Portfolio · Marketing-engine build
CoolRankings — a closed-loop go-to-market engine, designed and instrumented end to end.
CoolRankings sells a $499 search-visibility audit to local HVAC companies — and around that product sits the real work on display here: a full-circle marketing machine. Prospecting from map data, audit-first cold outreach, a tagged and attributed funnel, a CRM that pings a phone when a prospect replies — every stage designed, built, and operated by one person, as one connected loop.
Read this page as system design, not a results deck. CoolRankings is my own venture, launched mid-2026; the engine is live and still being tuned, and the numbers that appear below are its real, current, modest ones. What’s on display is the architecture and the reasoning — including what the instrumentation has already caught and changed.
00 · The loop
Eight stages, one circuit.
The design goal: no stage that doesn’t feed the next, and no result that doesn’t feed the strategy. Outreach creates a visit, the visit is attributed, intent is logged event by event, a reply becomes a CRM row and a phone push, and what the numbers say loops back into what gets sent next. Every stage reads from a niche configuration file, so the engine is a template — the next vertical is a config change, not a rebuild.
01 · The offer
A productized wedge, priced to say yes to.
The funnel is deliberately short: a free 10-second visibility audit (email-gated — that’s the lead capture), a $499 flat-fee report as the cash product, and a monthly retainer as the long tail. Two trust decisions carry the cold-traffic gap: real names on every report — no faceless brand — and a set of anti-promises (“no contracts, no guaranteed-#1 claims, money back”) aimed squarely at a category the snake-oil agencies got to first. The report itself is delivered on a 2-business-day SLA with a human review gate: the automation drafts it, a person approves it, and the buyer is told exactly that — the review is the product’s credibility.
02 · The artifact
Don’t ask them to run an audit. Run it for them.
The first cold campaign pointed prospects at the free audit form — and the tracking layer told the truth about it: a few dozen attributed sessions, zero form starts. Busy owners click, look, and leave; nobody does homework from a cold email. So the ask was inverted. The pipeline now runs the audit before the first email: every prospect in a batch gets a real, scored report minted at a personal link, and the email opens with their own number — “your company scored 63/100.” The report page carries the $499 upgrade in context, so the artifact does the selling. The full hand-built treatment is held back for warm replies; the batch never gives it away.
03 · Outreach
Deliverability is a design constraint, not an afterthought.
The list comes from an index, not from crawling: local businesses pulled from public map-listings data, enriched with validated email addresses, then hand-trimmed every batch — franchises, doorway sites, and any address whose domain doesn’t belong to the business get cut, because emailing the wrong company poisons the channel. The sends themselves are built to not look like what they are: plain text, no open or click tracking pixels, no tracking domain, a capped daily volume inside business hours, and stop-on-reply. That trade is deliberate — open rates are given up on purpose, and attribution moves to the landing side, where it’s measured properly anyway.
04 · Tracking
One container, one event contract.
The whole measurement layer hangs off three canonical events the site pushes — lead submitted, report viewed, checkout clicked. A tag-management layer fans those out to analytics, the search-ads conversion, and the social retargeting pixels, so adding or swapping a marketing platform never touches the site. The container itself is kept as configuration-as-code: the tag set is generated from one source file, which makes the setup reviewable, restorable, and cloneable to the next niche by swapping IDs. The links feeding this layer follow one discipline: cold emails carry a single clean /go short link, and the redirect appends the UTM parameters server-side — prospects see a clean URL, analytics still gets full attribution. And because a $499 sale often closes days after the click, on another device, ad click IDs are captured at the lead so a later purchase can be imported back into the ad platform as an offline conversion — the design assumes the sale and the click won’t be in the same browser session, because in this market they won’t.
05 · The learning loop
The engine audits itself — three finds so far.
Instrumentation you don’t interrogate is decoration. The rule here: verify measurement against raw evidence, not against whether the setup looks right — and let what it finds change the plan.
Four days after the tag-manager migration, every pixel on the site was silently dead — the loader was present in the page source but shipped in a form the browser treated as inert text. Nothing errored; dashboards just went quiet. A full-funnel audit caught it by checking for the network requests themselves rather than trusting the snippet’s presence, and the fix was re-verified end to end in a clean browser profile. The lesson is now house policy: “the tag is visible in the HTML” is not evidence that it executes.
The first search-ads audit traced roughly $400 of the first $737 spent to broad-match terms pulling generic profile-administration queries from every industry — clicks that could never become HVAC audit buyers, plus a conversion that was quietly counting page loads as leads. Broad match was paused, twenty negative keywords went into a shared list, the fake conversion was removed, and the ad groups were rebuilt around the two intents the funnel actually serves. The honest conclusion got kept too: on-intent search volume is too thin to carry the revenue goal — search ads support the engine; outreach drives it.
The checkout-click log — the funnel’s strongest intent signal — started showing clicks that no analytics event corroborated. Server logs settled it: link scanners and crawlers were following the buy button, one identifiable by probing the URL twice in the same second in a way no human browser does. The fix: the logging path refuses non-human traffic — probe requests, crawler user agents — and crawlers are told to skip the checkout route entirely, because an intent metric that counts robots is worse than no metric.
The evaluation gate was fixed before launch: a specific revenue floor decides whether the model is real, the free-audit-to-purchase rate is treated as a measured baseline, not a pass/fail test, and the possible outcomes — scale it, rebuild the back-end offer, or kill the channel — were written down in advance. The point of running it this way is that even a failure produces a transferable number, and nobody gets to move the goalposts after the fact.
06 · Where it stands
Live, early, and honestly measured.
As of late August 2026: roughly 1,800 cold sends across two campaigns, replies in the low single digits, the first organic leads arriving through the free audit — including one attributed to a search-ad click — and no $499 sales yet. The engine hasn’t reached its revenue gate, and this page doesn’t claim otherwise. What run one has already produced is what it was designed to produce either way: a clean, instrumented baseline — every visit attributed, every reply logged, every click distinguishable from a robot — and a strategy correction (the audit-first inversion) that came out of the data rather than out of opinion. The machine is built so that the next iteration argues from numbers; that discipline, not the current totals, is the deliverable.
Built end to end by Brian Fidler — fractional CMO & marketing technologist, Phoenix, AZ.