Marketing attribution

Editorial approval scope: The team checked current source fidelity, scope boundaries, dates, arithmetic, reader usefulness, practical workflow, and general-information limitations.

Attribution is a decision model

A person may see a local-search result, attend a community event, receive a pediatrician recommendation, read a review, and later call the practice. Each touch can matter. Attribution applies a chosen rule to this incomplete record.

Common models include:

• first touch: assigns credit to the earliest eligible recorded source
• last touch: assigns credit to the final eligible source before the defined event
• multi-touch: distributes credit across eligible recorded sources
• self-reported source: records what the person identifies as influential
• incrementality test: compares predeclared groups or periods under a suitable design

The first four describe recorded journeys. A well-designed comparison can address causal questions more directly, though operational and ethical limits still apply.

Define the outcome before the source

Attribution for a first inquiry differs from attribution for completed intake, assessment, first service, or collected revenue. Name the event and unit. Keep clinical recommendation, payer authorization, capacity, family choice, claim adjudication, and payment as separate states.

Use a maturity window long enough for the event. A channel that produces slower referrals will look weak when the report closes early. Keep open records visible by age and next action.

Build a controlled source taxonomy

Use durable categories such as organic local search, paid search, community partner, referring clinician, school, payer directory, social media, event, existing-family word of mouth, and unknown. Store campaign, content, location, and partner detail in separate fields.

Map tracking parameters, call numbers, form fields, referral records, and self-report to the taxonomy. Retain raw values so mapping changes can be audited. “Google” may mean an ad, Maps profile, organic result, or remembered search; preserve uncertainty instead of forcing precision.

The SBA market-research guide encourages analysis of demand, location, saturation, and competitors. Attribution is one first-party signal. It omits people who never contact the practice and undercounts unrecorded offline influence.

Protect identity and privacy

Collect only data needed for the stated decision. Define consent or permission for cookies, call recording, texts, email, analytics, and audience tools. Restrict access, retention, export, and vendor reuse. Avoid sending diagnosis, treatment, insurance, or free-text intake details into general advertising platforms.

For a HIPAA covered entity, HHS marketing guidance explains that certain marketing uses or disclosures of protected health information require authorization, subject to defined exceptions. Confirm entity role and each data flow. A vendor contract or tag-manager setting cannot decide legal scope.

Reconcile identities without double counting

One person may submit a form, call, and receive a partner referral. Apply a written identity rule and merge only when the evidence supports a match. Keep households, siblings, and separate service requests distinct where appropriate.

Record unmatched and ambiguous journeys. Overaggressive merging erases valid people; weak matching inflates leads and lets several channels claim the same outcome.

A fictional attribution example

Beacon ABA, a fictional practice, locks a cohort of 20 unique inquiries with a full 30-day review window. First-touch rules assign eight to local search, five to community partners, four to paid search, two to payer directories, and one to unknown. The counts total 20.

Last-touch rules assign six to local search, four to community partners, six to paid search, three to payer directories, and one to unknown. Paid search receives 4 of 20 first-touch credits and 6 of 20 last-touch credits.

Twelve inquiries complete intake review. Seven reach a conditional onboarding offer. The practice reports source mix for all 20, review outcomes for all 20, and offer results for the mature cohort. It avoids stating that paid search caused six inquiries or any offer.

Interpret gaps before moving budget

Compare platform-reported events with unique practice leads. Investigate missing tracking parameters, blocked cookies, shared devices, offline referrals, forwarding numbers, duplicate forms, delayed data, and staff overrides. Reconcile spend, source, and outcomes by the same dates and geography.

The FTC advertising FAQ says objective claims need support before release. Attribution results rarely justify public outcome or superiority claims. Internal budget decisions also deserve clear limitations.

Report models side by side

Useful fields include cohort, event, unit, model and version, lookback window, eligible touches, identity rule, raw source, mapped source, unknown count, numerator, denominator, cost, mature outcome, and refresh date.

Show first touch, last touch, and self-report side by side when they materially differ. Consistent disagreement is information: it can reveal a long journey, missing early data, or a final channel that captures demand created elsewhere.

Before shifting spend, require a reconciled mature cohort, a documented sensitivity check across plausible models, and a review of access and source quality. Record the decision, owner, amount, expected observation window, and reversal trigger so attribution remains an auditable planning input rather than an automatic verdict.