Reconcile audience counts before accepting a segment
Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.
An unexpectedly small or large audience needs an explanation, not just a second look at the total. Exclusions may overlap, missing values may behave inconsistently, and different views may count different things. A plausible number can still represent the wrong people.
Before accepting a segment, freeze the starting population and rule version, count distinct contact IDs, separate unresolved records from confirmed exclusions, and prove that every starting contact has exactly one final disposition. Then inspect membership differences rather than relying on changes in totals.
The fictional worksheet below demonstrates that process. All example names, counts, and rule versions are hypothetical. The exercise validates audience composition—not campaign performance, delivery expectations, attribution, or business forecasts.
1. Define what the count represents
Start with a count contract: a short statement of the population and rules being reviewed. Without it, reviewers can compare accurate totals that describe different things.
- Counting unit: distinct contact IDs, not imported rows, phone numbers, organisations, or activity records.
- Starting population: the named list or saved population under review.
- Snapshot: the data cutoff, including its timezone.
- Rule version: the inclusion conditions, exclusions, and treatment of unknown values.
- Final membership: included contacts that pass eligibility checks, with unresolved cases held outside the accepted group.
In this example, the starting population already contains 1,000 distinct contact IDs. Repeated import rows have been handled before this boundary. If your source contains duplicate rows, show row-to-contact deduplication as a separate preliminary reconciliation. Discarded rows are not necessarily excluded people.
Also decide whether multiple contact records representing the same person remain separate. Counting distinct IDs does not resolve identity duplication. Record the limitation rather than silently merging records during review.
A frozen snapshot makes the calculation reproducible, but it can become stale. Recheck current suppression and permission information before any authorised use. Acceptance of a historical count must not override a later opt-out.
2. Give every contact one final disposition
Use a decision order that partitions the starting population without losing records:
- Determine whether the contact belongs in the intended population.
- For included candidates, evaluate known exclusion reasons.
- For candidates without a confirmed exclusion, identify unresolved eligibility checks.
- Accept the remaining contacts as eligible members.
This is an accounting design, not an instruction to stop collecting evidence after the first failed check. Preserve all relevant reason flags, but assign one primary disposition for reconciliation.
In the hypothetical example, inclusion is fully resolved: 200 contacts are outside the target population and 800 are candidates. If inclusion itself were unknown, those records would need a separate unresolved-inclusion bucket—not an assumed match or non-match.
Within the candidates, a confirmed exclusion takes precedence over uncertainty. A contact with a recorded opt-out and conflicting address evidence belongs in confirmed exclusions, not both final buckets. Preserve the address issue as a secondary flag.
For WhatsApp, recipient permission matters and opt-outs must be honoured. Check the current WhatsApp Business Messaging Policy for precise requirements. Accepting an audience does not approve a template or establish delivery; campaign preparation, template review, and actual delivery are separate states.
3. Use this filled reconciliation worksheet
Fictional worked example: all details and counts below are hypothetical. A team is reviewing contacts for a product update. Its documented permission rule separates confirmed absence of qualifying permission from missing or contradictory evidence requiring review. Every exclusion count below is restricted to the included candidates.
FILLED COUNT CONTRACT
Population: Product-update review list
Unit: distinct contact ID
Snapshot: review-day export, 09:00 UTC
Rule version: Product update v3
Inclusion: active account and product interest recorded
Unknown inclusion results: hold for review
Unknown eligibility results: hold outside membership
Primary exclusion order: A, then B, then C
Review owner: Audience operations lead
POPULATION BRIDGE — HYPOTHETICAL COUNTS
Starting contacts 1,000
Outside inclusion conditions 200
Included candidates 800
Candidate dispositions:
Unique confirmed exclusions 200
Unresolved, with no confirmed exclusion 40
Final eligible members 560
Starting-population check:
1,000 = 200 outside + 200 excluded
+ 40 unresolved + 560 eligible
Candidate check:
800 = 200 excluded + 40 unresolved + 560 eligible
EXCLUSION FLAGS — OVERLAPS ALLOWED
A: recorded opt-out 120
B: confirmed no qualifying permission 90
C: fails documented channel-address check 70
Total reason occurrences 280
Distinct contacts with any exclusion 200
EXACT OVERLAP CELLS — MUTUALLY EXCLUSIVE
A only 60
B only 40
C only 30
A and B only 30
A and C only 20
B and C only 10
A and B and C 10
Total distinct excluded contacts 200
PRIMARY EXCLUSION BRIDGE
A, including contacts with other reasons 120
B, excluding contacts already counted under A 50
C, excluding contacts already under A or B 30
Total distinct excluded contacts 200
UNRESOLVED — NO CONFIRMED EXCLUSION
Permission evidence missing or contradictory 25
Channel-address evidence conflicting 15
Total unresolved 40
FINAL MEMBERSHIP
Included candidates 800
Less distinct confirmed exclusions 200
Less unresolved contacts 40
Accepted eligible members 560
REVIEW STATUS
Arithmetic residual 0
Unresolved records held outside membership 40
Record checks and change review: required
Acceptance: pending reviewer decisionThe unresolved reason rows are mutually exclusive here. If those reasons overlap in your data, count their distinct union too. Otherwise, uncertainty can be overstated just like exclusions.
Reuse the worksheet by replacing the population, rule descriptions, ownership, and counts—not merely the final total. Keep a record-level companion file with each contact ID, inclusion result, exclusion flags, unresolved flags, primary disposition, and rule version. Limit its contents and access to what reviewers need. A totals-only worksheet cannot show which records produced the result.
4. Explain why the apparent mismatches are valid
The hypothetical worksheet shows 280 exclusion reason occurrences but only 200 excluded contacts. These figures answer different questions: how many reasons were recorded, and how many distinct contacts were excluded?
The exact overlap cells explain the difference:
- The hypothetical 60 contacts in the three two-reason-only cells contribute 60 extra reason occurrences.
- The hypothetical 10 contacts with all three reasons contribute 20 extra occurrences: each appears three times instead of once.
- Those 80 additional occurrences explain the gap between 280 reasons and 200 contacts.
Subtracting all three reason totals from the hypothetical 800 candidates would leave 520 before unresolved records, or 480 after them. That is not a more cautious audience. It is an accounting error that subtracts some contacts repeatedly.
The primary exclusion bridge avoids that error. Under the stated order, its hypothetical counts of 120, 50, and 30 total 200. Changing the order may change reason attribution, but it must not change the excluded ID set or final eligible membership.
Another apparent mismatch is the hypothetical 600 candidates remaining after confirmed exclusions versus 560 accepted members. The difference is the 40 unresolved records. Calling all 600 eligible would conceal uncertainty.
Finally, document exactly what the channel-address check tests and what causes a pass, failure, or unresolved result. Use that description consistently in the worksheet. Do not treat a passed address check—or acceptance into an audience—as proof of delivery.
5. Investigate membership changes, not just totals
When a count surprises reviewers, compare contact IDs between the previous and proposed populations. A stable total can conceal substantial replacement of members.
New membership = previous membership - removed IDs + added IDs
In a hypothetical comparison, the previous audience contains 550 contacts and the proposed audience contains 560. There are 35 added IDs and 25 removed IDs. The net increase is 10, but reviewers must explain 60 changed memberships.
Classify the changes before sampling: changed source data, inclusion logic, exclusion logic, unknown-value handling, identity mapping, or unexplained differences. Do not assign an unexplained record to the nearest plausible category without evidence.
Where possible, evaluate old and new rules against the same snapshot to isolate logic changes. Evaluate unchanged rules against different snapshots to investigate data changes. If neither comparison is available, record that limitation rather than claiming a precise cause.
Common failure cases include blank values treated as false, an activity join multiplying contact rows, an exclusion calculated against the whole database instead of the candidates, or date boundaries evaluated in different timezones. Each needs record-level evidence. The shape of the total alone cannot identify the cause.
6. Test the explanation with record-level samples
Sampling should challenge the reconciliation, not merely find plausible contacts. Start with unexplained membership differences, then cover the decision boundaries.
- Added members: identify the condition that became true or exclusion that stopped applying.
- Removed members: identify the changed inclusion result, new exclusion, or unresolved check.
- Overlapping exclusions: confirm that all flags remain visible while the contact is subtracted once.
- Unresolved records: inspect the missing or conflicting evidence and assigned owner.
- Unchanged eligible members: check that stable membership still follows the documented rules.
- Boundary cases: inspect blanks, conflicting values, recent changes, and records near date cutoffs.
For each inspected record, capture the source values, expected disposition, actual disposition, and explanation. Mix random selections within groups with deliberately selected edge cases. Random selection reduces cherry-picking; edge cases challenge fragile logic. Neither proves every record is correct.
If a sample contradicts the rule, do not patch only that contact. Identify the affected condition and the full population exposed to it. Correct the rule or source data through the appropriate authorised process, then rerun the reconciliation. Otherwise, the same defect may remain elsewhere.
7. Keep unresolved records visible and owned
Every unresolved group needs an owner and a next review point. In the hypothetical worksheet, the 25 permission-evidence cases go to the permission-record owner; the 15 conflicting address cases go to the contact-data owner. Both groups remain outside accepted membership until their evidence is resolved.
Holding uncertain records makes the accepted audience smaller. Report that trade-off explicitly. The remedy for an unexpectedly small audience is to resolve evidence, not weaken eligibility rules until the total looks familiar.
When a case is settled, move it to confirmed exclusion or eligibility and regenerate the bridge. Retain the earlier snapshot so reviewers can explain the movement. Do not simply delete the unresolved row.
A zero arithmetic residual is necessary but insufficient: incorrect classifications can add up perfectly. Arithmetic establishes completeness of accounting; record checks establish whether the classifications are justified.
8. Apply a concrete acceptance gate
Use this audience count reconciliation checklist before accepting the segment:
- The counting unit, source boundary, snapshot, and rule version are recorded.
- Every starting ID has exactly one final disposition.
- Reason occurrences are separate from distinct excluded contacts.
- Overlap cells or an equivalent distinct-union check explain repeated reasons.
- Unknown values remain visible rather than defaulting into eligibility.
- The final membership ID set matches the accepted eligible count.
- Added and removed IDs have explanations; unexplained differences remain under investigation.
- Record checks support the classifications, and discovered defects have been corrected and rerun across the affected population.
Present the worksheet alongside its ID-level disposition file. Record the arithmetic residual, unresolved count, and unexplained membership changes separately. A reviewer can then distinguish a balanced calculation from justified membership—and accept the eligible group or return it for a specific investigation.
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