Dundir

For construction companies

2026-07-23

Your best procurement manager loses real hours every week to admin.

Not salary. Admin. Comparing quotes in Excel, chasing certificates by email, writing up reports nobody asked to write.

That’s the cost of not having a system.

Audience: procurement director, CFO, or operations director at a mid-size construction company. Let the third line land before moving on - this is the cost of the status quo, stated plainly, no invented number attached.

What we do

• Give it a specification - what you need, where, by when
• Get back a ranked comparison of your suppliers in minutes: certifications verified, reasoning explained, report ready for approval
• Runs on your own server. Your data never goes anywhere.

A procurement AI your team actually uses - not a chatbot bolted onto the spreadsheet you already have

Keep this concrete: specification in, ranked comparison out, on their own infrastructure.

How it works

• Two to four weeks integrating with your ERP and procurement history
• We fine-tune a local AI model on that history - it already knows your suppliers, your preferred materials, your past decisions
• From day one it’s not a generic tool. It’s your institutional knowledge, made searchable
• The longer you use it, the smarter it gets

Emphasize “day one” - this isn’t a blank model, it’s calibrated on their own history before it ever answers a real question.

Are you sure

• McKinsey’s megaproject research finds average cost overruns of roughly 80%, with poor execution - planning, design, project management, and procurement together - as the dominant cause (McKinsey & Company 2026)
• Construction is the second-least digitised industry on earth - most mid-size firms still run on spreadsheets
• When your best buyer leaves, that knowledge walks out the door

What a working prototype should demonstrate

Cutting a procurement cycle down from hours to minutes isn’t a marginal improvement - it’s a different way of working. That’s the bar we’re building to, not a claim about results we don’t have yet.

Note (2026-07-23): do not claim observed pilot results here - none exist yet. The “hours to minutes” framing is stated as a design target, not an observed outcome. Keep it that way until real pilot data exists.

Can you do it

• A Windesheim supply chain researcher who has run a local AI lab on this exact problem for two years
• A production ML engineer who has deployed models at Stedin, ABN AMRO, and bliq
• We’re not building this in theory. The infrastructure exists. The first pilots are the proof.

The ask: a paid pilot - two to four weeks of integration, a working system on your own procurement history, a report ready for your next real supplier decision.

Close on the pilot ask directly - this segment has the fastest sales cycle, don’t bury the ask under more narrative.

References

McKinsey & Company. 2026. “Megaprojects: The Good, the Bad, and the Better.” https://www.mckinsey.com/capabilities/operations/our-insights/megaprojects-the-good-the-bad-and-the-better.