Conversion and Revenue Tracking
Build conversion and revenue tracking that connects organic and AI-referred sessions to outcomes, catches broken funnels, and supports honest attribution.
SEO earns attention with rankings and visibility, but it keeps budget by showing business outcomes. If the team cannot connect search or AI visibility to qualified demand, orders, subscriptions, pipeline, or realized revenue, finance sees an expense with an interesting dashboard. That line is easier to cut than a program with a traceable result.
Phase: P15 · Conversion and revenue tracking. Stage: D · Measure. Timebox: 3–5 working days for a site with working analytics and a connected revenue system; allow 1–2 weeks when CRM stages, checkout events, consent behavior, or historical identities need repair. Owner: the analytics or revenue-operations lead is accountable, with SEO defining channel questions, engineering implementing events, and finance approving the revenue definition.
This phase builds a trustworthy chain of observations, identifies where it is incomplete, and makes attribution assumptions visible enough to challenge.
Why this phase, and why here
Conversion and revenue tracking follows the build and promotion work because it consumes the final URL map, releases, campaign dates, target segments, citation destinations, and frozen baseline measurement . Earlier goals, access, consent rules, and revenue definitions determine which outcomes matter and which comparisons remain valid.
It sits before reporting cadence for a dependency reason: a recurring report can only repeat the measurement system beneath it. If events fire twice, CRM opportunities cannot be joined to sessions, refunds are counted as new revenue, or AI referrals are folded into direct traffic without disclosure, a polished monthly deck scales the error. Three months later the team has a quarter of internally consistent but false history.
Running it too early instruments a draft funnel or obsolete URL structure. Run it after conversion paths are stable enough to test, but before the first result reallocates budget.
Inputs and outputs
The outputs are a contract with the next phase. Reporting may visualize them, but it may not quietly redefine them.
| Direction | Item | Acceptance condition |
|---|---|---|
| Input | Approved outcomes and funnel | Each stage has a business meaning, owner, source system, and valid state transition. |
| Input | URL, page-type, campaign, and release map | Organic and AI landing pages can be segmented, and material changes have timestamps. |
| Input | Analytics, consent, CRM, billing, and commerce access | Coverage dates, identifiers, timezones, currencies, retention, and known gaps are recorded. |
| Input | Frozen baseline and channel definitions | The comparison window, organic scope, brand rules, and starting revenue basis cannot change silently. |
| Output | Measurement plan and event dictionary | Every event names its trigger, parameters, deduplication key, owner, test evidence, and downstream use. |
| Output | Funnel integrity report | Critical paths have observed counts, stage rates, reconciliation results, defects, and retest status. |
| Output | Attribution specification | The primary model, comparison views, lookback windows, identity rules, exclusions, and limitations are explicit. |
| Output | Organic and AI outcome dataset | Sessions, leads, orders, pipeline, and realized revenue are segmented without treating unknown traffic as zero. |
| Output | Reporting handoff | Metric definitions, approved thresholds, evidence links, owners, and a dated sign-off are ready for recurring use. |
Ready to Monitor Your AI Visibility?
Track how AI chatbots mention your brand across ChatGPT, Perplexity, and other platforms.
The checklist
Complete these items in order. Each gate asks whether another person can reproduce the result, not whether the dashboard looks plausible.
1. Define the outcome hierarchy and economic source of truth
What: define primary conversions, supporting conversions, funnel stages, and the revenue values the program will report. A primary conversion is the business outcome being funded, such as a paid order, activated subscription, or sales-qualified opportunity. A supporting conversion is evidence of progress, such as a product demo request or checkout start.
Why: teams overstate impact when they add unlike actions together. Ten newsletter signups are not ten purchases, and booked pipeline is not realized revenue. The hierarchy preserves the distinction between intent, qualification, sale, and cash.
How: document the valid path from visit to outcome. For each stage, name the authoritative system, timestamp, status rules, currency, tax and shipping policy, refund treatment, and whether value means gross revenue, net revenue, recurring revenue, pipeline, or margin. Use finance-approved realized revenue for the primary view. If customer lifetime value is modeled, show its inputs and keep it apart from collected revenue.
Tool: analytics plan, CRM stage documentation, billing or commerce platform, and finance ledger.
Done when: every reported outcome has one definition, one source of truth, one owner, and one calculation; supporting actions cannot enter revenue totals; and finance signs off on currency, refunds, cancellations, and recognition timing.
2. Build the event and conversion dictionary
What: specify the event tracking needed to observe each funnel transition, then designate which validated events count as conversion tracking outcomes.
Why: event names alone do not define behavior. A generate_lead event might fire on a button click, a successful form response, or a thank-you page reload. Those implementations produce different counts and can reverse a performance conclusion.
How: create one row per event with its question, trigger, parameters, allowed values, systems, identifier, deduplication key, consent dependency, failure behavior, and owner. Prefer confirmed server outcomes for purchases and accepted leads; keep UI interactions diagnostic. Version definition changes rather than overwriting history.
Tool: tag manager or application instrumentation, analytics debugger, browser network panel, server logs, CRM, and billing or commerce webhooks.
Done when: 100% of primary and supporting conversions map to documented events; every revenue event has a stable transaction identifier and value/currency fields; every parameter has an allowed type; and a reviewer can distinguish intent from confirmed completion without reading implementation code.
3. Test every critical funnel path and failure path
What: run end-to-end tests for successful, rejected, repeated, canceled, and resumed journeys across the devices and consent states that matter.
Why: a happy-path test misses the failures that poison reporting: double submits, payment retries, thank-you page reloads, blocked scripts, validation errors, CRM deduplication, refunds, and cross-domain checkout. These defects often preserve believable totals, making them harder to notice.
How: test desktop and mobile, consent states, anonymous and logged-in users, organic and known AI-referral landings, form and order failure, duplicates, refunds, and cross-domain returns. Follow one identifier through browser event, analytics, CRM or order record, and revenue report, recording expected and actual counts.
Tool: analytics debug view, browser developer tools, server log, CRM sandbox, test payment or store order, and a QA evidence sheet.
Done when: every in-scope critical path passes with exactly one accepted conversion and the correct value; failed or abandoned attempts create no primary conversion; duplicate and reload tests add no second outcome; refunds and cancellations reach the approved reporting state; and every failed case has an owner and retest date.
4. Reconcile the funnel before trusting rates
What: compare event counts and values between adjacent systems and calculate stage-to-stage rates. Reconciliation means explaining why two sources that describe the same business activity differ.
Why: a conversion rate can improve because a start event stopped firing, not because more people completed. Revenue can rise because currency conversion changed, an import was repeated, or the selected date uses payment time in one system and order time in another. Integrity checks catch the break before it enters a quarter of reports.
How: reconcile analytics conversions to accepted CRM leads or orders, then reconcile subscriptions, refunds, and revenue to billing or finance. Compare counts, transaction IDs, values, currencies, timestamps, and statuses. Measure missing IDs, duplicates, impossible sequences, and unknown values. Document expected loss from consent, blocking, timezones, or latency; investigate rather than force equality.
Tool: warehouse query or spreadsheet, analytics export, CRM export, billing or commerce export, and Open Economics .
Done when: transaction IDs are unique, all primary conversions follow valid stage order, 100% of reported revenue has a recognized currency, daily source differences are within the approved tolerance, every difference outside tolerance is explained and owned, and the seven-day comparison has no unexplained break or step change.
5. Connect organic and AI-referred sessions to outcomes
What: preserve the acquisition evidence needed to segment outcomes from organic traffic and visits referred by AI answer products.
Why: AI traffic is not a clean, universal channel. Some products send a recognizable referrer, some use redirectors or embedded browsers, some strip context, and a buyer may return later through branded search or direct navigation. Calling every direct visit “AI” invents evidence; ignoring known AI referrals hides a real contribution.
How: maintain versioned rules for search engines, known AI referrers, campaign tags, redirects, and internal exclusions. Capture original and session source, landing URL, tags, a citation or prompt identifier when available, and the first-party lead/account ID. Persist original acquisition into CRM. Treat unrecognized traffic as unknown or direct, not inferred AI. Keep citation-page correlation separate from identified sessions.
Tool: analytics acquisition reports, server logs, CRM fields, AmICited Revenue Attribution , and Open Revenue Attribution .
Done when: 100% of observed sessions enter one documented channel bucket; known AI referrers have tested rules; original and session source survive the lead or order handoff where consent permits; unknown values remain visible; and a test organic visit and test tagged AI visit reach the correct outcome segment without overwriting one another.
6. Choose attribution views and state their limits
What: select one primary attribution model for stable trend reporting and define comparison views for first touch, last non-direct touch, and assisted outcomes. An assisted conversion is an outcome where a channel appeared in the observed journey but did not receive primary credit.
Why: attribution is allocation, not causation. Last-touch favors channels near the transaction. First-touch favors discovery. Multi-touch attribution distributes credit but depends on the observed touchpoints and weighting rule. No model sees every device, offline conversation, word-of-mouth exposure, or privacy-restricted interaction.
How: document lookback, direct handling, cross-device identity, offline imports, reporting time, and reopened opportunities. For short ecommerce cycles, compare order-level first and last touch. For long B2B cycles, preserve first acquisition, record opportunity creation and close separately, report lead-created cohorts, and separate open pipeline from won revenue. Use controlled holdouts, geographic tests, or time interventions to test incremental impact.
Tool: analytics attribution reports, CRM opportunity history, warehouse model, Revenue Attribution, and experiment documentation.
Done when: the primary model and lookback window are frozen for the reporting period; first, last, and assisted totals are labeled and never added together; open pipeline is separate from won revenue; model exclusions appear beside the result; and the same raw conversions reconcile across every crediting view.
7. Publish the decision-ready economic view and monitoring gates
What: combine validated conversion, revenue, cost, and attribution outputs into the views used for prioritization and recurring reporting.
Why: a technically correct dataset still fails if decision-makers cannot see which page, segment, prompt, or action produced an outcome—or whether the number is strong enough to act on. Conversely, a ranked list without data-quality status invites budget changes based on a broken feed.
How: report by landing page, page type, topic, business line, market, device, and identified source where volume permits. Show conversions, revenue, pipeline, refunds, costs, and return on investment with denominators. Put freshness, coverage, model, and reconciliation beside each result. Alert on disappearance, duplication, value shifts, unknown-channel growth, and connector failure. Suppress recommendations when a critical gate fails.
Tool: Cockpit at Open Cockpit , Economics, Revenue Attribution, warehouse reporting, and the issue queue.
Done when: each decision row links to its definition and source; every metric has a period and denominator; critical data failures visibly block recommendations; named owners receive an alert within one business day; and a second analyst can reproduce the page- or channel-level total from the approved exports.
Tools in AmICited
AmICited supplies three connected views. Use them after event and revenue sources pass integrity checks.
| Product step | Deep link | Use it for | Preserve as evidence |
|---|---|---|---|
| Revenue Attribution | Open Revenue Attribution | Connect trials, orders, subscriptions, and revenue with AI answers, prompts, and cited landing pages where the journey is observed. | Date range, model or method, confidence, prompt, cited page, outcome, revenue, and export time. |
| Economics | Open Economics | Reconcile orders, revenue, costs, status mappings, and the economic basis behind performance. | Currency, status rules, unmapped values, realized revenue, costs, refunds, and source coverage. |
| Cockpit | Open Cockpit | Review what moved economic performance and which rule-based actions crossed a threshold. | Comparison window, driver values, data-health warnings, action threshold, and report timestamp. |
The matching feature pages explain Revenue Attribution and Cockpit . Product attribution and a platform’s claimed conversion count remain separate columns; neither overwrites the revenue system of record.
Stay Updated on AI Visibility Trends
Get the latest insights on AI mentions, brand monitoring, and optimization strategies.
Decision rules
These are integrity gates, not industry benchmarks. Change a tolerance only with data-owner approval; do not relax it to make a report pass.
| Finding | Bad threshold | Decision | Done when |
|---|---|---|---|
| Duplicate primary conversion | More than 0 for the same transaction or lead ID | Block affected conversion and revenue reporting | Duplicate rate is 0 in test and every production duplicate is removed or explicitly explained. |
| Missing transaction or lead ID | More than 0.5% of primary conversions | Investigate; block page-level attribution above 2% | The trailing seven days are at or below 0.5%, or the limitation is approved and affected detail suppressed. |
| Analytics-to-system-of-record count variance | More than 5% daily for 2 consecutive complete days | Open incident and suspend trend claims | Variance returns within 5% or every difference is reconciled to consent, latency, exclusions, or status rules. |
| Revenue reconciliation variance | More than 1% against the finance-approved total | Block revenue and ROI publication | Currency, refunds, taxes, cancellations, and timestamps reconcile within 1%. |
| Unknown currency | 1 or more revenue records | Block affected value | Every included record has a supported currency and approved conversion rule. |
| Invalid funnel sequence | 1 or more primary outcomes before their required prior stage | Block affected funnel rate | All records follow valid state transitions or a documented exception. |
| Unknown channel share | Above 10% of outcome value, or increase of 5 percentage points week over week | Investigate classification and identity handoff | Cause is explained, rules are corrected where possible, and unknown remains labeled. |
| Event volume discontinuity | Drop above 30% day over day with no matching traffic or release explanation | Treat as possible tracking failure | Deployment, seasonality, outage, or genuine behavior explains the movement and a test event passes. |
| Stale connector or export | No successful update for more than 24 hours on a daily report | Mark data stale and suppress recommendations | Freshness is restored and missing periods are backfilled or visibly marked. |
| Long B2B lookback | Shorter than the 90th percentile of observed lead-to-close time | Do not use the model for channel exclusion | Window covers the observed cycle or the excluded tail is quantified beside the result. |
| Assisted versus primary credit | Values added together | Reject the report | Primary and assisted views are separate, labeled, and reconcile to the same unique outcomes. |
A threshold catches likely defects; it does not establish causality. Post-release revenue movement remains an association without an incremental design.
Deliverable
Hand over one versioned measurement package with exportable tables. It contains five artifacts:
OUTCOME AND EVENT DICTIONARY
Outcome | Event | Trigger | Required parameters | Allowed values
Source system | Destination | Deduplication key | Consent rule | Owner | Version
FUNNEL INTEGRITY REPORT
Test case | Device/consent state | Expected events | Actual events
Analytics count | CRM/order count | Revenue count | Variance | Defect | Retest evidence
CHANNEL AND ATTRIBUTION SPECIFICATION
Organic rules | Known AI referrers | Campaign rules | Unknown handling
Primary model | Comparison models | Lookback | Identity rule | Exclusions | Limitations
ECONOMIC DATASET
Period | Segment | Landing page | Source | Outcomes | Assisted outcomes
Pipeline | Realized revenue | Refunds | Costs | Currency | Coverage | Quality status
MONITORING AND SIGN-OFF
Check | Threshold | Frequency | Alert owner | Response time
Evidence links | Analytics approval | Revenue-operations approval | Finance approval
The handoff is accepted when an analyst can reproduce totals, finance can trace revenue, engineering can rerun critical tests, and SEO can distinguish identified outcomes from assisted, inferred, unknown, and direct activity.
What goes wrong
The thank-you page is treated as the sale. Reloads and failed payments create conversions. Use the accepted server transaction and deduplicate its ID.
Every form interaction becomes a lead. Keep clicks and errors diagnostic; count only a lead the receiving system accepts.
The dashboard matches itself. Comparing two analytics views repeats the same defect. Reconcile to CRM, commerce, billing, or finance.
Unknown visits are relabeled as AI. A post-citation spike is context, not session-level proof. Report identified AI referrals separately.
Last touch erases discovery. Keep assisted and first-touch views when branded search or direct return gets final credit, without calling allocation causation.
Open B2B pipeline is reported as revenue. Show pipeline by stage and cohort; keep won and realized values separate.
Lookback ends before buyers convert. Base the window on observed lead-to-close time and show the open cohort.
Refunds and cancellations disappear. Apply approved status and recognition rules; distinguish gross from net.
A consent or connector change creates a performance story. Annotate tracking changes, monitor unknown and missing-ID rates, and suppress conclusions until integrity returns.
The model changes when inconvenient. Freeze the primary model for the period; label alternate views.
Next phase
The next phase is Reporting cadence and annotations. It needs a signed measurement package, not screenshots copied from live dashboards. The reporting owner receives:
- the outcome and event dictionary, including version dates and owners;
- the approved primary attribution model, alternate views, lookback window, and explicit limitations;
- the organic and AI channel rules, including unknown handling and identity constraints;
- reconciled baseline and current datasets with revenue, refunds, costs, pipeline, coverage, and quality status;
- the integrity thresholds that suppress a claim or trigger an incident;
- the release, campaign, connector, consent, and tracking annotations needed to interpret change.
Recurring reporting can begin when the same inputs reproduce the same totals and a failed integrity gate is visible before any recommendation. It waits when finance has not approved the revenue basis, critical tests are failing, or attribution views cannot reconcile to unique outcomes.
FAQ
Frequently asked questions
Which conversions should SEO report?
Which attribution model is best for SEO?
How should we measure a long B2B sales cycle?
Can we identify every visit from an AI answer engine?
When is revenue tracking ready for budget decisions?
More tutorials in this section
Ready to put it into practice?
Free check · 7-day trial · no credit card