Dundir

AI-Powered Construction Procurement

2026-07-23

Every AI procurement tool in the world runs in the cloud.

Dutch municipalities are legally prohibited from using cloud tools for procurement data.

That is a €4.7M annual market with zero compliant solutions.

Don’t introduce yourself yet. Stand still. Let them read the three lines.

Pause for three seconds after they’ve read the third line.

Then say: “That’s not a technical problem. It’s a structural gap. And we’re the only team building inside it.”

Only then introduce yourself and move on.

“Construction procurement managers make million-euro decisions from Excel spreadsheets.”

• Procurement teams lose real time every week to manual comparison work: emails, PDFs, supplier portals, and spreadsheets
• Every supplier choice is a documentation liability: most decisions cannot be reconstructed if challenged
• The AI tools that exist for this? Every single one is cloud-based. A Dutch municipality cannot legally use a single one of them (Rijksoverheid 2024)

The cost of the status quo

Manual comparison work is one of the largest hidden line items in a procurement team’s week - and none of it shows up as a cost anywhere in the budget.

“I’ve sat with procurement managers at construction companies across the Netherlands. The story is always the same: three supplier emails, an Excel sheet they’ve been updating for six years, and a gut call that gets written up after the fact, if anyone ever asks.

For municipalities it’s worse. They’re legally accountable for every euro they spend. One auditor question they can’t answer and there’s a political problem. They know AI would help. They also know they can’t touch it: the moment data leaves their server, they’re exposed under the Aanbestedingswet.”

“A procurement assistant trained on your data. Runs on your server. You own it.”

• Structures procurement decisions: the AI ranks and documents each option, the procurement professional reviews and decides. The human is always in control.
• Runs locally and learns: trained on the client’s own procurement history; improves with every cycle, not less
• Not a chatbot: purpose-built software trained on real procurement data. No invented suppliers, no invented certifications.

Why this framing matters

Public sector clients are legally accountable for every procurement decision. An AI that “generates” an answer is a liability.

An AI that structures and documents a decision the manager makes: that is a compliance tool.

“Think of it as a procurement expert who has read every contract your company has ever signed, knows every supplier you’ve ever worked with, remembers every certification and delivery failure, and structures the comparison for the manager to review.

The manager still decides. The AI does the retrieval, the ranking, the documentation: everything that currently takes three hours in a spreadsheet.

It doesn’t live in our cloud. It lives in theirs. That’s not a limitation: for our clients, that’s the product.”

“Three steps. One report. Full audit trail.”

STEP 1

Describe the need

Material · specification
quantity · delivery window

→

STEP 2

AI ranks & scores

Price · delivery · quality
certifications · history

→

STEP 3

Manager reviews

Checks · adjusts
approves or rejects

→

STEP 4

Decision documented

Audit trail built at
point of decision

The system improves with every decision. The more they use it, the better it knows their suppliers, their preferences, their standards. That’s not a feature: that’s why clients don’t leave.

“The procurement manager types in what they need: the material, specification, quantity, delivery window. The AI searches their supplier history, their current catalogue, current certifications. It returns a ranked list with reasoning attached to every result.

Then it generates a structured report, ready for manager approval, legal sign-off, or an auditor who shows up three years later.

Walk through each box. Emphasise the green box: the audit trail is built AT the point of decision, not reconstructed after the fact. That is the key for municipality clients.”

“Local deployment isn’t a feature. For our clients, it’s a legal requirement.”

Every other AI tool	Dundir
Cloud: data leaves the building	Local: data never leaves the server
Non-compliant for Dutch public sector	Audit-ready by design
Generic model, same for every client	Trained per client on their own data
No switching cost	Deep switching cost: client data stays in the system

The closest AI competitor: Parspec (PR Newswire 2025)

$20M Series A. 4× revenue growth. Hundreds of distributors and sales agents, including four of the five largest US electrical distributors.

Cloud-only. US-focused. No EU presence.

Cannot serve a single Dutch municipality. Does not train per client. No audit trail for public law.

The intersection of local + construction + per-client training + audit trail has no current occupant in Europe.

“Parspec is the best-funded AI procurement tool in the world. They’re doing the closest thing to what we’re building.

They are cloud-only. They cannot serve a single Dutch municipality. They are US-focused and have no EU presence.

The combination of local deployment, per-client training, and construction domain specialisation: that intersection has no occupant. We checked.”

“€710M EU opportunity. We’re starting where we have the unfair advantage.”

€710M EU TAM (Dundir bottom-up estimate)AI procurement, Segments A+B, cross-checked against MarketDataForecast’s €2.92B EU procurement-software TAM (MarketDataForecast 2025)

€20M NL SAMA + B

913 NL construction cos. with 50+ employees(Dundir calculation: CBS/Statista size-class data (Statista / CBS StatLine 2023))

342 Dutch municipalities (Rijksoverheid 2024)

0 compliant local AI procurement solutions in NL today

Why NL first: Uniform regulation. English-language business culture. High digital adoption. Local deployment is non-negotiable, not just a preference. One NL municipality reference unlocks the conversation in Belgium, Germany, and the Nordics. The same problem, a different flag.

“We’re not trying to take on the world in year one. The Netherlands is a self-contained proof point.

375 qualifying construction companies. 130 qualifying municipalities. And a legal environment that makes our local deployment story non-negotiable.

Once we have reference clients in the Netherlands, the doors to Belgium, Germany, and the Nordics are the same conversation.”

“We deliver. They own. We’re available when needed.”

	Construction	Municipality
Implementation (one-time)	€55,000	€58,000
Annual maintenance (optional)	€8,000	€9,000
Retraining engagement	€12,000	€12,500

No subscription. No platform dependency. No ongoing permission to use.

The client owns the trained model, the integration, everything.

Revenue per client (5-year LTV)

€139K LTV per client · 5.6× LTV:CAC · <6 month payback

Revenue path

Workshop → Paid pilot → Implementation → Maintenance → Retraining on demand

Workshops generate revenue before the product is built and seed the demo dataset. Revenue from day one.

“This is not SaaS. It’s how this market already buys enterprise software: ERP implementations, IT deployments, consulting engagements. We fit into that procurement pattern naturally.

The maintenance contract and retraining are optional. Both are genuinely valuable. Neither is a condition for the system to function. That’s important: our clients are not locked into paying us to keep using what we sold them.

What keeps us in the relationship is that the trained model holds their own procurement data. Leaving means losing their accumulated history. That’s not artificial lock-in: it’s structural.”

“Profitable from Year 1. €192K contracted maintenance base by Year 3. No large equity round required.”

	Year 1	Year 2	Year 3
Active clients	5	14	23
Contracted maintenance base	€42K/yr	€117K/yr	€192K/yr
Total revenue	€321K	€697K	€1.02M
Total costs	€251K	€549K	€719K
Net result	+€70K	+€148K	+€298K
Cash position	€120K	€269K	€566K

{.total-row}

Base: €50K founders’ working capital. No external investment assumed. Year 1 costs include founder salaries at 40–60% FTE (bridged by existing income).

Non-dilutive grant pipeline

Grant	Amount	Deadline
MIT R&D AI	up to €200K	⚠️ May 26
KIEM Arbeidsbesparende AI	€40K	Sep 2026
RAAK-PRO	€700K/4yr	Late 2026

Base case is profitable without the grant. MIT R&D triples Year 1 closing cash to €195K.

“This model is profitable from Year 1, not Year 2. The reason is simple: implementation fees are front-loaded. We earn the bulk of our revenue per client when we deliver the system, not spread over a subscription.

The contracted maintenance base, €42K/year from Year 1 clients, growing to €192K/year by Year 3, is the recurring revenue floor. It grows with every implementation, even if we sign no new clients in a given quarter.

This business does not need a large equity round to reach sustainability. We’re here because the right investor at this stage accelerates the EU expansion timeline, not because we need to keep the lights on.”

“The university anchor. The AI depth. The operations track record. The municipality relationships.”

Christiaan Verhoef, CEO Project Manager, Supply Chain Finance · Windesheim University of Applied Sciences Founded Value Chain Hackers AI lab. Co-founded 10 startups. Helped raise €2M. Windesheim anchor for grant access.

Dr. Milan Jelisavčić, CTO PhD Evolutionary Robotics, VU Amsterdam · Head of AI @ Salesteq Production ML at Stedin, ABN AMRO, bliq. Solved real-time energy optimisation at grid scale. Built the decision engine.

Kirsten Coppoolse, COO COO @ Open Food Chain 2018–2024 First employee to 20-person team. Doubled revenue every year for 6 years. Has built a company before.

Dr. Nina Gluhović, Domain Expert / Research Assistant Professor · Faculty of Civil Engineering, University of Belgrade PhD Structural Engineering. 12 years in construction academia. Knows CROW RAW, UAV, CPR 2024, KOMO, SKG-IKOB. The reason the AI understands a reinforced concrete specification.

Gerard Tunteler, Head of Government Sales Local Government Lead Netherlands @ HPE · 19 years Founded HPE Roundtable for Municipalities. Direct relationships across all 342 Dutch municipalities. Has sold enterprise software to every municipality we’re targeting.

Institutional anchors: Windesheim University of Applied Sciences (NL) · University of Belgrade, Faculty of Civil Engineering (RS)

Named prospect: ED. Züblin AG (STRABAG), insider contact Dusko Stojanovic, Project Lead Bid Processing, Stuttgart.

“This is not a team that needs to find the market. Christiaan is the Windesheim connection: we walk into grants with institutional credibility from day one. Milan has shipped production AI at Dutch infrastructure scale. Kirsten has built the ops system before. Nina knows construction procurement from the inside: she’s the reason the AI understands a reinforced concrete specification.

And Gerard has already sold software to every municipality we’re targeting. He knows the procurement officer, the alderman, and which approach works. That’s not a network you build: that’s 19 years of trust.”

“We’re raising €250K pre-seed to own the NL market before Parspec arrives.”

€250,000 pre-seed

Allocation	Use
60%	Founder salaries: full-time from day one
25%	AI infrastructure: GPU, training pipeline, deployment stack
15%	Go-to-market: workshops, pilots, industry events

What €250K unlocks vs. grant-only path

	€250K + grant	Grant only
Year 1 clients	8+	5
First municipality ref.	Q3 2026	Q1 2027
EU expansion	Year 2	Year 3

We don’t need this money to survive. The grant track funds Year 1 either way.

We want this money because Parspec raised $20M and will eventually look at Europe. The window to establish NL reference clients is open now. It closes the moment a well-funded US competitor decides the EU market is next.

Grant track funds Year 1 either way. We want this money because Parspec raised $20M and will eventually look at Europe. The window to establish NL reference clients is open now: it closes the moment a well-funded US competitor decides the EU market is next.

Parspec proved the market exists. The question is who owns the EU version. We think that should be us. We’re looking for a partner who sees that, not just a cheque.

Appendix A: Competitive Matrix

	SAP / Coupa	Procore	Parspec	Onventis	Mercell	Dundir
Local / on-premise	✗	✗	✗	✗	✗	✅
Construction domain	✗	✅	✅	✗	✗	✅
AI specification matching	✗	✗	✅	✗	✗	✅
Trained per client	✗	✗	✗	✗	✗	✅
Explainable / audit trail	✗	✗	✗	✗	✗	✅
Dutch / EU market focus	⚠️	⚠️	✗	✅	✅	✅
Municipal compliance	⚠️	✗	✗	✗	✅	✅
Data sovereignty	✗	✗	✗	✗	✗	✅

The combination of all eight criteria exists only in Dundir.

Parspec is the closest AI competitor: cloud-only, US-centric, no EU presence (PR Newswire 2025). Onventis is the most likely name in a Dutch shortlist: Benelux salesforce, but cloud-only and horizontally generic. Mercell manages the process; we provide the AI decision layer. Complementary, not competing.

Appendix B: Grant Timeline

%%{init: {'theme': 'base', 'themeVariables': {
  'primaryColor': '#1b4fd8', 'primaryTextColor': '#fff',
  'primaryBorderColor': '#0d2b7a', 'lineColor': '#5a5a5a', 'fontSize': '16px'
}}}%%
gantt
    dateFormat  YYYY-MM
    axisFormat  %b %Y
    section Grants
    MIT R&D AI (⚠️ Apply now, up to €200K)    :crit, 2026-05, 2026-08
    KIEM Arbeidsbesparende AI (€40K)           :2026-09, 2027-01
    RAAK-PRO preparation                       :2026-09, 2027-06
    RAAK-PRO application (€700K/4yr)           :2027-01, 2027-06
    Horizon Europe Cluster 4 (€2–5M)          :2027-06, 2028-06
    section Commercial milestones
    First 5 clients + pilots                   :2026-03, 2026-12
    First municipality reference               :2026-06, 2026-12
    Full-time team                             :2027-01, 2027-12
    EU expansion begins (BE, DACH)             :2027-06, 2028-06

Appendix C: Key Assumptions

Assumption	Impact	Validation
5 clients Year 1	HIGH	First LOIs by Q2
Blended fee €55K / €58K	HIGH	First pilot pricing
60–70% retraining rate Year 2+	HIGH	Track with Year 1 clients
Founders at partial FTE Year 1	HIGH	Co-founder agreement
MIT R&D AI granted	MEDIUM	Submit before May 26
5% annual churn	MEDIUM	Trained model retention
4 workshops × €10K Year 1	MEDIUM	Pilot with Windesheim network

Conservative scenario (3 clients Year 1)

	Year 1	Year 2
Revenue	€208K	€430K
Costs	€251K	€549K
Net result	-€43K	-€119K

The conservative case requires the MIT R&D grant or an angel bridge to remain cash-positive. This is why the May 26 deadline is the single highest-leverage action right now.

References

MarketDataForecast. 2025. “Europe Procurement Software Market: Size, Share, Forecast.” https://www.marketdataforecast.com/market-reports/europe-procurement-software-market.

PR Newswire. 2025. “Parspec Raises $20 Million Series a to Modernize the Construction Supply Chain with AI.” https://www.prnewswire.com/news-releases/parspec-raises-20-million-series-a-to-modernize-the-construction-supply-chain-with-ai-302498075.html.

Rijksoverheid. 2024. “Aanbestedingswet 2012.” https://wetten.overheid.nl/BWBR0032203.

Statista / CBS StatLine. 2023. “Netherlands: Number of Construction Enterprises by Size Class (Table 81589NED).” Statista. https://www.statista.com/statistics/431961/number-enterprises-construction-industry-netherlands/.