Set an attribution window that matches the campaign decision
Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.
A campaign can keep collecting credit long after it has stopped influencing the reporting decision. The opposite problem is just as costly: a short cutoff can make ordinary response delays look like campaign failure.
Choose a campaign attribution window by defining the decision first, measuring response and outcome delays in fully observed cohorts, and comparing plausible cutoffs. Document separate response and outcome windows, including what starts each clock. Then test whether changing those cutoffs changes the action you would take.
The objective is not to find the window that produces the most credited outcomes. It is to establish a consistent, useful measurement rule. Attribution records an association under that rule; it does not establish that the campaign caused the outcome.
1. Start with the decision the report must support
Before choosing a duration, finish this sentence: We will use this report to decide whether to… The answer determines which delays matter and how long the business can wait.
- Assess immediate interest: Focus on qualifying responses. A final sale may be too distant to inform the decision.
- Compare campaign approaches: Use consistent eligibility rules and equally mature observation periods. Do not give one campaign more time to collect outcomes.
- Review commercial results: Allow time for the relevant outcome to occur, while keeping a bounded relationship to the original campaign.
- Make an early operational decision: Use a clearly provisional response report rather than presenting incomplete outcome totals as final.
For example, a hypothetical appointment campaign might generate replies quickly but require additional time for appointments to be completed. A response window appropriate for measuring interest would not automatically be suitable for measuring completed appointments.
Write down the decision owner, decision date, and counted outcome before inspecting alternative windows. Otherwise, it is easy to choose the cutoff that makes a preferred campaign look strongest.
2. Define two clocks, not one ambiguous window
A response window answers: How long after the campaign anchor can a qualifying response enter this campaign's measurement cohort? An outcome window answers: How long can an eligible outcome take to occur? These are different questions.
Define the following fields explicitly:
- Campaign anchor: The recorded event from which elapsed time is measured. Delivery can be a useful anchor when reliable delivery records are available. Do not silently substitute preparation, approval, or attempted sending.
- Qualifying response: A recipient action that indicates the interest being measured. Exclude administrative replies and opt-outs from positive response counts; retain them separately for their proper handling.
- Qualifying outcome: A specific recorded event, such as a completed appointment, rather than an undefined status such as engaged.
- Clock convention: Whether the outcome deadline runs from the campaign anchor or from the qualifying response.
- Counting unit: Unique recipients, accounts, orders, or another defined unit. State how repeats are treated.
An anchor-based outcome window gives everyone the same elapsed-time limit from campaign exposure. A response-based outcome window gives each qualifying responder the same time to progress, but later responders extend the final reporting date. Either can be useful; switching between them without disclosure makes reports incomparable.
Also specify the time zone and boundary rule. For example, include events occurring at or before the exact deadline timestamp, using elapsed hours rather than an ambiguous end-of-day cutoff.
3. Estimate ordinary lag from mature cohorts
Use cohorts that have already had enough observation time to cover your longest candidate window. A recent campaign cannot tell you how many late outcomes it will eventually record.
- Collect the campaign anchor timestamp, first qualifying response timestamp, qualifying outcome timestamp, and identifiers needed for deduplication.
- Calculate elapsed time from the anchor to the response and, separately, to the outcome. If considering a response-based outcome clock, calculate that interval too.
- Count cumulative unique responses and outcomes at several candidate cutoffs.
- Inspect what is added between cutoffs, not just the final total.
- Check whether late records reflect genuine delay, delayed data entry, or questionable linkage.
A flattening response curve suggests that extending the window adds relatively little measured interest. It does not prove that all meaningful responses have arrived. Likewise, a long outcome tail may reflect a normal sales process, but it may also contain unrelated activity that happens to involve the same recipient.
Keep event time separate from entry time. If a completed appointment is entered into the CRM later, its completion timestamp should determine eligibility where that timestamp is trustworthy. A reporting-processing delay is not necessarily a customer-response delay.
Inspect distributions for materially different campaign purposes before combining them. Pooling a short-lived offer with a considered purchase can produce a cutoff that represents neither. Create separate rules only when the operating difference justifies them, not whenever a campaign's results look inconvenient.
4. Filled hypothetical window-selection worksheet
Fictional worked example: All details and numbers below are hypothetical. They illustrate a measurement design, not customer results or a recommended universal duration.
A team is reviewing an appointment campaign. It wants to decide whether to repeat the same approach in its next planning cycle. Its selected decision criterion is at least 35 completed appointments among eligible responders, subject to a separate capacity and cost review.
Worksheet: definitions and readiness
- Decision owner: Campaign operations manager.
- Decision: Whether the campaign meets the appointment-count criterion for consideration in the next cycle.
- Population: 1,000 unique recipients with recorded delivery anchors.
- Observation readiness: Every recipient has at least 35 elapsed days of observation after delivery, and records have been reconciled.
- Response event: First reply explicitly requesting an appointment. Opt-outs and administrative replies are excluded from this measure.
- Outcome event: First completed appointment after the qualifying response.
- Clock: Both response and outcome deadlines run from each recipient's delivery timestamp.
- Counting rule: Each recipient contributes at most one response and one completed appointment.
- Linkage: The appointment must link to the campaign response record. Ambiguous links are reported separately, not automatically credited.
Worksheet: candidate windows
- Short — response through day 3; outcome through day 7: 90 cumulative qualifying responders; 18 attributed completed appointments. Fast to finalise, but excludes substantial later response and completion activity.
- Medium — response through day 7; outcome through day 21: 132 cumulative qualifying responders; 39 attributed completed appointments. Captures most observed responses and allows additional time for completion.
- Long — response through day 14; outcome through day 35: 144 cumulative qualifying responders; 43 attributed completed appointments. Adds 12 responders and 4 completions relative to the medium pair, while delaying final reporting.
The short response window captures 62.5% of qualifying responders observed by day 14. The medium window captures approximately 91.7%. These are shares of the responses observed within this hypothetical observation horizon—not estimates of all responses that could ever occur.
Worksheet: written choice
Use a 7-day response window and a 21-day outcome window, both measured from recorded delivery. The short pair excludes substantial observed activity and changes the appointment-count decision. The long pair adds comparatively few completions and does not change that decision. Treat this retrospectively selected rule as provisional, freeze it before the next comparable campaign, and review it after the next three mature comparable cohorts.
This choice is useful because it states what is counted, why the extra waiting time matters, and when the rule will be reconsidered. It does not claim that the credited appointments were caused by the campaign.
5. Change one cutoff at a time
The worksheet compares practical short, medium, and long configurations. Because both windows change together, it does not show which cutoff drives the difference. Run a second check holding one window fixed.
In the same hypothetical example, keep the response window at 7 days:
- Outcome through day 7: 22 completed appointments.
- Outcome through day 21: 39 completed appointments.
- Outcome through day 35: 41 completed appointments.
Then keep the outcome deadline at day 35:
- Response through day 3: 29 completed appointments.
- Response through day 7: 41 completed appointments.
- Response through day 14: 43 completed appointments.
These checks show that both very short response eligibility and very short outcome observation suppress the hypothetical result. Extending beyond the medium choice adds relatively little and does not change whether the 35-appointment criterion is met.
Always translate sensitivity into a decision statement. If a ranking, threshold result, or recommended action changes under plausible windows, label the conclusion cutoff-sensitive. Report the range and investigate the delayed records before making a strong recommendation. Do not conceal sensitivity by publishing only the most favourable configuration.
6. Resolve overlap without extending credit indefinitely
A longer window increases the chance that recipients encounter other campaigns before responding or completing an outcome. A cutoff alone cannot resolve that overlap.
Choose a credit rule separately from the window. For example, a team could assign primary credit to a campaign with a directly linked qualifying response and keep other touches as context. If several campaigns have equally plausible links, place the outcome in an ambiguous category or apply a documented allocation rule.
Whatever the choice, avoid giving multiple campaigns full primary credit for the same outcome in an aggregated total. If a report intentionally shows overlapping associations, label it as non-additive so readers do not sum it into a unique-outcome count.
Keep late activity visible without reopening credit forever. An outcome outside the window can remain in the customer record and appear in a late-outcomes report. Excluding it from the campaign's primary measure does not mean deleting it or denying its business value.
7. Recover when the data or reporting process fails
Immature cohorts: Mark results provisional and show how much observation time is complete. For comparisons, use cohorts that have all reached the relevant deadline, or compare the same elapsed-time checkpoint. Never treat not-yet-observed outcomes as confirmed failures.
Missing anchors or broken links: Separate unmeasurable records and report their count. Repair records where evidence supports the correction; do not invent timestamps or attach outcomes merely because they occurred nearby in time.
Backfilled outcomes: Document an ingestion allowance after the event deadline. This gives records time to arrive but does not extend event eligibility. Publish a reconciliation date and distinguish revised totals from the original snapshot.
Results change after a rule revision: Version the definition and, where feasible, recalculate the comparison period under the new rule. If recalculation is not possible, identify the break rather than presenting one continuous trend.
Late activity keeps growing: Inspect its source before extending the window. A longer genuine completion process may justify a longer outcome window. Repeated campaigns, weak linkage, or delayed entry may instead require better attribution rules or data repair.
8. Publish a window contract with the next report
Attach a compact measurement contract to the reporting specification: decision, population, anchor, response definition, outcome definition, window lengths, clock convention, duplicate handling, overlap rule, maturity requirement, reconciliation date, owner, and version.
Your next action is to fill that contract for one campaign and reproduce the three-window worksheet using a mature cohort. Record the extra responses and outcomes each extension adds, then write whether the business decision changes.
For the next comparable campaign, freeze the rule before results arrive. At the planned review, check three observable things: whether late eligible events remain material, whether missing or ambiguous records undermine confidence, and whether plausible alternative cutoffs change the recommended action. Those checks provide a concrete reason to retain, repair, or revise the window.
Put the next conversation to work.
Explore Alactic Communications for WhatsApp conversations, campaigns and AI-assisted workflows. Choose the product and plan that fit your team.
Explore Communications