in the gaps between them. Finding those gaps today means opening each document and comparing by hand, which is the manual effort AEC Magazine identifies as the target.
The reason that work goes undone is not that it is difficult to understand. It is that it is unbounded. Checking every quantity in a specification against both a drawing set and a model is a task whose cost scales with the size of the document set, and a bid team with a deadline will check the line items it suspects rather than all of them. The orchestration framing matters because it changes what the user has to know: instead of selecting a check from a menu, the user states the outcome, and the system decomposes it.
Herman B. Smith, Volve’s co-founder and chief executive, is quoted in AEC Magazine on what made it possible:
“With orchestration on top of the project graph structure we already had, we are seeing completely new possibilities, sooner than we had planned for.”
That is a claim about sequence, and it is the most interesting sentence in the announcement. The orchestration layer is presented as something the underlying data structure enabled, not as a feature bolted on top. Neither the word “orchestration” nor the phrase “project graph” appears anywhere on Volve’s own homepage, which was checked for both — so the framing in this piece comes from AEC Magazine’s reporting and the chief executive’s quote there, not from the company’s published product material.
What the Company Already Said It Had
The “structure we already had” is documented, and it is consistent with the claim.
Volve’s homepage describes a platform “Built on contract and document hierarchy logic to model obligations, precedence, time bars, and risk. With traceable sources for review and sign-off.” Its advertised outputs are “Coverage checks, risk and deviation flags, scope control, bid development, and a baseline that carries into execution,” with support for “large, complex document sets across PDFs, Word, and Excel.” The homepage also already describes “construction-specific agents and checks” that a team can use predefined or extend.
Two things in that description do load-bearing work for the orchestration story. The first is precedence: contract documents are not a flat pile, they are a hierarchy in which one document governs another when they conflict, and a system that models that hierarchy can decide which answer wins rather than just reporting a disagreement. The second is traceability — the site says “Every answer and finding links back to the source text.” A multi-step automated check is only usable in a commercial process if its conclusions can be audited back to a document, because the finding will eventually be argued about with a client or a subcontractor.
That traceability requirement is also why this is harder than it looks. An orchestration layer that chains several inference steps together accumulates error at every hop, and the output of step three is only as good as the extraction in step one. Attaching sources to the findings is the mitigation, and it puts a human back in the loop at exactly the point where that is cheap.
The Company, and the Round Behind It
The company trades as Volve Technologies AS, per the copyright line in its own site footer. It was founded in 2024 by Herman Smith, Abyl Ikshanov and Alf Jørgen Dovland and is headquartered in Oslo, according to Tech.eu’s report on its seed round.
That round was announced shortly before this launch. Volve announced on 25 August a seed of USD 3 million (NOK 30 million), led by Skyfall Ventures, with participation from J12, Norrsken Evolve, OBOS Ventures, Antler and StartupLab. Daniel Kjørberg Siraj, former chief executive of the Norwegian housing developer OBOS, joined as an angel investor. The company said the money would deepen the product for contractors and public clients, take it into more European markets, and “build out the connected data structure beneath it.”
Six weeks between a seed announcement and a capability launch of this kind usually means the work predates the round, and the chief executive’s “sooner than we had planned for” reads the same way. On the company’s own account the round funded the data structure; the orchestration layer is what that structure appears to have made possible.
Worth noting on the investor list: OBOS Ventures is the corporate venture arm of OBOS, a Norwegian housing cooperative and developer, which makes at least one investor a participant in the industry Volve sells into. That is common in construction technology and is not a conflict, but it is the kind of relationship worth keeping in view when reading a vendor’s customer claims.
The Traction and the Figures, as the Company States Them
Volve’s own numbers, from the same seed-round post and reported here with attribution rather than audited: “Volve has been used on more than 2,500 construction projects with leading contractors across the Nordics. On one recent public infrastructure tender comprising 213 documents, Volve surfaced 20 risk findings and supported the comparison of 73 subcontractor bids.”
The single-tender figures are the more informative of the two, because they describe a shape rather than a volume. Two hundred and thirteen documents is, on any reading, a great deal to get through on a bid deadline; 73 subcontractor bids is the part of tendering that is pure comparison work and the part most likely to be done on a spreadsheet at speed. Twenty risk findings is the company’s count on one job, and the company’s post offers no baseline for what a human team would have found, so it is a description of output and not a measure of improvement.
The company also publishes three outcome claims under the heading “Proven to deliver results in construction tendering”: “60% more time on early mitigation and optimisation”, “100% bid coverage and stronger submissions”, and “30% less rework from clearer scope and early visibility.” These are the company’s figures, presented without a stated methodology or sample, and they are reported as its claims.
The site carries a single named testimonial: “Volve addresses a challenge we face in every project. From day one, we’ve seen real value, both in time saved and stronger decision support.” It is attributed to Christoffer Hernæs, described on Volve’s site as “CDO (at the time), Skanska” — the qualifier is the company’s own, and is reproduced here because it changes how the endorsement should be read.
Why the Tender Phase Keeps Attracting the Money
Volve’s argument for its own market, also from the seed-round post, is the one most preconstruction companies make, and it is worth stating because it is largely correct: “Around 80 percent of project cost is locked in before the contract is signed, and the decisions made in that first stage set everything that follows.” The company puts European construction investment at “more than EUR 1.6 trillion” annually and says the sector accounts for “nearly one in five public tender notices,” while describing tendering as the phase “where digitisation has advanced least.”
That reasoning has drawn a remarkable concentration of capital in the past few weeks alone. Quotr raised $4M on the bids contractors never get around to submitting; Kuro raised €10M on twelve customers to check tender documents against a contractor’s own rule book; LightTable raised $22M to catch construction’s costliest mistakes before anyone breaks ground. Volve’s $3M is the smallest of those, and its differentiator on the available evidence is the document-hierarchy modelling underneath — precedence and time bars are contract-administration concepts, not estimating ones.
Analysis: Orchestration Is the Word to Watch, and to Discount
What follows is this publication’s assessment rather than reporting.
A pattern is forming across this sector’s last two months of announcements, and it is not about models getting better. It is about vendors who spent two years building a structured representation of a project now putting an agentic layer on top of it and discovering they can answer questions they never built a feature for. Volve describes exactly that sequence. So, in a different corner of the stack, does the MCP connector work appearing at Arcadis and in what AEC Magazine reports of CupixWorks 5.0.
The discount to apply is that “orchestration” is, for now, a word rather than a benchmark. Nothing in the available reporting says how often the system picks the right chain of steps, what it does when the specification and the model disagree for a legitimate reason, or how a bid team verifies a finding it did not ask for. Those are the questions that decide whether a multi-step check is used on a live tender or quietly abandoned after the first confident wrong answer — and a plain-language instruction is precisely the interface that makes a wrong answer hardest to notice, because the user never saw the steps.
The structural reason to take it seriously anyway is that the traceability requirement is already in the product. A system that must link every finding to source text is constrained in a useful way: it cannot present a conclusion it cannot point at. If orchestration in this sector turns out to work, it will probably be at the companies that built the audit trail before they built the agent, and on current evidence that is the order Volve did it in.
