Choose a marketing attribution window from observed time between an eligible interaction and the conversion you are evaluating. Start with the conversion event, measure the lag distribution, select a documented primary view, and show the remaining long tail separately. This method is more defensible than applying a generic channel or industry benchmark.
An attribution window changes which interactions may receive credit. It does not identify every conversion, prove that an interaction caused a conversion, or guarantee a change in ROAS, CAC, revenue, or channel performance. Those outcomes must be measured in the relevant dataset.
This guide separates attribution eligibility from report date ranges and available interaction history. It also provides a practical workflow for comparing platform reports, CRM milestones, and mobile measurement settings without treating unlike metrics as if they were the same.
What is an attribution window?
An attribution window is the eligibility period during which a specified interaction may receive credit for a specified conversion under a reporting setup. Name both parts before choosing a duration. A purchase, lead creation, deal creation, and closed-won revenue are different conversion events. A click, impression, view-through interaction, email engagement, and mobile reattribution are different interaction pathways.
Keep these four concepts separate:
- Attribution window: the period in which an eligible interaction can qualify for credit under the selected model and settings.
- Report date range: the dates used to select or display records. The date basis might be conversion date, deal creation date, or deal close date.
- Lookback or interaction-history scope: how far a system can inspect or include interaction history. Its meaning and available controls vary by tool, so do not assume it is identical to an attribution window.
- Conversion lag: elapsed time between an interaction and a defined conversion milestone, or between two CRM milestones such as lead creation and deal close.
A conversion event, a credited touchpoint, and attributed revenue are also different measures. One purchase is one conversion event. A model may assign credit to one or more interactions for that purchase. Revenue attribution concerns the revenue value associated with a conversion under the report rules. Their totals do not have to match.
A 30-day report range, a 30-day interaction eligibility setting, and 30 days of available customer history are not automatically the same period. Record each setting separately before comparing reports.
Click-through and view-through measurement should remain distinguishable. View-through credit can be relevant when impression delivery is reliable and the campaign objective makes impression influence worth testing, but it is not proof of incremental impact. Mobile reattribution is different again. Adjust describes reattribution using conditions such as inactivity and qualifying engagement. Deep linking and deferred deep linking describe routing a person to app content, including after an install; they do not by themselves decide whether an existing user qualifies for reattribution. See Adjust’s reattribution documentation and its deep-link support information.
Choose a window from observed conversion lag
Use timestamped interaction and conversion data to compare candidate windows. For a purchase decision, compare eligible interaction timestamps with purchase timestamps. For a sales decision, use the milestone that matches the question: lead creation, opportunity or deal creation, or closed-won revenue. A lead-to-close interval is not a substitute for a touchpoint-to-lead interval.
A practical analysis uses one row per conversion-touchpoint observation: one eligible interaction paired with one conversion, with elapsed time recorded. Keep conversion events and report aggregates in separate datasets. If one conversion has many interactions, a touchpoint-level distribution gives that conversion more weight. Alongside it, report conversion-level coverage: the share of conversions with at least one eligible touchpoint inside each candidate window. Deduplicate repeated source events before calculating either view.
Illustrative fields for an appropriately joined event or CRM dataset include conversion_id, conversion_type, conversion_timestamp, touchpoint_id, touchpoint_type, touchpoint_timestamp, campaign_id, and source_system. Calculate elapsed time only when timestamps are valid, the interaction type is eligible, and the touchpoint is no later than the conversion. This is a proposed analysis design, not a claim that a vendor supplies a universal export or API with these fields.
Segment the results by conversion type and, where sample size permits, by campaign objective, customer segment, or sales process. Select a primary window that captures the portion of observed conversions your team intends to evaluate. Show interactions or conversions outside that window in a separately labeled long-tail diagnostic view.
For example, a hypothetical purchase analysis might show most conversion-level coverage inside three days and a smaller tail extending to 21 days. The team could use three days for its primary comparison and inspect the 21-day tail separately. The shorter view would change which interactions are eligible for credit. Measure the resulting change in reported channel credit, ROAS, CAC, or revenue using the actual dataset rather than assuming its direction.
When the decision concerns CRM milestones, review missing or inconsistent dates and associations with the CRM owner before selecting a window. Teams investigating that data quality may find CRM systems consulting relevant.
Set a definition contract before comparing platforms
Meta, HubSpot, and a CRM can report different totals for the same calendar dates without any one report necessarily being faulty. Differences may come from conversion definitions, eligible interactions, attribution models, settings, date bases, data access, deduplication, sampling, or modeled results.
Attach a compact definition contract to every report run. Record:
- Conversion event, source, ID, timestamp, and revenue currency where applicable.
- Attribution model, click setting, view-through or engaged-view setting, and eligibility period.
- Date basis, report range, filters, interaction types, and deduplication rule.
- Report run ID, retrieval timestamp, and whether the result is raw, sampled, modeled, or aggregated.
Do not compare a platform conversion count directly with fractional touchpoint credit or CRM revenue attribution as if they were the same metric. Preserve the underlying grain: conversion event, touchpoint, attribution observation, and aggregate report are separate records.
For concurrent imports or reprocessing, retain source IDs and enforce uniqueness at the actual record grain with a database constraint or transactional upsert. A lookup followed by an insert is not sufficient when two workers can process the same record at the same time.
HubSpot’s current documentation describes attribution report types, models, dimensions, filters, data requirements, and report setup. It does not establish a universal 30-day lookback. HubSpot also notes that campaign Performance-tab data and attribution-report data can differ because they use different data sources. Treat a comparison as directional when conversion definitions, interaction eligibility, date basis, or coverage do not match.
HubSpot supports contact-creation attribution in specified Marketing Hub and Content Hub Professional and Enterprise editions. Deal-create and revenue attribution require Marketing Hub Enterprise. Revenue attribution requires closed-won deals, an associated contact, and known Amount, Create date, and Close date values. External interactions generally need qualifying tracking, such as tracked URLs and HubSpot tracking code, to appear in attribution. Verify the account’s current prerequisites in HubSpot’s attribution reporting guidance and report creation instructions.
Use a window that fits the decision, not just the channel
Choose the outcome first, then inspect its timing. A purchase, qualified lead, trial start, deal creation, and closed-won revenue can have different lag distributions within the same business. Business category alone is not a reliable substitute for observed data.
| Scenario | Decision window | What to validate |
|---|---|---|
| Fast purchase | Test a short primary view against observed purchase lag and report the tail separately. | Purchase timestamp, eligible clicks, repeat events, and duplicate handling. |
| Lead or trial | Measure touchpoint-to-lead or touchpoint-to-trial timing rather than time to eventual revenue. | Conversion definition, campaign objective, and segment differences. |
| Extended sales cycle | Use CRM deal velocity and touchpoint data to test a longer view. A 30 to 90 day range is a hypothesis, not a prescription. | Deal-create versus close date, contact associations, and missing CRM dates. |
| Cross-channel journey | Keep one documented primary view and a separate diagnostic for other eligible interactions. | Click, impression, lifecycle, and source coverage. |
Keep the setting used for campaign optimization distinct from the reporting window used for historical comparison. Record the optimization event, selected optimization setting, reporting window, model, and any view-through or engaged-view setting. Platform controls can vary by campaign and event type, so confirm the actual account configuration.
A practical reporting workflow
Use this sequence before changing a reporting definition, campaign budget, or CRM revenue field. It can be carried out with available exports, reports, and CRM data; it does not assume a direct integration between vendors.
A compact definition record can be stored with each report run. The following is an illustrative implementation design, not a vendor-prescribed schema:
{
"source_system": "example_export",
"report_run_id": "run_illustrative_20261013_01",
"record_grain": "one attribution observation per conversion-touchpoint-model-window",
"conversion_definition": "closed_won_revenue",
"conversion_id_scope": "source_system plus source_conversion_id",
"attribution_model": "Time Decay",
"window_definition": "record actual configured scope",
"date_basis": "deal_close_date",
"interaction_types": ["tracked interactions"],
"deduplication_key": "source_system plus conversion_id plus touchpoint_id plus model plus window_definition",
"sampled_or_modeled_status": "verify and record",
"retrieval_timestamp": "2026-10-13T00:00:00Z"
}The proposed key identifies an attribution observation, not a conversion event or a daily aggregate. A conversion-event table would use a source event ID at its own grain. A daily aggregate would also need its account, date, platform, campaign, model, window, and aggregation period. Do not reuse one key across these datasets.
In HubSpot, the documented workflow is to select an available report type such as Contacts, Deals, or Revenue, choose a supported attribution model, dimensions, and filters, then save or export the report. Availability depends on subscription and data requirements. High-volume interactions may be sampled in attribution reports, so use non-attribution reports when exact raw activity counts are required. HubSpot also documents reprocessing after applicable interaction-type changes, which can take up to two days. See its guidance on selecting interaction types.
AI is optional in this workflow. It should not calculate a deterministic timestamp difference, approve a financial attribution record, or select a window without review. If used, its defined output can be a summary of already validated segment-level lag findings for an analyst. The destination should be a reviewed lag report, saved report, dashboard, warehouse table, or comparison workbook, with exceptions routed to the marketing analytics, CRM, paid media, or budget owner responsible for the decision.
When view-through or mobile attribution needs separate treatment
View-through credit is a channel-specific measurement hypothesis. Consider it when impression delivery is recorded reliably and the campaign objective makes impression influence relevant. Keep click-through results visible separately. For high-stakes budget decisions, consider a holdout or other incrementality design rather than treating view-through credit as proof that impressions caused conversions.
Mobile measurement requires a separate distinction. A deep link routes a user to app content. A deferred deep link can route a user after an install where the selected setup supports it. Reattribution concerns whether an existing user becomes attributable again under defined eligibility conditions. Adjust’s documentation describes product-specific settings and support conditions, not a universal deep-link duration. Test the actual configuration using the relevant Adjust deep-link testing guidance.
- Confirm impression delivery is recorded and click-through results remain separately reportable.
- Write down the campaign’s actual view-through, engaged-view, attribution, and optimization settings.
- Test installed-user routing to the intended in-app destination.
- Where supported, test deferred routing after a new install and verify the destination.
- Test reattribution separately from link routing, including inactivity and qualifying-engagement conditions.
- Name the paid media or mobile measurement owner for failed or ambiguous cases.
Frequently asked questions
Can changing the attribution window affect multi-touch attribution?
Yes. It changes which interactions are eligible for the model’s credit assignment. It does not establish that an eligible interaction caused the conversion.
How often should we revisit the window?
Reassess it after a meaningful change in observed conversion timing, conversion definition, campaign setup, sales process, or platform configuration. Preserve the earlier definition so historical comparisons remain interpretable.
Does a longer window always improve measurement?
No. It includes a broader set of eligible history, but additional credit does not by itself establish incremental impact. Compare the primary view with the long-tail diagnostic and use causal testing when the decision warrants it.
Which system should be the source of truth?
Declare a system of record for each business decision. Use another platform as a comparison only when its conversion definition, model, interaction coverage, date basis, and data limitations are understood. HubSpot can be a useful comparison layer, but it is not automatically the universal source of truth.