Proof of work · September 2026

Booked calls down 60%.
Revenue up 28%.

Between June and August 2026 my booked calls fell from 84 to 34. My cash rose from $49,535 to $63,544. The gap was filled by 1,278 dials into a database I already had, chosen by software I built.

June 2026

$49,535

84 booked calls · 631 dials

TWO MONTHS

August 2026

$63,544

34 booked calls · 1,278 dials

Built and operated by Ian Ryan Kirk
System window 2026-07-18 to 2026-09-16 · measured against a June 2026 baseline

What I build

A system that decides who to call, when, what to say, and when to shut up.

I am not the most polished salesperson in the room and I do not claim to be. I am dyslexic. So I built software to do the three parts of selling that are reading, remembering and restraint, because those are the three things a machine does better than I do and the three that used to cost me deals.

Everything below is that system running for sixty days on a real book, inside a B2B company selling courses, a paid community and custom build work, with real money on it and every mistake mine. I built it and I operated it. Where a number got smaller under checking, the smaller number is the one printed.

I build it

Scoring, refusal rules, follow-up lanes. Tested logic, not prompts

I run it

Unattended on a server, logging every decision including the ones to do nothing

I work the list

I dial what it finds and I close. The system decides which calls are worth having

The closing record behind it, 12 months and $558K, is on its own page.

01 · The mechanism

The Database Juice Squeeze.

Ian's supply of booked Zoom calls was constrained. Rather than wait for the calendar to refill, he filled the pipeline by dialling the database that was already sitting there and by recovering abandoned checkouts. Revenue moved to a channel that does not require a booked call at all.

That is what the next four months look like when you put the activity next to the cash.

Jun 2026

Ian

631

dials

84 booked calls

$49,535 cash

The control

167

dials

56 booked calls

$54,189 cash

Jul 2026

Ian

222

dials

67 booked calls

$63,262 cash

The control

93

dials

85 booked calls

$43,276 cash

Aug 2026

Ian

1,278

dials

34 booked calls

$63,544 cash

The control

92

dials

60 booked calls

$34,531 cash

Sep 2026(partial)

Ian

754

dials

27 booked calls

$41,689 cash

The control

98

dials

36 booked calls

$28,526 cash

What the control did instead

The control's activity stayed flat all year, at roughly 90 to 170 dials a month.

Ian built a second channel. In August he dialled 14x what the control dialled, and that is where the money came from that the calendar no longer supplied.

02 · Who to call

73% of scored leads are killed before anyone is contacted.

Speed to lead only pays if you are fast to the right lead. So the first job is not outreach. It is elimination.

The system read 13,164 historical transactions across a 96,000-record database to work out what a buyer actually looks like. Not opinions about the ideal customer. Measured lift against the people who really paid.

Trait measured on people who paid

Lift

Sells lead gen or cold email as a service

4.00x

Revenue $25k to $100k per month

3.16x

Revenue $5k to $25k per month

3.03x

Has a business and needs more leads

2.67x

Actively looking right now

2.01x

Business email rather than a personal one

1.39x

On the last run it scored 1,529 leads, killed 1,109, routed 111 already-paying customers off the outreach list entirely, and handed back 297 ranked names split into 193 to dial and 104 to email.

Then speed

An abandoned checkout is answered within about 30 minutes, around the clock, because the monitor runs every half hour and does not sleep. Inbound replies are swept every 15 minutes, so nothing sits overnight.

That is the whole thesis. Profile the entire database to decide who is worth a conversation, then be first to them. The other closer is working the same market with the same offers, from a list nobody scored, at the speed a human can manage.

03 · The proof

There is a control group.

Most AI case studies cannot separate the tool from the market. This one can, because a second closer worked the same book, the same offers and the same months without it.

Feb 2026

Ian

$54,668

48 calls

Control

$61,338

73 calls

Apr 2026

Ian

$20,193

28 calls

Control

$48,136

57 calls

Jun 2026

Ian

$49,535

84 calls

Control

$54,189

56 calls

Jul 2026

Ian

$63,262

67 calls

Control

$43,276

85 calls

Aug 2026

Ian

$63,544

34 calls

Control

$34,531

60 calls

Sep 2026(partial)

Ian

$41,689

27 calls

Control

$28,526

36 calls

August is the cleanest head-to-head on the page. Same month, same company, same offers. Ian took 34 calls and produced $63,544. The other closer took 60 calls and produced $34,531.

June was the crossover. Before it, the control out-earned Ian in six of nine months. The two-closer pool grew about 39% in the same stretch, so this is not one book eating the other. Both men moved. They moved in opposite directions.

And the mechanism is documented. An automated screener removed 48 weak calls from Ian's calendar during the window. Fewer calls is not a side effect of the system. It is the system.

04 · The honest arithmetic

Two denominators. Both shown.

Divide cash by booked calls and the number looks spectacular, because the booked calls are exactly what went away. So here is that figure and the conservative one next to it. Both windows compare October 2025 to May 2026 against July to September 2026.

Cash per booked call

Ian

$622to$1,3162.1x

The control

$665to$5870.88x

Read this as what happens when booked calls stop being the only channel. It is not a claim that he got better on Zoom. The denominator shrank because the revenue moved.

Cash per human conversation

Ian

$468to$6331.35x

The control

$595to$4220.71x

Booked calls plus answered dials. This is the honest denominator and the conservative, like-for-like measure. It is the number to argue with.

On the conservative measure Ian is 1.35x and the control is 0.71x. That is the smaller claim, and it is the one that holds when you count every human either of them actually spoke to.

05 · What actually runs

Sixty days, fully logged.

Every figure below comes from a write-log the system keeps on itself. Nothing here is an estimate, and where a rail could not be measured it is named rather than filled in.

6,104

Messages delivered to people outside the company

1,510

Sends the system stopped itself from making

8,957

CRM record changes, each verified after writing

1,879

Rehearsal runs executed before live sends

480

Reports and worklists produced

13

Automations running unattended, 24/7

Rail 01 · Community DMs

2,876 sent · 23.4% reply

673 replies across 1,634 conversations. The highest-volume and highest-replying rail in the system.

Rail 02 · CRM messaging

2,093 sent · 783 people

1,348 SMS, 523 email, 222 WhatsApp. 216 provider failures were caught and 190 retried on another channel.

Rail 03 · Direct email

1,135 sent · 1,063 addresses

160 of those were written for one named person and sent to nobody else. The bespoke end of the work, not the batch end.

Behind those rails: 240 named automations, roughly 55,000 lines of code, 25 test suites, and a reply watch that sweeps every fifteen minutes so an inbound message never sits overnight.

06 · What it produced

$56,837 put in front of Ian in 60 days.

He collected $36,837 of it. The other $20,000 the system flagged on 6 August, nobody chased, and a colleague closed it.

Ian’s sales in the window

Sales · Cash

All Ian’s sales in the window

95 · $141,074.30

System messaged, carded or screened them first

34 · $36,837.50

Appeared only in a generated report

26 · $37,628.30

No system trace at all

35 · $66,608.50

26% of cash had the system in the path. First touch: 14 a message, 13 a dial card, 7 a call screen.

The abandoned-checkout funnel

1,030 emailed, 33 replied, 31 booked, 10 held the call, 29 purchased after, 13 credited to Ian at $14,436.95.

Throw out every same-day purchase as unattributable, because the monitor fires within 30 minutes of the abandon and those people were finishing anyway. $11,055.95 still lands on days 2 to 30.

13 of the 29 bought with no reply and no call. No human spoke to them. Nobody can credit that to sales skill, because no selling happened.

The complete loop

Abandon email 28 Aug → replied → booked → held the call → phone close → paid $1,497.15 on 14 Sep → the system filed the commission claim the same day → approved and paid 17 Sep.

Found by the machine, closed by the human, claimed by the machine. Cross-referencing which sales still needed a claim used to mean reading five separate spreadsheets by hand. 7 claims filed, $13,979.25.

07 · The loss that built it

The system identified him. That part worked.

On 6 August it flagged the lead. $20,000 of revenue, a $2,000 commission. Ian never got him reassigned into the system as his own, so five weeks later the prospect came back on his own, went to a colleague, and the commission went with him.

Missing · Never put into a call log or dial queue

Leads route into the queue automatically, and the contact is reassigned so replies notify the right rep

Missing · Never texted

The messaging lanes, with the full pre-send rule set

Missing · Never emailed after the first one

The 24 to 26 day chase, so nobody goes quiet inside the window

Missing · Nothing of value to send him

A searchable library of recorded coaching calls, used as a reciprocity offer: their own objection, answered on the record, quoted with its timestamp

None of those four fixes are personal to Ian. A rep who does not get a lead into the system loses it, whoever they are. The system closes that gap for anyone on the team, which is the argument for it existing at all.

“I lost a $2,000 commission because I never got him into the system as mine. Everything we built after that exists so it cannot happen to anyone else.”

08 · The differentiator

It refused 42% of its own outbound.

Anyone can buy software that sends. The sixty days of work is in the software deciding not to.

On the CRM rail it prepared 3,603 messages and sent 2,093. Before every send it pulls that person's live record, their full message history, their notes, their tags, their timezone and their purchase history, and runs fifteen tests. Any one failing stops the message.

Why it stopped

Times cited

Wrong hour where they live

770

Already sent this one

619

They wrote and nobody answered yet

266

Contacted too recently

178

Opted out of that channel

88

A personal email already went to them

85

WhatsApp window closed

27

Junk number

22

Marked do-not-contact

18

Already said no

15

Marked dead by hand

5

No channel on record

2

The most common reason a message did not go out is that it was the wrong hour where the person lives. Not a rule about us. A rule about them. The counts total more than the messages stopped, because one message can fail several tests at once.

The rule that cost the most to learn

On 30 July a prospect replied "No longer need" at 22:08. A message asking whether he still wanted to proceed went out at 22:30, twenty-two minutes later, while Ian was on the phone with him.

The pre-send check already verified opt-out status, tags, phone validity and lead ownership. It passed all four. It had never been asked whether the human had spoken.

It is now. Those two new rules have since stopped 281 messages that would have talked over somebody mid-conversation.

That is the distinction worth holding onto. A filter checks a database field. This reads the last thing a person actually said, decides whether it was a door closing, and refuses to talk over it.

09 · The fifteen tests

Here they are. All of them.

Every one of these was bought with a mistake. The counts are how many times each stopped a message in the same sixty days. Grouped by what the test reads, because that decides whether it survives a move to your CRM.

Reads our own record of what we have already done

A CRM cannot answer these. It does not know what we sent, only what it sent.

Have we messaged this person recently, on any lane

178

Have we already sent this exact thing

619

Has a personal email already gone to them today

85

Did mail to this address bounce before

0

Did a human mark this one dead by hand

5

Reads the person

Universal fields. Every CRM has them under some name.

What time is it where they live

770

Do we have a working channel at all

2

Is there a blanket do-not-contact on the record

18

Did they opt out of this specific channel

88

Reads what they actually said

The expensive ones to build, and the only ones that stop a message for a human reason.

Did they write to us and nobody has answered yet

266

Was the last thing they said a door closing

15

Is there an opt-out buried in a free-text note

not counted separately

Reads the list and the channel

Hygiene. Cheap to run, and they catch what the tidy fields miss.

Do they carry an opt-out tag

not counted separately

Is the phone number junk

22

Is the WhatsApp reply window still open

27

Two things to notice

Bounce suppression fired zero times. A rule that never fires is not a rule that failed. It is what a clean list looks like, and it stays in because the month it is needed is the month nobody is watching.

Seven of the fifteen move to your CRM unchanged. Five of them because they read a ledger I keep rather than anything your system has to be truthful about, and two because a clock and a phone number mean the same thing everywhere. That split is deliberate. Safety that depends on a CRM being accurate is not safety. One test is genuinely shaped around this CRM and would need real work to move.

10 · The lesson

Every rule was bought, not designed.

In sixty days the operating manual records 25 separate dates as the day something was learned. A rule bought roughly every 2.4 days. The pre-send check went from about four tests to fifteen in the same period.

None of those tests were anticipated. The message that created the two most important ones passed every check that existed on its way out the door. It was not a careless system. It was a careful system with a blind spot invisible from the inside.

The transferable part

"Assume the first version is wrong in ways you cannot list, and make it log its own reasoning, so the wrongness turns into rules instead of damage."

What turned a lost deal into an asset was that the system had written down why it sent. The failure was legible. The logging is the decision. The rules are just what falls out of it.

11 · What this does not claim

The limits, stated first.

Rigour about what is missing is what makes the rest of it worth reading.

  • Cash per booked call is not a skill measurement.

    It is a channel-mix measurement. The denominator shrank because revenue moved to a channel that does not need a booked call at all. Cash per human conversation is the like-for-like number, and it is the smaller one. Both are on this page for that reason.

  • It is one person against one person.

    Two closers is not a sample. Individual performance moves for reasons that have nothing to do with tooling.

  • Inbound routing is not random.

    Discovery calls are not round-robined. An external router decides who sees a lead, so lead quality between the two books is not guaranteed equal.

  • Cash figures are rep-filed, not audited.

    The finance sheet is the authoritative record and will differ. These are for shape and trajectory.

  • No campaign-level causal claim is made.

    The abandoned-checkout lane has 11 sales credited totalling $12,610.40 with an automated message dated before purchase. That is traceable money with a machine touch in it. It is not proof the machine caused the sale.

  • September is excluded from the cleanest comparison.

    6,023 opportunity cards were reassigned on 11 September, which moves the book. July and August are clean, which is why August leads.

The defensible sentence

"His booked-call supply fell 60% and his revenue rose 28%, because the database became the pipeline."

That claim is fully supported by the figures above. Anything stronger is not, and is not made here.

12 · On your book

What this actually looks like if you hire me.

I build the system and then I operate it. Not a consultant who leaves a deck. Not an agency that runs campaigns at you. Not a closer for hire.

First. The audit

I read your database against your money. Who actually bought, what those people had in common before they bought, and who matching that description is sitting there being ignored. You get a ranked list and the size of it.

Paid pilot. It has never run on a database that is not the one it was written against, so I sell it as a pilot rather than as a known quantity.

Then. The build

The scoring, the refusal rules and the follow-up lanes, standing up against your stack. It runs unattended and logs every decision it makes, including the ones to do nothing.

Scoped after I have seen your stack. I will not quote a timeline for work I have not looked at.

Then. I work it

I dial the list I built. I close. You get a monthly attribution number measured from the money backward, and a random control group held back from day one.

The measurement is part of the product. The one avoidable mistake of the sixty days below was not keeping a holdout.

The boundary, before you ask

This works the volume end of a book. Courses, community, abandoned checkouts, reactivation, the dial queue. It finds and warms the people worth a call.

It does not close five-figure builds. In the window below, 62% of the revenue it never touched was high ticket, and that is by design. Those deals are human conversations from the first minute to the last. Any version of this pitch that implies otherwise would not survive the first client.

Worth a conversation

  • A CRM with years of contacts in it and a suspicion that money is buried in there
  • A team where the follow-up is the leak, not the closing
  • A founder still personally involved in pricing who suspects the offer is underpriced
  • Anyone who has hired an agency, received activity reports, and never once received attribution

Not worth either of our time

  • Pre-product, or no list. There is nothing to squeeze
  • You want leads bought rather than a system built
  • You want a coach for the team rather than an operator in the seat
  • You need someone who will tell the room the AI did it all

The Ask

Bring your numbers. I will bring mine.

No price on this page, because the honest answer depends on your stack and I have not seen it. What I can tell you in one call is roughly what is sitting unworked in the database you have already paid for, and what it would take to get at it. If the answer is that there is nothing there, I will say so and we both save a quarter.

Reply to the message I sent you.

That is the whole ask. Same thread, wherever this link reached you. Tell me what you sell and roughly how many people are sitting in your CRM, and I will tell you whether this is worth a call.

Compiled 16 September 2026, block counts re-derived from the send ledgers on 21 September after the first version ranked a figure that would not reproduce. Every number traces to a log file, a weekly platform report or the team activity record. No prospect names, email addresses or phone numbers appear anywhere on this page.

Contents