A measurement system that follows a visitor from the ad to the approved customer. Site events, ad spend, and CRM outcomes land in one warehouse the client owns, and every campaign and landing-page test carries a verdict — keep, kill, or spend this much more to know — computed with proper statistics, not read by eye.
Designed for a direct-to-consumer telehealth brand running Meta campaigns across several product tracks. The client is named here once they consent; until then this page describes the system, not the client.
Nothing is judged on clicks. Verdicts are computed from cost per approved customer against an allowable figure the client sets from margin and payback.
| The question | How it is decided | What the client sees |
|---|---|---|
| Is this campaign earning its spend? | A probability model of cost per approved customer, seeded from the client’s own history and updated daily; the rule of three (3× allowable cost with no customers) is the floor before any kill. | Keep · Kill · Spend $X more to decide — with a probability, never a bare average. |
| Which landing page wins? | Visitors are split at random and stay on their variant; each test is registered up front with its metric, duration, and stopping rule; approval rate and plan mix must not fall. | Adopt · Drop · Keep running — with the confidence level and the cost of being wrong. |
| Which ads produce customers? | Only the ad platform’s own split tests are fair comparisons; ads are judged on first-party cost per approved customer, not clicks. | Creative ranked by customers produced — with a range. |
| How far can a winner scale? | Budget rises in steps held for two weeks; the cost of the extra customers is read separately from the average. | Marginal cost per customer — before the next step. |
| What have we learned? | Every test’s effect is pooled by element type — headline, image, offer, price framing — across tests. | A running list of what moves this audience — and by how much. |
The same evidence standard every DaVeenci case exposes: input, the manual workflow it replaces, systems, gates, finished output, exceptions, and current status.
Pages, clicks, quiz steps, photo upload, and checkout, plus the campaign that brought the visitor — recorded first-party, only after opt-in consent.
Ad spend, the CRM export, and a working sheet that never quite reconciled. Whether a campaign was working was a judgment call made weekly.
To be pulled into one client-owned Postgres warehouse every day. Customers appear as pseudonymous IDs; names and emails never enter the analytics store.
The rules produce a queue of actions — raise or cut a budget to a stated amount, turn a campaign off or on. A person executes each one in the ad platform and marks it done. Nothing changes automatically.
What changed, which tests concluded, what to do next — reviewed by a person. Every change to a campaign or page is logged with its reason.
The acceptance test: a deliberately broken tag raises an alert in under an hour, and spend running ahead of plan or a funnel step dropping does the same — instead of surfacing at month-end.
Each landing-page test states its metric, duration, and stopping rule up front, with approval rate and plan mix as guardrails. Results pool by element type — headline, image, offer, price framing — so the client learns what moves conversion.
Package one — the measurement plan and data boundary — is the gate for everything after it. Status flips to "in delivery" when the client kicks off, and the client is named here only with consent.
Each package has a plain description, a purpose, and an acceptance test the client runs. Tiers are cumulative; anything can be bought one at a time.
One trustworthy view of the funnel and a keep / kill rule the client can calculate.
Learn faster: experiments, creative-level attribution, alerts, and a weekly decision brief.
Keep the pipeline fed and every platform change deliberate, explained, and traceable.
Every package ships with a named acceptance owner on the client side, a measurable acceptance test, reconciliation against the source system, documented rollback, and alerting — the same definition of done as every other build.
Health and regulated-data engagements are designed to the strictest state, then checked against what the tags actually fire. The design reduces risk; it does not replace counsel.
Customers appear as pseudonymous IDs. Quiz answers, photos, and diagnoses are on the list of things never recorded, written down before package one is accepted.
The ad pixel and tag manager load only after consent; every tag is held to an enforced allowlist of what it may send, audited quarterly against what actually fires.
Public, internal, or sensitive — decided per event, with the privacy policy reconciled to the real tag behaviour.
Outcomes and decisions are never overwritten; corrections are new rows. Code, documentation, and accounts belong to the client from day one.
Not legal advice — whether a given law applies to a client is a question for their counsel.
If the answer lives in three spreadsheets and one person's judgment, that is the workflow to bring. Thirty minutes, no deck.
Verdicts with probabilities, gates before actions — the same discipline applied to trading research.