PublicationsDominate IgniteResearch 09

Research 09 / Attribution

Attribution Is Not Causation

This paper examines the limits of marketing source data in a healthcare consultation funnel. It separates what a recovered campaign or partner field can prove from what budget, partner credit and return claims require.

Setting
A multi-location healthcare services organization
Period
November 2026
Scope
Campaign and partner attribution across booking and form journeys
Evidence
Source code inspection; source-reported test outcomes; synthetic journeys

Published By Dominate Ignite / Engineering Research

The Question

When a booking system reliably records a campaign or partner source, what does that record establish about whether the source caused the appointment or produced incremental revenue?

The Thesis

Recovering source identifiers is a necessary data-integrity intervention, but it is not proof that a campaign caused an appointment or that an appointment produced incremental revenue. In a healthcare acquisition environment, consent, sensitive data, multiple touchpoints, clinical eligibility, appointment attendance and downstream care decisions all constrain what marketing telemetry can legitimately establish. The final causal inference does not follow automatically from successful attribution capture.

01

The Institutional Decision Problem

The source package describes a WordPress, Elementor and NitroPack site with Zoho Bookings and Zoho Forms, advertising pixels, GTM and a strict Zoho CRM partner picklist. It reports that paid campaign identifiers were lost on some booking journeys and that partner visits could overwrite campaign context. The core runtime script introduced decoupled paid and partner storage, URL decoration and interception of booking and form links. These mechanisms address whether the organization can observe an acquisition path. They do not independently determine which exposure changed a patient’s behavior.

For an executive allocating acquisition budget, the distinction separates three questions. The first is whether a source was recorded. The second is whether that source was a meaningful contributor to the observed conversion. The third is whether the conversion would have occurred without that source. The first can be tested through software and records. The second requires a defensible attribution model. The third requires causal identification, usually a credible counterfactual.

The analytical sequence therefore runs from capture to persistence to association to incremental effect to economic contribution. Each step supports a narrower claim than the one after it, and none of the later claims follows automatically from success at an earlier step.

Claims hierarchy: the minimum evidence each claim requires and the inference it permits.
ClaimMinimum EvidencePermissible Inference
Tracking parameter was attachedDecorated URL or controlled DOM testBrowser-side propagation only
Field was persistedRedacted Zoho record and field lineageRecord-level capture
Channel drove appointmentsComplete denominator, deduplication and channel policyDescriptive association
Campaign increased bookingsRandomized holdout or credible quasi-experimentEstimated incremental effect
Campaign increased profitIncremental revenue, margin and full costsEconomic contribution under assumptions

02

What The Implementation Actually Does

The version 3 source of the core runtime script uses sessionStorage for campaignKeys and cookies for selected identifiers and partner state, along with a URL compiler that places parameters inside Zoho Bookings hash-routed URLs or Zoho Forms query strings. The implementation can preserve a paid-media session while storing a partner independently. It also uses a fallback click_id that prioritizes fbclid, then gclid, then ttclid. This is an explicit operational precedence, not evidence that the first available click identifier caused the booking.

The source register reports synthetic end-to-end journeys and 100% URL decoration (as reported in the source package; not independently measured). Those are source-reported outcomes, not independently audited CRM population rates. The code further reveals that a new URL containing any campaign key clears the prior session campaign set and then repopulates only the keys present. A partial campaign visit can therefore produce a hybrid state in which historical context has been discarded. That behavior may be intentional, but it needs a written attribution policy and test cases.

Observed design decisions and their tradeoffs.
MechanismBenefitResidual Failure Mode
Separate paid and partner slotsAvoids automatic partner overwriteMultiple credit claims can coexist
90-day partner cookieCarries partner across sessionsConsent, ITP and retention variability
Hash-router URL decorationPreserves Zoho route semanticsDoes not prove CRM commit
First nonempty click_id precedenceProvides deterministic field valueSuppresses multi-channel click history
Generic slug partner inferenceSupports partner landing pagesFalse partner classification possible
Global DOM interceptionCovers several mutation pathsCompatibility and ungoverned paths remain

03

Causal Architecture And Competing Hypotheses

Confounding and measurement mechanisms can each distort channel-level conclusions. A booking following a paid click can arise because the ad induced new demand, because an already motivated visitor sought the brand, because a partner introduced the patient before the ad or because multiple channels jointly affected the decision. These hypotheses produce similar observed CRM records but materially different budget recommendations. A deterministic source field resolves none of them without further design.

Counterfactual hypothesis register.
HypothesisPredicted EvidenceDiscriminating Test
H1: Paid media creates incremental demandBookings rise beyond baseline under exposureRandomized geographic or audience holdout
H2: Paid media captures existing demandAttributed bookings rise without total liftBranded-search and holdout analysis
H3: Partners generate first interestPartner-first journeys precede paid clicksConsent-permitted journey cohort study
H4: Technical repair changes visibility onlyAttribution completeness rises while bookings stay flatPre and post record completeness and volume
H5: Faster booking changes conversionCompletion rate rises at a stable mixControlled page or time-series design

04

Measurement Design: Define The Denominator

A source-reported 100% recovery can mean that every governed URL was decorated in test cases, that every eligible form submission carried the expected hidden fields or that every resulting CRM record was populated. These are distinct denominators. The source materials do not furnish a complete production-level denominator for all booking sessions, all consent states, all devices or all CRM outcomes. The research must therefore avoid substituting one coverage metric for another.

Recommended measurement dictionary.
MetricNumerator And DenominatorInterpretation
URL decoration coverageDecorated eligible targets over eligible targets inspectedClient-side technical coverage
Field persistenceRecords with expected fields over eligible submitted recordsCRM data integrity
Source reconciliationRecords matching journey evidence over tested recordsCross-system correctness
Booked consultation rateConfirmed bookings over eligible sessionsFunnel conversion
Attendance rateAttended appointments over confirmed bookingsOperational yield
Incremental bookingsTreatment bookings minus counterfactual estimateCausal contribution
Incremental contributionIncremental gross margin minus incremental acquisition and service costEconomic impact

05

Identification Strategy And Test Protocol

The strongest design would use an ethically and operationally appropriate randomized holdout, with clinical services and patient access unaffected. Where randomization is impractical, a difference-in-differences or interrupted time-series design may be considered only if its identifying assumptions are plausible and checked. A before-and-after chart alone is not sufficient, because campaigns, seasonality, service offerings, pricing, site content, device mix and referral programs can change simultaneously.

The evidence matures in stages, from code inspection to controlled journeys to record reconciliation to causal design. Each stage supports a narrower claim than the stage that follows it, so a publication should state which stage its evidence has reached.

Recommended controlled test cases.
JourneyExpected Technical OutcomeFailure To Monitor
Paid, then partner, then bookingPaid fields and normalized partner both presentOverwrite or false precedence
Partner, then direct returnPartner persists within policy windowCookie blocked or expired
Paid campaign A, then paid campaign BNew campaign policy applied consistentlyPartial UTM loss
Consent denied, then bookingNo impermissible tracking; booking still worksPremature SDK initiation
Mobile cold or warm reloadSame attribution and a single booking iframeRace or duplicate event
Booking confirmation reloadOne logical conversion event per bookingDuplicate Schedule or Lead
Form submission to CRMSynthetic fields reconcile to final recordHidden field not committed

06

Privacy, Consent And The Evidentiary Boundary

Healthcare attribution data can be sensitive even when a URL contains no explicit clinical keyword. The source code initializes Meta and TikTok tracking before the consent management platform loads, and the source package itself identifies consent gating as an outstanding requirement (see Case Study 02). The same source notes that Google Ads, GA4 and GTM intentionally receive native URLs. A claim of universally sanitized outbound telemetry would therefore exceed the supported evidence.

Whether HIPAA applies depends on covered-entity and business associate status, on the character of the information and disclosure and on the surrounding transaction; those are questions for counsel. Whether FTC and state consumer-health privacy obligations apply independently is also a question for counsel. Technical field capture does not authorize collection or sharing. A defensible measurement design must start with a data inventory, a lawful-basis and permission assessment, vendor agreements, data minimization and tests across granted, denied and withdrawn consent states. The source package included no external legal review.

Data minimization controls.
ControlRequired DecisionAssurance Artifact
Collection allowlistWhich fields are genuinely neededApproved field taxonomy
Consent enforcementWhen each vendor may initializeConsent-state test evidence
Identifier lifecycleHow long each ID may persistRetention and deletion policy
Cross-border and vendor routingWhich processors receive which dataVendor data-flow register
Event deduplicationWho owns Lead, Schedule and CRM eventsEvent ownership matrix
Sensitive-page handlingWhich page contexts must not be exportedNetwork captures and negative tests

07

Adversarial Examination

A CFO may challenge whether recovered attribution fields justify greater media spend without demonstrated incrementality. A regulator may ask whether the tracking itself was permissible, regardless of sanitization. A technical auditor may challenge the absence of raw HAR files and final CRM reconciliation. A skeptical buyer may ask whether the observed improvement reflects engineering changes or concurrent marketing activity. Each challenge is legitimate and requires a different evidentiary response.

Challenge, failure mode and fallback.
ReviewerPrincipal ChallengeDefensible Fallback
CFOAttribution is not return on investmentReport data quality and run a separate lift study
Privacy counselCollection and disclosure may be impermissibleRestrict tags pending legal assessment
Technical auditorSynthetic journeys do not prove production coverageRun controlled end-to-end reconciliation
Growth leaderLast-touch credit obscures partner valueUse multi-touch descriptive views with caveats
Client executiveMeasured speed does not prove revenue gainReport verified speed separately from conversion

08

Economic Model Without Invented Outcomes

The engagement’s commercial register provides indicative remediation value ranges and a hypothetical rebuild comparison (see Research 11). It does not provide independently verified incremental bookings, realized revenue or contribution margin. A defensible financial model must therefore remain parameterized.

For illustration only, incremental contribution can be expressed as ΔB × A × M - C, where ΔB is incremental booked consultations, A is attendance-to-revenue realization, M is contribution margin per realized patient and C is incremental campaign, platform and servicing cost. No numeric return should be published until the variables are supported.

Sensitivity dimensions.
VariableLow-Case PressureHigh-Case Pressure
Incremental appointmentsNo causal lift despite better captureDemonstrated holdout lift
Appointment attendanceHigh no-show or cancellation rateStable verified attendance
Revenue realizationConsultation without paid careMeasured conversion to service
Attribution qualityMissing fields and duplicatesReconciled, consent-permitted records
Incremental costHigh media and operating costEfficient acquisition and delivery

09

Governance And Implementation Architecture

The organization should maintain separate owners for tracking configuration, consent rules, CRM schema, conversion-event definitions, engineering release control and business interpretation. One person may perform several of these functions, but the decision rights must remain explicit.

Every material change to the core runtime script, the legacy multi-vendor header snippet, GTM, Zoho forms or booking configuration should trigger a synthetic journey suite, negative privacy tests, cache-state checks and CRM reconciliation before release.

Operational decision register.
DecisionAccountable RoleTrigger Or Gate
Approve attribution precedenceRevOps ownerNew channel or partner
Approve vendor data exportPrivacy and legal ownerNew tag or payload
Release runtime changesEngineering ownerRegression suite passed
Approve performance claimsAnalytics ownerComparable traces available
Approve incremental ROI claimFinance and analyticsCausal design and cost evidence
Approve named publicationClient executive and Dominate IgniteWritten release and redaction review

10

Sources And Evidence Boundaries

The paper draws on three source families. The first is the full version 3 source of the core runtime script, including campaignKeys, partner normalization, decorateTargetUrl, pre-hydration and SDK wrappers. The second is the engagement evidence register. The third is a code supplement containing the legacy multi-vendor header snippet before and after remediation and the form-message bridge. No independent production access, HAR capture or CRM extract was obtained for this paper.

Pixel IDs, portal IDs, raw patient data and private administrative screenshots are excluded, and the organization is not identified.

  • Documented: the code contains URL decoration, state persistence and event interception.
  • Source-reported: 100% governed URL decoration, improved calendar readiness and improved CRM visibility.
  • Not independently verified: universal field completeness, absence of sensitive third-party disclosures, causal appointment lift, financial returns and any legal determination.

What The Publication Establishes

  • Recovering campaign and partner identifiers improves what the organization can observe, but it does not establish which exposure caused a booking.
  • The implementation applies an explicit click-identifier precedence and a campaign reset rule that can discard earlier context, so both need a written attribution policy and test cases.
  • A reported 100% recovery figure is meaningful only with a stated denominator, and the source materials do not supply a production-level denominator.
  • Incremental bookings and economic contribution require a counterfactual design, such as a randomized holdout, and finance-approved inputs before any return can be published.
  • Privacy permissibility is a separate gate that the source package itself leaves open, and it belongs with counsel rather than with the attribution model.

What It Does Not Claim

  • This paper does not report a measured lift in bookings, revenue or return on advertising spend.
  • It does not claim complete attribution; the 100% decoration figure is source-reported and was not independently measured.
  • It is not a legal determination of HIPAA, FTC or state-law applicability to any party.
  • It relies on supplied source code and a narrative evidence register, without independent production access, HAR captures or CRM extracts.
  • The economic formula is illustrative and carries no supported numeric inputs.