Example report

This is what lands in your inbox.

Below is a complete Leak Report, laid out exactly as we send it. We don't publish client reports, so the business here is a stand-in and the figures are illustrative — but the format, the arithmetic, and the habit of showing which numbers are soft are all real.

Anthracite AILeak Report · Example
Prepared for

A mechanical contractor — 31 staff, two locations, about $6.2M a year. Six in the office, the rest in vans.

Leak Report

Pipeline diagnosis · lead → retention · one month of operating data

01

The short version

This business is losing about $18,900 a month — roughly $226,800 a year — to four problems, none of which are about how well the work gets done. Two are lost revenue, two are hours going into things a computer should handle. Nothing here needs new customers, more staff, or a price rise. It is all demand already paid for and then dropped.

The diagram shows where it goes and how it adds up. The widest band is the most expensive problem, and it is the one to fix first.

WHERE IT LEAKS TODAYRECOVERABLECalls nobody answersAfter 5pm, weekends, and while already on a call · $9,200/moQuotes that go coldSent, then never followed up · $5,400/moThe same questions, all dayHours, pricing, arrival windows, invoice copies · $2,300/moPaperwork retyped by handWork orders, supplier invoices, permits · $2,000/mo$18,900per monthif all four are closedEXAMPLE SCENARIO — NOT CLIENT DATA
Ribbon width is proportional to monthly cost, so the widest band is the most expensive problem. Every figure is broken down, with its working, in the ranked sections below.
Total identified$18,900 / mo · 75 hrs / mo
02

Where it's going, worst first

Ranked by what each one costs, not by what is most interesting to build. Every figure shows its working so you can argue with it, and each ends with the part we are least sure about — a report that only carries good news is a sales document.

2.1Calls nobody answers

$9,200 / mo

The phone is the business, and it stops being answered at 5pm. Nights, weekends, and any moment the office is already on another call all go to voicemail — and a caller with a broken system does not leave a message, they dial the next company on the list.

How we got there
  • 412 inbound calls a month, from the phone system's own logs
  • 118 of them unanswered — 29%, concentrated after 5pm and on Saturdays
  • Of calls that ARE answered, 34% become booked work
  • Assume only half the missed callers were real, reachable demand: 59
  • 59 × 34% ≈ 20 jobs a month × $460 average gross margin ≈ $9,200
What we would build

A voice agent that answers every call around the clock, works out whether it is an emergency or routine, books the routine work straight onto the schedule, and rings the on-call phone for anything urgent.

Where we could be wrong

The 50% assumption is the soft number here. We would tighten it in week one by calling a sample of missed numbers back and asking what they did next.

2.2Quotes that go cold

$5,400 / mo

Quotes go out well and then nothing happens. There is no second touch unless the customer starts it, so the ones who were merely busy get lost alongside the ones who said no.

How we got there
  • 96 quotes issued a month, averaging $2,100
  • Median time to first follow-up: 3.4 days; 23 quotes a month get none at all
  • Quotes that receive a follow-up close at 31%; quotes that don't close at 12%
  • 23 × the 19-point gap ≈ 4.4 extra jobs × $1,230 margin ≈ $5,400
What we would build

A follow-up sequence that runs in the company's own voice on a schedule matching how this trade actually buys — and stops dead the moment a human replies, handing the thread to a person with the history attached.

Where we could be wrong

Worth doing second, not first: it depends on the quote data being complete, and here about one quote in six was missing an issue date.

2.3The same questions, all day

$2,300 / mo31 hrs / mo

Roughly fifteen questions account for most inbound contact: opening hours, service area, ballpark pricing, arrival windows, copies of invoices. Each is quick. Together they are most of a working week, and they land on the person who is also supposed to be scheduling.

How we got there
  • About 340 calls and emails a month asking one of the same 15 things
  • 5.5 minutes average handling time, sampled over a fortnight ≈ 31 hours
  • 31 hours × $38 fully-loaded office cost ≈ $1,180
  • Plus the scheduling those hours displace — dispatch errors run about $1,100 a month in return visits
What we would build

The same voice and inbox agent answering these from the company's real information, with anything it is unsure of going to a person rather than being guessed at.

Where we could be wrong

Half of this value is time rather than cash. It only becomes money if the freed hours go into scheduling — worth agreeing up front what those hours are for.

2.4Paperwork retyped by hand

$2,000 / mo44 hrs / mo

Work orders, supplier invoices, permits and warranty claims arrive as PDFs and photographs, and someone retypes them into the system of record. It is the least glamorous number in the report and the most reliable one.

How we got there
  • About 260 documents a month across four types
  • Timed at 10 minutes each end to end ≈ 44 hours
  • 44 hours × $38 fully-loaded cost ≈ $1,670
  • Plus rework: a 4% keying-error rate on supplier invoices ≈ $330 a month in credits chased
What we would build

Extraction that reads each document as it arrives and lands the fields in the system of record, with anything low-confidence flagged for a human rather than silently guessed.

Where we could be wrong

Lowest value of the four and the one we would sequence last — but also the easiest to prove, because the before-and-after is just a stopwatch.

03

What we would do about it

One at a time, worst first. We would build the call answering, get it live, and let it run for a month against the number it is meant to move — calls answered, and jobs booked from them. If it does what this report says, the case for the second build makes itself. If it doesn't, you have lost a month rather than a year, and you would hear it from us first.

A real report ends the same way: nothing to sign, and a first build you can say no to. Deeper consulting and solution builds start at $1,000.

An example, not a client report — we don't publish those. The business shown is a stand-in and every figure is illustrative; yours is built from your own numbers, with the estimates flagged exactly as they are here.Anthracite AI · Example

Want one of these on your business?

The first call is free and the report that comes out of it is yours to keep — your pipeline, your leaks, your numbers.

Get your free Leak Report

For real numbers from a system we built and ran ourselves, see the PlumbDesk build story.

Anthracite AI is an AI consulting and engineering firm that builds AI systems for small and mid-sized businesses. Based in Charlotte, NC, working with businesses anywhere in the US.