HALDCO

The Operations Map & Three Moves Memo · Worked sample

How a job moves through Meridian Mechanical

Built from 23 closed jobs (Sep 2025 – May 2026), 11 interviews, and a half-day on the floor. June 2026. Claims stay inside that sample. (The default instrument reconstructs 10–20 units; Meridian's Procore discipline made the full closed-job set usable, so we took it.)

Interviewed: the owner, all four project managers, both estimators, the office manager, the controller, the service manager, one field foreman. No individual is quoted anywhere on this page — findings describe the pattern, not the person.

$22M commercial mechanical (HVAC + plumbing) 92 people · owner-led Jobs $80K–$2.5M Procore · QuickBooks Desktop · Excel estimates · Outlook Service division: 6 techs

The Operations Map

One page: the path a job takes through the building, left to right, with the leaks pinned where they occur. Each marker keys to its evidence card below (L1–L4). No score, no grade — the mechanism is the message.

The Six Leaks: Chasing Re-keying Key-person Flying blind Waiting Redoing — four found here at material size. Waiting and Redoing surfaced only inside other leaks' mechanisms (cross-referenced on the cards, never double-counted).

The takeaway: Most of Meridian's drag lives where the field meets the books — costs are committed during the build but land in QuickBooks weeks later, so margin problems surface at closeout instead of at 60% complete. That one gap is worth roughly $96K–176K a year; the four leaks together run roughly $125K–256K.

What we couldn't see: QuickBooks Desktop is batch-entered — weekly for field time, month-end for some sub invoices — so job-cost dates reflect bookkeeping cadence, not when the cost happened. Every "when was this knowable" claim is reconstructed from Procore commitment dates and corroborated in interviews, and the ranges are wider for it. No call logs exist, so the chasing counts are self-reported — flagged as such, triangulated against a half-day floor observation. The bid log records win/loss but not the reason; the register entry below stays rough for that reason.

The leaks, one card each

Every range shows its multiplication and names its sources — which data, which interviews. If you can't recompute a number from what's on the card, it doesn't ship. All rates are Meridian's own loaded payroll, not industry benchmarks.

Chasing

$17K–51K/yr

Build — field ↔ office status calls

What the data showed

On 14 of 23 jobs, Procore's schedule and status fields were 10+ days stale at some point mid-build — so looking doesn't answer, and people call. The CO log showed 9–16 open change orders at any given time, the most-chased item.

What the people said

All four PMs independently described fielding "where's it at?" calls from the field and making the same calls to the office — 6–10 rounds a week each, 10–15 minutes per round-trip including the re-find. The office manager described the mirror image. In a half-day observed on the floor, the office manager fielded six status calls — the self-reports held up.

The arithmetic

5 people (4 PMs + office manager) × 6–10 interruptions/wk × 10–15 min = 5–12.5 hrs/wk × loaded rate $70–85/hr (their payroll, their burden) = $350–1,062/wk × 48 working weeks = $16,800–51,000 → call it $17K–51K a year

Each person's own minutes on the same call are separate payroll — counting both ends is not double-counting. The field side (six foremen) is excluded: we interviewed one of six crews, too thin to cost. The range is conservative for it. The CO cycle-time days behind many of these calls are a separate metric — costed at L4, cross-referenced here, never summed.

Sources: Procore status-field export (23 jobs); CO log, Sep–May; interviews: all four PMs, office manager; half-day floor observation.

What it means on Monday: a week of skilled time each month spent asking questions a current status field would answer.

Re-keying

$10K–26K/yr

Build → books — Procore into QuickBooks Desktop, by hand

What the data showed

Sub invoices, POs, and field time are entered in Procore, then typed into QuickBooks Desktop again. Comparing QBD entry dates against Procore commitment dates across the 23 jobs: costs land in the books 2–5 weeks after they're committed. 6 of 23 closed jobs carried cost-coding corrections found at closeout.

What the people said

The office manager and the controller both walked us through the same double entry, unprompted, and estimated the weekly volume within 20% of each other — 35–50 documents a week between them, 6–10 minutes each. When the two systems disagree, QuickBooks wins, and finding out takes a phone call (that call is L1's problem; the linkage is noted, the dollars stay here and there separately).

The arithmetic

35–50 documents/wk × 6–10 min = 3.5–8 hrs/wk × loaded rate $52–60/hr × 48 weeks = $8,736–23,040 → $9K–23K/yr in entry time + errors it seeds: ~8 jobs/yr with coding corrections × 2–6 hrs of controller tracing × $58–70/hr = $1K–3.4K/yr Total: $10K–26K a year

Per the tie-breaker rules: rework traceable to a transcription error rides this leak's error line, not Redoing. And the 2–5-week lag this entry pattern creates is the mechanism behind L3 — cross-referenced, costed once, there.

Sources: QBD job-cost entry dates vs. Procore commitment dates (23 jobs); closeout correction entries; interviews: office manager, controller.

What it means on Monday: a day a week of office time typing what a system already knows — and books that trail the job by up to five weeks.

Flying blind

$96K–176K/yr

Closeout — margin fade discovered when the job is over, knowable at 60%

What the data showed

21 of 23 closed jobs finished below their bid margin (median drift −1.9 points — some drift is the trade; that's not the finding). The finding: 9 jobs finished 2 or more points down, and on 7 of the 9, the first written flag was the closeout job-cost review — even though the driving cost (a busted sub buyout, unbilled scope creep, equipment re-orders) was on the books, or committed in Procore, by the job's 60% mark. The contrast group is the evidence: on fade jobs the PMs caught mid-stream, action followed — a recovery change order, a scope correction — and those jobs landed 0.5–1.5 points down. The late-discovery seven landed 3.2–4.6 points down. Same PMs, same trade, different discovery date.

What the people said

All four PMs described the same closeout surprise pattern without prompting; the controller described assembling final job costs "after the crew's already on the next one." The owner's own calibration guess at kickoff — "we find out at the end" — matched the data. Nobody in the building disputes the mechanism; the fresh cost feed to act on it just doesn't exist (see L2).

The arithmetic

Late-discovery fade jobs, annualized from the 9-month sample: 7 observed → 8–9 jobs/yr × average contract value of those jobs, $600–650K (they skew large — the big jobs fade quietest) × 2–3 margin points recoverable by mid-stream discovery (their own early-caught jobs are the comparable — not an industry benchmark) = 8 × $600K × 2 pts = $96K low · 9 × $650K × 3 pts = $175.5K high → $96K–176K a year

Range honesty: "recoverable" uses Meridian's own early-caught jobs as the comparable, and the batch-entry lag (see the map's honesty note) blurs exactly when each cost was knowable — hence the wide range. A First Win here instruments the real number in week one.

Sources: job-cost closeout reports, all 23 jobs; Procore commitment dates; bid recaps from both estimators; interviews: all four PMs, controller, owner.

What it means on Monday: the margin conversation happens at the funeral instead of the sickbed. Two to three points on eight or nine jobs a year is the single biggest number in this building.

Key-person

$2K–3K/yr + a named risk

Change orders — pricing queues behind a single senior estimating role

What the data showed

Of 61 change orders in the sample, 44 were priced by the same senior estimating role. In weeks with two or more active bids, CO pricing sat 4–9 days from request to priced; in quiet weeks, 1–2 days. 21 of those 44 COs missed the monthly billing cycle they would otherwise have made. And the log shows a 16-day stretch last October with zero COs priced — matching a documented vacation. The queue is visible in their own data.

What the people said

Three of four PMs and both estimators described the same queue, in the same shape: CO pricing waits when bids are hot, because bids eat first. Nobody described a person failing — everybody described a structure with one seat where two are needed.

The arithmetic

21 of 44 COs missed a billing cycle in 9 months → ~28/yr × average CO value $9–14K = $252K–392K billed ~30 days late × carrying cost at their 9–10% line of credit × 30/365 = $1.9K–3.2K/yr → call it $2K–3K

The dollar figure is small on purpose, and we're saying so: the countable cost is delayed billing, and it's minor. The finding is the single point of failure — CO pricing stopped for 16 days when one person was out, and the pricing knowledge for Meridian's largest CO category lives in one head. That exposure is described, not priced; inventing a number for it would be theater. Filed as Key-person, not Waiting, per the tie-breaker: the countable baseline (queue days) forms behind one role. And per the attribution rule, this finding names the role and the structure, never the person — the fix is a chartered second pricer, and the person in the seat heard this finding, worded exactly this way, before the delivery session.

Sources: CO log — 61 COs, request→priced gaps, the October gap; interviews: both estimators, three of four PMs.

What it means on Monday: every change order is one vacation, one resignation, or one busy bid season away from a stopped queue.

The "Not a leak" register Bid accuracy on public plan-and-spec work

What we found: the bid log shows 74 bids in 9 months, 26 won — 35% overall. On public plan-and-spec work: 19 bids, 2 won — 11%. On 11 of those 17 losses a winning number was public, and Meridian's bid averaged 9–16% above it. Rough cost, register-grade: roughly 25 such bids a year × 25–40 estimating hours × $62–75/hr ≈ $39K–75K of estimating time a year riding on an 11% hit rate — a wider, softer number than the leak ranges, and labeled that way.

Why it's on the register and not the map: this is a pricing-accuracy finding, not drag on the path — the work moves fine; the number on the bid is the problem. Forcing it into a leak would be the taxonomy lying to protect itself.

Routing — a process fix, yours to run: a quarterly bid post-mortem against the public winning numbers, feeding a unit-cost refresh in the estimating workbook. It doesn't need us, it isn't one of the three moves, and it's on this page anyway because you paid for what we found, not what we can sell.

✓ Where it runs clean — we looked

Contract & buyout: median 6 days from signature to buyout complete; subs bought out before mobilization on 20 of 23 jobs. No card, no cost — this handoff works, and saying so is part of the evidence ("we looked there too"). Also examined and clean: the service division's dispatch-to-invoice loop — 138 tickets sampled, median 2 days from completed work to invoice, techs entering time same-day on phones. A map that's all leaks is an indictment; this isn't one.

Four leaks, each dollar costed once: roughly $125K–256K a year ($17–51K + $10–26K + $96–176K + $2–3K), plus the register item and one named exposure. Ranges, not points — single audited numbers come only after a First Win instruments a baseline.

The Three Moves Memo

Meridian Mechanical · June 2026 · one page · yours to keep and run without us

Three moves, picked from eight candidates by value, absorbability, and countability. Each is tagged with what kind of fix it actually is — that tag is the honesty mechanism: it shows you which recommendations aren't AI and which cost nothing but discipline. Each move's worth is priced before our fee appears anywhere near it.

Put job margin in front of the PMs every Monday, while it's still moveable

AI-assisted

The move: A weekly per-job readout — budget vs. committed vs. billed, assembled from the Procore and QuickBooks exports you already produce — on the table at the Monday PM meeting, with variance flags drafted for human eyes. Kills L3 (Flying blind); the assembly rig also blunts L2's lag. AI does the assembly and flagging; your PMs and controller do all of the judging. We checked the process-fix answer first: a manual weekly cost meeting was tried twice at Meridian in 2025 and died both times, because assembling the numbers took half a day. The assembly is the part that gets automated; the judgment isn't.
What it's worth: 8–9 late-discovery fade jobs/yr × $600–650K × 2–3 margin points recoverable = $96K–176K a year (L3's arithmetic, traceable to the map)
What it takes: Six weeks. Your people: the four PMs (20 added minutes in a meeting they already hold), the controller (~2 hrs/wk during build-out, ~30 min/wk after), the owner at weeks 1 and 6. Tools touched: Procore exports, QuickBooks Desktop exports, one shared readout page — nothing new purchased. Our fee: $52K (sample figure from the provisional $45–65K First Win band; 90-day measurement tail included; your real quote is fixed, in writing, before you commit).
How it's measured: Baseline, instrumented in week one from data you already have: share of fade jobs first flagged before 60% complete — today, 2 of 9. Read by the owner and PMs every Monday; formal reads at week 6 and day 90.

What building it yourself doesn't include: the instrumented baseline (what gets flagged and when, counted before anything changes), the one-habit-per-person absorption plan, the 90-day drift check, and a reported miss — in writing — if the number disappoints.

Make Procore answer the status question, so people stop calling the office

Turn on what you own

The move: You already pay for Procore's change events module, mobile field app, and status dashboards — they're configured off or kept stale. Turn them on, set one rule (status current by Friday, checked in the Monday meeting), and give the office manager the dashboard. Kills most of L1 (Chasing). This is configuration and discipline, not a build — your Procore admin, or your IT provider, can run this from this page, and we're putting that in writing per our own rules.
What it's worth: 70–75% of L1's $17K–51K range (status calls don't go to zero; some chasing is the trade) = $12K–38K a year
What it takes: 8–12 hours of configuration plus three weeks of the Friday rule sticking. No new licenses, no new spend. If you buy the First Win, this folds into its first two weeks at no separate fee; standalone, it doesn't need us at all.
How it's measured: Two counts: status calls per week (a one-week tally by the PMs and office manager, before and at week 6) and Procore fields more than 7 days stale (today: 14 of 23 jobs had them; target: near zero).

What building it yourself doesn't include: honestly — not much. This one is discipline. What we add if it rides inside the First Win: the before/after call tally actually getting run, and the follow-through in week three when the Friday rule starts slipping. If your team holds the rule without us, you'll get the same result.

Charter a second change-order pricer and put a 48-hour clock on pricing

Process

The move: A leadership decision plus about twenty hours of work: name one PM as the chartered second pricer, draft a CO pricing playbook from your own last 61 change orders (the pricing logic, written down for the first time), and set a 48-hour clock from request to priced. Kills L4 (Key-person) — the queue and the exposure both. Structural options, not mandates: the leadership team picks the second seat; a dotted-line review by the senior estimating role keeps quality where it is.
What it's worth: Countable: the $2K–3K/yr of billing-cycle carrying cost, plus ~28 COs/yr billed a month sooner — the smallest dollar figure on this memo, and we're saying so. The real worth is the exposure it closes: CO pricing stopped for 16 days last October when one seat went empty. After this move, that stops mattering.
What it takes: Your call on the seat, ~20 hours total (training + playbook drafting), zero new tools. No fee required — this one is yours. If you want the playbook drafted from the CO log by us, it folds into the First Win; it doesn't justify its own engagement and we won't pretend it does.
How it's measured: From the CO log you already keep: median request→priced days (today 4–9 in bid weeks; target ≤2), and share of priced COs making the next billing cycle (today 21 of 44 missed; target >90% made).

What building it yourself doesn't include: the playbook drafted from your own 61 COs rather than from memory, and the queue-time baseline that proves the clock is being kept after month one.

What we're not recommending (and why)

You've been quoted a Procore↔QuickBooks connector platform at $11,400/yr plus setup. We'd skip it, for now. The arithmetic: L2's whole low end is $10K a year — the connector's carrying cost eats the fix. QuickBooks Desktop is the real constraint (the connectors that work well assume the hosted version), and Move 1's readout rig removes the decision-driving half of the re-keying pain — the stale books — leaving the keystroke half, which is a real but smaller problem that belongs to your bookkeeper's process, not to a platform purchase. If you move to hosted accounting later, revisit; that's a genuine integration case then. We are also not recommending a reorganization: nothing on this map is an org-chart problem.

The First Win pick: Move 1, the Monday readout

Biggest costed range on the map, strongest evidence behind it, and it passes all four gates:

Countable Baseline exists in week one from job-cost data you already produce: fade jobs flagged before 60% — today, 2 of 9.

Absorbable One habit per person: 20 minutes added to a Monday meeting the PMs already attend.

Inside your tools Procore and QuickBooks exports you already run. Nothing purchased, nothing installed on your systems.

Six-week-shaped Wk 1 baseline + design · wks 2–4 build with your people in the loop · wk 5 your team runs it, we watch · wk 6 measure and hand over.

The low-end-clears-fee check, on the page:
Combined fees: Assessment $34K + First Win $52K = $86K (sample figures from the provisional bands).
Move 1's worth at the bottom of its range: $96K/yr → clears $86K inside month 11. At the midpoint ($136K/yr), inside month 8.
With the fee credit below, your combined outlay is $69K → the low end clears it inside month 9.
If no move had cleared this bar, this memo would say so and recommend against buying the First Win. This one clears it at the bottom of the range, which is the only place we run the test.

Fee credit: half the Assessment fee — $17K here — credits toward a First Win signed within 60 days.

The standing rules this memo was built under: each move's worth is priced before our fee is; any move your existing IT provider or your own team could execute says so, in writing (see Moves 2 and 3); and this memo is yours — every move on it can be run without us.

What an Assessment costs your team: about an hour per person interviewed, one data pull by the person who runs your systems, and two leadership sessions. That's the promise — and it's the standard this sample is built to.

Reminder — this is a worked sample. Meridian Mechanical is a fictional composite built to show the method: the map, the visible arithmetic, the fix-type tags, the things we'd tell you not to buy. Your Assessment would look like this, with your jobs, your numbers, and your names for things.

Start a conversation: cory@haldco.io · HALDCO · The business you already run, running better.