We build AI systems that answer leads, chase follow-up, and fill a calendar. Before selling one to anybody, we ran it on our own retirement-planning business for five months. This page is that business's production database, opened up. Every figure below was pulled live on August 17, 2026 — and every one carries the query it came from.
Pulled from production August 17, 2026 · no rounding up, no borrowed logos
The database was empty on March 10, 2026. Everything in it since arrived through a system that runs whether or not anyone is awake — Instagram DMs answered in Marc's voice, website tools that qualify and file the lead, TikTok DMs, quiz results. The DM agent alone captured 1,642 of them.
| As of | Running total | Added that month |
|---|---|---|
| Mar 31, 2026 | 22 | 22 |
| Apr 30, 2026 | 208 | 186 |
| May 31, 2026 | 974 | 766 |
| Jun 30, 2026 | 1,747 | 773 |
| Jul 31, 2026 | 2,141 | 394 |
| Aug 17, 2026 | 2,272 | 131 |
| Where the lead came from | Leads | Run by |
|---|---|---|
| Instagram DM agent | 1,642 | AI |
| Website tools & calculators | 402 | AI |
| Website quiz | 77 | AI |
| TikTok DM agent | 45 | AI |
| Direct bookings | 20 | By hand |
| Referrals | 17 | By hand |
| Cold calls | 17 | By hand |
| Paid ads & lead forms | 13 | By hand |
| Everything else | 39 | By hand |

The busiest front door · 727 visits in 90 days · every completion files itself as a lead, scored and tagged, with no one watching
Source: Neon production database, table prospects · SELECT count(*), source, created_at · run 2026-08-17. Two test rows excluded from the source table above.
They commented on a post. The AI replied in Marc's voice, delivered what it promised, captured the email — and in the third one, the lead asks for the meeting herself. Names and faces are blurred because these are real people.
Red circle · the captureHer email address, handed over willingly. Value first, ask second.
Red circle · the askIt works out she's newly retired, then asks for the meeting — politely, at the right moment.
Red circle · the outcomeDifferent lead, same machine. She asks for the consultation herself.
Source: screenshots taken from the live Instagram account, 2026-08-12. Names, handles, avatars and email addresses redacted before publishing.
One conversation, start to finish, on a Sunday morning. She commented on a post. The AI qualified her, took on a genuinely hard cross-border retirement question, declined to talk fees in a DM, agreed a time, corrected that time when she caught her own typo, asked for her email — and put the meeting on the calendar. Marc found out it existed when the invite showed up.

One · it earns the askShe raises retiring abroad. It names the actual traps — non-resident withholding, OAS eligibility, the TFSA problem — then asks for the call.

Two · it knows its limitsAsked for fees, it refuses to quote in a DM. She picks a time, then says the time was a typo — it updates without being told twice, and asks where to send the invite.

Three · the receiptThe event, written by the system: “Booked via IG DM” — and it quotes her own words back as the reason the time was set.
Source: screenshots from Marc's phone, 2026-08-17, of one continuous conversation and the calendar entry it produced. Guest name, photo, email address, Instagram handle, meeting link and organiser address redacted before publishing — she is a real person who has not agreed to appear here.
Capturing a lead is the easy half. The system that follows is the one that earns money: a sequence that emails, waits, texts, nudges toward the calendar, and stops the moment somebody replies — so nobody ever gets the same message twice.
Source: Neon tables drip_sends (2026-03-30 → 2026-08-17), email_events (2026-06-15 → 2026-08-17, engagement events only), sms_conversations (2026-03-22 → 2026-08-17). Reply rate computed per distinct phone number.
This is the number that matters. Not impressions, not "engagement" — signed clients and the assets they moved over, sitting in the same database that captured them in the first place.
Source: Neon tables prospects (stage = Closed-Won, sum of assets), calendar_events, meetings · run 2026-08-17. Assets are recorded on 15 of the 21 won deals; the $1.53M is the sum of what is recorded, not an estimate of the rest.
The lead machine needs something to eat. That's the content engine: scripts written against what actually performed, an AI avatar that films them, an automated render, and a scheduler that posts them. 307 posts published — and half a million views in the last thirty days, on the platform's own dashboard.
| Date | Followers |
|---|---|
| May 14, 2026 | 3,099 |
| Jul 12, 2026 | 7,920 |
| Aug 17, 2026 | 10,404 |

The profile · 506.9K views / 30 days

Creator dashboard · 2.4K new followers

Meta's own dashboard · 484.3K views · 169.6K accounts reached · last 28 days
| Best-performing posts | Views |
|---|---|
| "Everyone waits until 71 to open a RRIF" | 194,612 |
| "Comment RRIF" — the follow-up | 160,751 |
| "The RRIF trap hits every Canadian at 71" | 144,034 |
| "Canada has no gift tax, so give while you're alive" | 134,282 |
| "You take CPP at 60 and keep working" | 132,769 |
Source: Zernio account analytics pulled live 2026-08-17 (followers, posts, per-post views); Instagram and Meta Business Suite dashboards screenshotted 2026-08-12; rendered-video count is a file count of the render output folder.
Search is moving inside the AI assistants, and almost nobody is measuring whether they get named. We built an engine that publishes to the site, then interrogates the models every day to check whether they mention us and whether they cite the site as a source. It is scored, not guessed — and the search traffic followed.
| Month | Search visits |
|---|---|
| March 2026 | 23 |
| April 2026 | 101 |
| May 2026 | 187 |
| June 2026 | 422 |
| July 2026 | 559 |
| August 2026 (17 days) | 313 |

Microsoft's own AI Performance report · 7.2K citations · 3 months · the late-July fall-off is real: we cut publishing volume deliberately after a quality change, and it shows

Google Analytics, last 28 days · 2.2K active users · the quiz and the planner tools sit at the top, which is the whole point

A slice of the library · 370 articles published · researched, written, fact-checked and posted by the system, each through a quality gate
Source: Google Analytics property for the site, read live 2026-08-17 (sessions, users, organic-search channel, AI-assistant channel); the site's own citation-tracking tables in Neon, same day (mention and citation rates, trailing 28 days vs prior 28); article count is a file count of published posts on disk; Microsoft citation figures from Bing Webmaster "AI Performance" (beta), trailing 3 months as at 2026-08-17. The chart shows complete months only — August is 17 days old at the time of writing and is excluded rather than shown as a short bar.
Not a subscription to somebody's dashboard. A set of scheduled jobs and agents that boot on a machine in London, Ontario, watch each other for failure, and do a specific job at a specific time.
Market-typical prices for the equivalent tool or service · our actual running cost is an AI subscription and a server
Source: scheduled jobs verified as loaded on the machine 2026-08-17, not merely installed. A further 18 jobs exist but are switched off, and are deliberately not counted above.
“He saved me six grand in one session.”
The build: a site visit becomes a scan, a voice note and a checklist — and comes back as the design, the estimate and the client presentation, exported straight into the software his team already uses. What took his designer ten hours happens while the crew drives back.
The full case study publishes when the numbers are in. Real ones, like everything else on this page.
The build: 1.15 million Canadian mortgages come up for renewal this year and most brokers never mine their own database for them. The renewal date is already sitting in their system; the work is turning it into a compliant, well-timed conversation.
Scope of work signed, compliance pack written, access checklist ready. Nothing has been built yet — the engagement is held at a licensing gate we insisted on clearing first. We would rather show you a stalled project honestly than a finished one we invented.
Anyone can build a page of big numbers. The useful question is what got left off it — so here is ours, in the same size type.
Take the ten-minute leak audit. Qualify, and we install your first small system on one video call — missed-call text-back, a follow-up sequence, your socials on autopilot. On your accounts. Yours to keep, even if we never speak again.
Start with the leak audit No invoice · no catch · two free installs a week