What builders say when nobody's watching

We're an 85-person electrical contractor and we were bidding two jobs a week because that's what our estimating team could handle. I was ready to hire two more seniors at $140K each. Instead we brought in Pilrs, gave it to our existing team, and now we're bidding seven a week. The surprise wasn't speed — we expected speed. The surprise was that our hit rate went up about four points because we stopped submitting tired numbers on Friday afternoons when our estimators had been staring at plans for fifty hours straight.

The NEC 2023 logic was the convincer for me. I'm old enough that I don't trust software that doesn't cite its sources. Pilrs flags which article it's applying for ampacity correction, conduit fill, and demand factors, and it caught an ampacity correction on a healthcare job that our senior would have caught — but in 90 seconds instead of two hours of table lookups. When I can verify the AI's reasoning against Chapter 9 in front of me, I'll trust the output. That was the shift.

My biggest objection going in was that sheet metal is a weight game, not a count game, and I didn't think an AI could handle SMACNA gauge logic cleanly. I gave Pilrs a 220,000 sq ft distribution-center HVAC package and a stack of our historical takeoffs on similar jobs. The weight came in within 1.8% of what my lead estimator independently produced. That's well inside my noise floor. I stopped arguing and put it into our bid pipeline the next Monday.

I run a 32-person concrete outfit. We bid on tight margins — three, four points on a good week — and waste factors are where we bleed. I never trusted the default waste numbers in our old estimating tool because they were averages across types of work we don't do. Pilrs lets me pin waste factors per pour type and per crew, which means our concrete numbers actually reflect the crews running them. First six bids after switching, we picked up an average of 1.1 points of margin on accepted jobs.

Structural steel fab estimating is not a forgiving discipline. Connections are where the hours hide, and they're the part the AI tools I'd tried before just skipped over. Pilrs handled AISC connection logic on a moment-frame job better than I expected — it caught three detail-level items on the shop drawings I would have missed on a first pass, and it flagged them as low-confidence so I could go verify, rather than pretending it had it figured out. The honesty about confidence is what separates this from the last generation of takeoff software.

I came in thinking this would automate my senior estimator out of a job. What it actually did was free him up to spend his time on the parts of the bid where his judgment matters — risk assessment, alternates strategy, buyout planning, sub selection. He still owns the number. Pilrs handles the mechanical part of the takeoff that used to eat 60% of his week. He's happier, we're bidding more, and I haven't laid anyone off. That's the honest story most of your customers aren't going to tell publicly because it was a political concern internally.

Here's what sold me. I ran the same 48,000 sq ft office shell through Pilrs and through my manual process as a test. Pilrs came in at $2.84 per sq ft total. My manual estimate came in at $2.91. The difference was buried in conduit LF where Pilrs had a more accurate path routing because it was working from the CAD file instead of my PDF markup. When the bid came back and we won, the actuals tracked within two percent of Pilrs's quantities. The manual estimate had been carrying a margin of error I didn't even know was there.

Our pain wasn't speed — it was consistency. We had three estimators and three different ways of counting devices, applying labor burden, and handling alternates. Our bid spreads on identical scope between estimators were running eight to twelve percent apart, which meant our numbers didn't really mean anything. Pilrs normalized the takeoff layer. Our three estimators now produce quantities within 1% of each other on the same set. Whatever premium or discount they apply on top is a deliberate judgment call, not an accident of methodology.

I was the skeptic at our firm for about four months. I kept setting traps — feeding in plans I knew had errors, missing schedules, weird scope notes on sheet A-000 — expecting to catch Pilrs being confidently wrong. It caught every trap I laid. The one that got me was a plans addendum we'd uploaded where the spec and drawing conflicted on fixture count. Pilrs flagged the conflict and asked me which to trust. That's not something our old workflow did. That's what made me stop running tests and start running bids.

The onboarding is what sold my team. I'd been burned by two prior estimating tools where we spent three months loading our assembly database and never fully got there. Pilrs pulled our historical data in a week, mapped our cost codes to theirs without me having to hand-hold it, and had a test bid going within ten days. I put two of my apprentice estimators on it first as a low-stakes trial. Both said it's the first piece of software they've used that didn't fight them. That's what adoption looks like.

Government work is unforgiving on documentation. We bid federal GSA and VA projects, and the audit trail on our estimates has to survive a procurement officer asking questions three years after award. Pilrs logs which revision of which drawing was used for which quantity, which addendum was incorporated, and which sections of the NEC were applied. That level of traceability was the difference between a tool we could pilot and a tool we could institutionalize. Our compliance officer signed off faster than anyone expected.

Small GC perspective here — we're a design-build firm with one in-house estimator (me) handling concrete, masonry, and rough carpentry. Pilrs let me bid three projects in the time it used to take me to bid one. That's not a marketing number, that's what actually happened on my calendar. The change for a solo estimator is existential: the work I couldn't physically do before — going after jobs I would've turned down for lack of capacity — I can actually chase now. It changed what our company can go after.