Why the usual models fail here

Last-click attribution credits whatever the family touched immediately before the inquiry, which in a long cycle is often brand search or direct, and tells you nothing about what created the demand. First-click credits the earliest touch the platform happened to see, usually on a different device from the one that converted. Platform-reported conversions count the form fill, not the retainer, and each platform counts its own. None of these knows that the inquiry was screened out two weeks later, or signed four months later.

What to record, and where

At the inquiry

The source page, the campaign and click identifiers present at the moment of submission (read from the URL, not from a cookie that your privacy policy says you do not set), the referrer's hostname (never the full search URL, which can contain the family's words), and an event identifier you generate. This travels with the inquiry into the CRM as fields, not as a note.

At every stage transition

Contacted, screened, under review, signed, declined, each with a timestamp, each tied to the inquiry's identifier. This is the whole system; without it nothing downstream can be computed.

Back to the platforms

Offline conversion imports that carry the event identifier, the stage and the date, and nothing else. No condition, no narrative, no child's details. Where a platform cannot accept a stage months later, it receives the screened event instead, which is the closest signal that arrives inside its window.

How to compute it

  • Trailing windows. Cost per signed case for a month is spend in that month divided by cases signed from inquiries that arrived in that month, computed only once the review cycle for that cohort has mostly closed. Report cohorts, not calendar months, and say which cohorts are still open.
  • Leading indicators. Screened inquiries and records requested reconcile within weeks and predict signed cases. Make channel decisions on them while cohorts mature.
  • Assisted paths. Where the same family appears in more than one channel's records (an organic visit, then a paid click), record the path and report both first and last touch side by side rather than choosing a weighting you cannot defend.
  • Brand versus non-brand. Separate them. Brand search inquiries were created elsewhere; crediting them to paid search inflates the channel that merely collected them.

What to print on the report

The measurement gaps are part of the result, not a footnote. Print: the share of inquiries with no source information (phone calls from printed numbers, walk-ins, referrals); the share of cohorts still open; the consent and device limits that make cross-device paths invisible; and any period where tracking failed. A report that shows 100 percent attribution is reporting a model's assumptions as facts.

What not to do

  • Do not set cookies or local storage to "remember" a campaign if your privacy policy promises not to; under-counting your own advertising is cheaper than making that promise false.
  • Do not send medical or case details to any platform to improve matching. The improvement is marginal and the exposure is not.
  • Do not resolve ambiguity by picking the attribution model that makes the current agency look best, including us.

Built this way, the system cannot tell you everything. It can tell you, reliably and defensibly, what a screened inquiry and a signed case cost by channel over a trailing year, which is the only attribution question a managing partner needs answered. Our analytics and attribution service implements exactly this, and nothing more.