Quoting is costing about $155,000 a year in gross margin.

Margin varies by estimator on jobs that are genuinely comparable, because the adjustments that make the difference are not written down.

Across the last twelve months Holcombe realized 22.1 percent gross margin on $8.6M of install and retrofit work. Had every job been priced and run the way your senior estimator prices and runs his, the figure would have been 23.9 percent. That difference is 1.8 points, or about $155,000 a year.

The cause is specific. Doug Wetzel adjusts the workbook price for the things the workbook cannot see: rooftop access on grocery sites, prevailing-wage jobs, customer payment history, which crew is available. Those adjustments are sound and they are consistent. They exist nowhere except in the bids he personally prices, so the other two estimators price without them, and the margin follows.

Not everything here is broken, and it matters to say so. Your takeoff tooling feeds pricing cleanly. Your win rate is normal for the trade and does not vary meaningfully by estimator. The workbook itself is a real, competent pricing instrument. The problem is confined to how the workbook is kept and what sits on top of it.

What we recommend

Write down the adjustment layer, fix how the workbook and vendor pricing are kept, and put a drafter in front of the estimators that prices from both and cites every number. Pilot it on rooftop-unit replacement retrofits, where the gap is widest and the jobs repeat. Of the $155,000, we would expect to reach $85,000 to $110,000 inside twelve months, and we would agree the measures up front so you can judge it in your own numbers.

We watched nine live bids and then reconciled what we saw against twelve months of records.

Interviews about a workflow return the procedure. We wanted quoting as it is done today, so the three weeks were spent alongside Doug Wetzel, Alan Prewitt and Cody Sturgill through nine bids in progress: the inquiry, the site walk, the takeoff, the workbook, the adjustments, the proposal and the follow-up. Where what we saw differed from what the procedure said, we recorded both.

The observed work tells you how and why. It does not tell you how much, so the second half of the diagnosis reconciled it against your records: 214 bid files for the trailing twelve months, job costing out of QuickBooks for the 58 that were won and closed, the estimating workbook and its two variant copies, the labor rate sheet, and the vendor quote traffic in the shared mailbox.

What we could measure, and what we could not

• Measured from your recordsQuote turnaround, bids submitted and won by estimator and job type, bid margin against realized margin on closed jobs, and equipment cost variance against the workbook.
• Measured by observation, and therefore a sampleEstimator hours per bid and how those hours are spent. Nine bids is enough to show the shape and not enough to be precise. The hours figures in this document carry that caveat wherever they appear.
• Could not be measuredWhy a bid was lost. Holcombe does not capture loss reasons, and neither our observation nor your records can reconstruct them. This is the single largest hole in the picture and it is worth closing whatever else you decide.

Two of the nine observed bids were rooftop-unit replacements at Bexley Foods, including BEX-118, which appears again in the build section as the pilot example.

A quote takes 6.4 days, and the two slowest steps are the two that wait on information.

The workflow itself is sound and everyone runs it the same way. What varies is how long each step takes, and the variation concentrates in the two steps that depend on information held somewhere else: pricing from the workbook, which waits on current equipment costs, and the adjustment step, which waits on Doug.

Twenty-two of the 214 bids went out after the customer's stated deadline. That is 10.3 percent of everything Holcombe bid last year, and it is the finding Brian described on the first call.

Everything quoting needs is written down, and four of the nine sources have no owner and no review date.

Holcombe is not a company that keeps its pricing in somebody's head. The pricing is written down, in the ordinary ways companies write it down, and most of it is good. What is missing is custody: which copy is current, who owns it, when it was last checked, and where the reasoning behind a number is recorded.

The pattern is worth naming plainly, because it decides what the build has to do first. The three maintained sources are the three that a piece of software already keeps for you. The four with no custody are the four that live in documents and habits, and those are the ones every bid actually turns on.

These are the numbers any later work gets judged against.

Captured before anything changes, so that improvement is measurable in your records and not in our claims. All four are drawn from the trailing twelve months.

The four measures, as of today

• Quote turnaround: 6.4 business days, medianNinetieth percentile 14 days. 22 of 214 bids went out after the customer's stated deadline.
• Estimator hours per bid: 5.2 retrofit, 14.8 install and tenant improvementMedians from the observed sample.
• Win rate: 27.1 percent overall31.1 percent on retrofits, 20.7 percent on install and tenant improvement.
• Margin against estimate: 23.8 percent bid, 22.1 percent realizedA blended fade of 1.7 points, examined in the next section.

The estimator spread survives when you control for the job mix, which is what makes it a real finding.

Realized margin does differ by estimator, and on its own that proves nothing, because the three of them bid different work. Doug takes the large installs, Cody takes small retrofits, and job types carry different margins for reasons that have nothing to do with who priced them. So the raw comparison is the beginning of the analysis and not the end of it.

To settle it we isolated one job type all three estimate: rooftop-unit replacement retrofits between 5 and 25 tons with no controls scope. Thirty-four of them closed in the last twelve months. On that like-for-like set the spread does not disappear. It narrows from 5.7 points to 4.5, and 4.5 points on identical work is the finding.

Two things follow from the like-for-like cut. The first is that the gap is not seniority: Alan Prewitt has eleven years in and prices 1.8 points below Doug on the same job type. The second is that fade tracks the same order. Doug loses 0.9 points between bid and realized, Alan 1.7, Cody 4.2, which says the adjustments are not only pricing the job better, they are pricing it more accurately.

The mechanism behind the fade is separately visible in the cost data. On the 34 comparable jobs, actual equipment cost ran 3.1 percent above the workbook figure on average. That is the two vendor price rounds the master workbook never absorbed, and it lands hardest on whoever is relying on the workbook alone.

The gap is $155,000 a year, and about $95,000 of it is reachable.

Holcombe realized 22.1 percent on $8.6M. Doug realized 23.9 percent. Closing the whole distance would be worth $155,000 a year, and closing the whole distance is not a realistic target, because some of what Doug does is twenty-eight years of pattern recognition that no written rule will fully capture.

So the honest number is a share of that gap. What share is a judgment, and we cannot measure it for you. Below is the arithmetic with the judgment left in your hands. Change the assumption and the number moves.

What is deliberately not counted

Turnaround is the obvious one. On-time bids won 29.2 percent last year and late bids won 9.1 percent, and it is tempting to price the 22 late bids at the difference. We have not, because bids are often late for the same reasons that make them hard to win: a complicated scope, an RFQ that arrived with no time in it. The relationship is real and the direction of cause is not established, so it stays out of the number and goes on the list of things the build should make measurable.

The 535 hours of estimating time is also not counted as money. Freed hours only become margin if they go to more bids or fewer estimators, and which of those happens is your decision, not an outcome we can book on your behalf.

Of the fifteen tasks in a quote, seven go to software and five do not move at all.

This is the part of the diagnosis that decides what gets built. We broke the workflow into its actual tasks and put each one where it is handled best, on the evidence of what we watched. Some tasks want software. Some want a person with a better tool. Some want a process fix and no technology whatsoever. Some are exactly right as they are, and the correct recommendation for those is to leave them alone.

Six positions are fixed before any build starts, and none of them is negotiable against a measured result.

A diagnosis that only says what could be automated is half a diagnosis. These are the limits we would hold to, and we would rather you hold us to them in writing.

• Final pricing judgment and sign-off stay with the estimatorNo exception, no threshold at which this relaxes, and no measured improvement that would justify revisiting it. The price is a commercial commitment your company makes to a customer, and a person makes it.
• No proposal leaves the building without a named human approverThe approval is recorded against the bid with the name and the time.
• The margin floor is a stop, not a warningA draft below 22 percent on retrofit work cannot be sent. It goes to an exception queue and needs an explicit override from a named person, recorded with the reason.
• Stale vendor pricing cannot silently price a bidAnything past 45 days is flagged before the draft is produced, not after.
• Nothing is priced from an unapproved baselineDoug approves the workbook and the adjustment rules before anything reads from them. If he has not approved it, the software does not use it.
• Nothing leaves your Microsoft 365 tenantEverything runs inside Holcombe's own environment under Holcombe's own permissions. No customer pricing, no job costing and no vendor terms move to an outside service.

There is a general position underneath these six. Your estimators hold the job; software does tasks inside it. However much of the drafting a machine takes on, every task rolls up to a person who owns the result, and the system is designed so that person can always see what the machine did and why.

Four things have to be true before a drafter is worth building, and three of them are yours to do.

Software that prices from a source nobody owns produces confident, wrong numbers faster than a person can. The order below is not a preference, it is a dependency: each step is worthless until the one before it is done.

• One workbook is marked as the masterDoug decides which of the four copies is current, the master is marked and owned, and the other three go read-only. The tweaks in the personal copies get reviewed, and the ones he endorses go into the master. Roughly a day of his time. Yours to do, and you can start this week.
• Vendor pricing has one intake path, dated on arrivalOne destination for every emailed quote, with the arrival date recorded. Until this exists, task 06 in the allocation cannot be built at all. Yours to do.
• The adjustment layer is written down and approvedThis is what we would do with Doug: eliciting the rules, writing each one next to the numbers it changes, recording the reasoning, and getting his name against the result. Estimated at about two hours a week of his time over three to four weeks. Ours to run, his to approve.
• The four measures are being loggedTurnaround is the easy one and it can start immediately. The other three come out of records you already keep. Without them the build cannot be judged and we would be asking you to take our word for it.

The first, second and fourth need nothing from us. If you did those three and stopped, quoting would be measurably better than it is today, and that is a fine outcome.

Build the drafter, pilot it on rooftop replacements, and let the measures decide whether it widens.

Rooftop-unit replacement retrofits are the right pilot for three reasons: it is the job type where the margin gap is widest, the jobs repeat often enough to produce a signal inside a quarter, and it is the set we already measured, so the baseline exists. BEX-118, the Bexley Foods store 118 replacement we watched during the diagnosis, is a representative example.

The outcome we would agree up front

• Realized margin on rooftop replacement retrofits priced by Alan and Cody reaches 22.5 percent blendedFrom 21.8 and 19.1 today, measured on closed jobs, two quarters after the pilot begins.
• Median quote turnaround falls below 4 business daysFrom 6.4 today.
• Estimator hours per retrofit bid fall below 4From 5.2 today.
• Above 95 percent of the numbers in a draft cite a source that is currentFrom no citation at all today.

Agree this scope and its outcome, or stop with this document. At the twelve-week gate the measures decide what happens next, and stopping is one of the four available answers. Everything built to that point runs in your tenant, belongs to you, and comes with a written handover.

Five things sit next to quoting that we did not examine, and one of them is probably the next place to look.

The diagnosis was scoped to project quoting from inquiry to signed proposal, and we held to that. What follows is what we noticed at the edges without investigating, recorded so that it is on the table rather than in our heads.

• Service quoting and the rate schedule markup tiersThe tiers were adjusted over the years and the reasons were not recorded beside them, which is the same failure as the adjustment layer in a different document. This is the strongest candidate for the second piece of work.
• Maintenance agreement pricingA different model with different economics. Renewals appear to be priced off the prior year, and we did not test whether that still tracks the cost to serve.
• Change ordersSeveral people mentioned that change orders are where retrofit margin is recovered or lost. Nothing in the diagnosis touched them and we have no view.
• Purchasing and buyoutThe gap between the quoted equipment cost and the purchased equipment cost sits here, and our 3.1 percent variance figure cannot distinguish a stale workbook from a buyout problem.
• The QuickBooks positionQuickBooks Desktop Enterprise has no cloud interface, so job-cost data reaches anything else as an export. That is workable and it constrains what can be automated later. Any Sage Intacct decision should be made on its own merits and not on account of this work.

We also did not form a view on the loss reasons, because the data to form one does not exist yet. Capturing why a bid was lost is the cheapest piece of new measurement available to you and it would change what the next diagnosis could answer.

Everything runs in your environment, belongs to you, and stops cleanly at any gate.

• A named owner on each sideOurs, and yours or your delegate. Weekly working session during the build, and a short written note after each one.
• Inside your tenantEverything runs in Holcombe's Microsoft 365 environment under Holcombe's permissions. Nothing moves to an outside service.
• Yours, including the parts we wroteThe workbook, the adjustment rules, the intake path, the drafter configuration and this document belong to Holcombe. Where a third-party licence is involved it is in your name.
• Stopping is a documented handoverAt any gate, in writing, with whatever has been built left running. It is not a negotiation.
• No vendor tiesWe do not resell anything and we take nothing from any tool we recommend. Our interest is in the measures moving, because that is what would earn the next piece of work.

What it would ask of you during the build is roughly what the diagnosis asked: about two hours a week from Doug while the adjustment layer is written down, less after that, an hour a week from Alan and Cody during the pilot, and a standing half hour from Terry when the measures are reconciled.