The Procurement Middleman Was Always Going to Get Automated. I Just Didn't Expect It to Be Me.

A fair chunk of my career has involved doing something a computer will eventually do faster than me: reading a set of drawings, working out what packages they break into, and writing a scope of works that tells a subcontractor exactly what they're pricing. It's skilled work. It's also, if I'm honest with myself, mechanical enough in places that I can see exactly how it gets automated — because I've spent enough evenings thinking about how to automate it myself.

That's an odd thing to admit as someone who's built a chunk of his career on doing that job well. But I think pretending otherwise is worse.

What the job actually is, underneath the title

Quantity surveying gets described as if it's one skill. It isn't. It's judgment — knowing whether a valuation's fair, whether a variation's justified, whether a subcontractor's claim holds up — bolted onto a huge amount of repetitive, structured extraction work. Reading a drawing and turning it into a measured scope. Cross-checking that scope against a spec. Flagging where the drawing and the spec disagree. That second half isn't judgment. It's pattern matching against a large, consistent rule set. Which is exactly the kind of task that's been getting automated across every industry for the last five years — construction was just slower to get there, mostly because our drawings and our documentation have historically been such a mess.

Why I don't think this is the bit worth being precious about

The value I actually bring a client isn't that I can read a drawing faster than software eventually will. It's that I know when a scope of works is going to cause a dispute eighteen months later, because I've seen that exact ambiguity cause a dispute before. It's that I can look at a valuation and know something's off before I can even articulate why. That's experience, not extraction. No amount of AI reading drawings replaces having sat across the table from a subcontractor arguing a variation, or having watched a poorly worded scope turn into a six-figure claim.

What AI is good at — genuinely, uncomfortably good at, if you've tried it — is the part of the job that was never really the interesting part. Reading a drawing set and flagging where the scope has gaps. Cross-referencing a spec against a subcontract package. Surfacing the value engineering options a busy QS doesn't have time to fully explore on every single package. That's not replacing judgment. It's freeing up the hours that used to go into extraction so more of them go into judgment instead.

Why I think the system needed correcting anyway

Construction procurement has run on a genuinely inefficient structure for a long time — multiple layers between the person who needs the work done and the person who does it, each one adding cost and time without necessarily adding value. Some of those layers exist for good reasons: risk transfer, cash flow buffering, specialist knowledge. Some of them exist because that's just how it's always been done, and nobody's had a strong enough reason to change it.

I don't think AI in procurement kills the middleman role. I think it kills the version of the middleman role that was only ever justified by how slow and opaque the process used to be. If your value was purely "I know how to read this drawing and you don't," that's a genuinely vulnerable position to be in over the next few years. If your value is judgment, relationships and knowing where the risk actually sits on a project — that gets more valuable, not less, once the mechanical extraction work stops eating your week.

That's the bet I've made, building something in this space rather than just watching it happen. I'd rather be the person who helped build the correction than the person explaining, in a few years, why the old way was worth defending.