You have written an invitation you care about. Now comes the uncomfortable part: deciding who should receive it. Your list contains people who downloaded a worksheet, people you coached last year and people who bought the programme yesterday. They should not all be treated as if they are at the same point.
For your Manchester training practice, useful segmentation starts with the message you intend to send. You are not trying to label every detail about every person. You are trying to explain why this invitation is relevant, who must be excluded and what should happen when someone’s situation changes.
Build the rule in ordinary language first. Then check whether your current sending workflow can express it and show you the resulting audience. If a required condition is not available, make it a configuration or build requirement; do not quietly replace it with “send to everyone”.
Can you explain this audience before the send?
Use the invitation you are actually preparing.
- What does this message help the recipient decide?
- Which evidence establishes relevance, and which establishes eligibility?
- What exact offer-specific state excludes an existing buyer?
- How are overlap, restrictions and late changes handled?
- Can the operator explain each test record’s inclusion or exclusion?
If the rule cannot be expressed or verified, keep the campaign unactivated while the necessary configuration or build is completed.
This is an audience-design method, not proof that every described rule already exists in a particular product.
- Topic candidates
- Assumed count: 200
- How to interpret it: Matches the assumed topic rule; eligibility still needs checking.
- Existing buyers
- Assumed count: 30
- How to interpret it: Exclude from this acquisition invitation.
- Restricted contacts
- Assumed count: 20
- How to interpret it: Apply the relevant restriction independently of topic interest.
- Overlap
- Assumed count: 5
- How to interpret it: Appears in both exclusion groups; count only once.
- Distinct exclusions
- Assumed count: 45
- How to interpret it: 30 + 20 − 5.
- Remaining candidates
- Assumed count: 155
- How to interpret it: 200 − 45; not a certified sendable audience.
All inputs are assumptions. The rule must also account for current permissions, data quality and any relevant changes before sending.
1. Begin with the reason for the message
Write a one-sentence purpose: “Invite people interested in giving feedback to consider the next programme, unless they already bought it.” That sentence gives you the ingredients of a segment. It also exposes what you do not yet know: how the interest was recorded, which purchase counts and whether this person is eligible for marketing.
Keep relationship, interest and permission separate. “Past learner” can help you choose an example; “requested feedback worksheet” can explain relevance; neither automatically answers whether this marketing message is appropriate. For a UK audience, confirm the applicable recipient and message rules using current ICO guidance rather than assuming every business address follows the same rule.
Choose the minimum information needed to make this decision. A segment that depends on a private inference about someone’s confidence or career risk is harder to explain and maintain than one based on a topic they explicitly requested. Relevance should feel useful, not as if you have been watching the person too closely.
- Why this matters
- A message-led rule connects selection to a legitimate purpose and prevents segmentation from becoming a collection of labels with no operational use.
- Work through the example
- Use an explicitly selected “feedback skills” interest for the invitation. Keep coaching delivery notes out of the marketing rule.
- When the approach changes
- A general service announcement to affected customers has a different purpose from a promotional launch. Classify the message before applying the marketing segment.
- Your decision
- If you cannot explain why the person belongs in the audience in plain language, refine the rule before sending.
2. Separate the next useful step from the last label applied
Start with a small set of operational states: interested but not yet a buyer, currently participating, and finished the relevant programme. These are examples, not a universal lifecycle. Define what event places someone in each state and what event moves them out.
Be precise about the offer. A person who bought an earlier course may still be a suitable prospect for a different one. A person who bought this programme yesterday should not receive another “last chance to join” message just because a general interest tag remains on their record.
Use the most reliable evidence available for the decision: a confirmed purchase record, a stated preference or an agreed current engagement. Do not use a lifecycle label as a substitute when its update rules are unclear. If the state depends on data in another module, check that relationship rather than assuming the two screens stay in step.
- Why this matters
- State rules prevent a useful historical label from overriding what is true now. The relevant event and the specific offer matter more than a broad “customer” badge.
- Work through the example
- Define “already bought” as the verified purchase of the programme being promoted, not any purchase at any time. Decide how a cancellation or refund changes the communication plan.
- When the approach changes
- Some existing participants may need an invitation to a separate follow-on programme. Excluding every customer would remove a relevant audience.
- Your decision
- Use offer-specific evidence and explicit entry/exit conditions; keep uncertain records for review.
3. Apply exclusions before you celebrate the audience size
Write exclusions alongside inclusions. Restrictions come first. Then consider offer-specific exclusions such as existing buyers, people receiving the same invitation through another route, or people whose situation makes the message irrelevant. Review the final audience, not only the group that matched the first condition.
Overlap is ordinary. One person can be an alumnus, a worksheet reader and a coaching client. Decide whether they should receive one message, a different version or none. Do not solve overlap by deleting their useful history; solve it in the rule and deduplicate the resulting recipients.
Check changes that happen after audience preparation. Someone may opt out or purchase while a campaign is waiting to send. Your requirement should describe what happens then, not only what the preview looked like an hour earlier. Verify that behaviour in a safe environment before relying on it.
- Why this matters
- Inclusion finds possible relevance; exclusion removes people who should not receive this message. Applying both to current evidence is what makes the segment operationally useful.
- Work through the example
- Create test cases for an interested prospect, an existing buyer, an opted-out alumnus and a person matching two interest groups. Write one expected outcome for each.
- When the approach changes
- An exclusion can be too broad. Removing all alumni may prevent a useful advanced-course invitation. Tie the rule to this offer and purpose.
- Your decision
- Treat the final deduplicated audience and late-change behaviour as acceptance checks, not optional housekeeping.
4. Use behaviour as a clue, not a personality diagnosis
Open tracking is not a dependable statement that a person read your message. Mail privacy features can fetch remote content independently of a deliberate read. That makes “opened three times” a weak foundation for deciding someone is eager, and “did not open” a weak reason to assume they are uninterested.
A click, reply or confirmed purchase may be more directly relevant to a particular decision, but keep the limits visible. A click is not automatically a human buying signal; a reply needs context; a purchase of one offer is not interest in every other offer. Prefer explicit preferences and useful actions where you have them.
Avoid building an intricate engagement score before you can maintain its inputs. A small, explainable rule can be more useful than a sophisticated-looking label whose behaviour nobody can describe. Give stale or uncertain signals an expiry or review point instead of letting them define the person forever.
- Why this matters
- Tracking records events generated by systems, not a complete account of human intention. Matching the signal to the decision reduces false certainty.
- Work through the example
- Use a requested topic preference to shape the invitation. Review a reply personally rather than treating an open-count threshold as a substitute for that conversation.
- When the approach changes
- Behavioural data can still identify a broken link or a broad pattern. The limitation is the inference, not a requirement to ignore every event.
- Your decision
- Name the observed event and its limits. Do not turn it into an unsupported claim about the person.
5. Make the rule usable by the person sending the campaign
Give each segment a purpose, an owner and a plain-language definition. Include the data source, exclusions and review date. Your assistant should be able to explain why a test record is included without asking you to reconstruct a decision from three months ago.
Gbeya’s broadcast workflow includes particular source-based groups. Check the available controls against your intended rule: those groups do not establish every purchase, interest and engagement combination described here. Keep any missing condition as a specific requirement before activating the campaign.
After the campaign, review the intended outcome and the audience quality separately from delivery diagnostics. Did suitable people ask the next question or buy the relevant programme? Did current participants receive an unnecessary invitation? Record what you can verify, fix the rule and retire segments that no longer serve a useful purpose.
- Why this matters
- A segment is a maintained business rule. Ownership and test cases make it repeatable; outcome review shows whether it is worth keeping.
- Work through the example
- Name a group “Feedback programme invitation: eligible, interested, not already bought” and attach its test cases. Keep the internal name descriptive rather than clever.
- When the approach changes
- A one-off campaign may not need a permanent saved segment. Save the rule and evidence without adding another live automation to maintain.
- Your decision
- Keep a segment only when its purpose, inputs and expected behaviour remain understandable and useful.
Useful segmentation is explainable restraint
Sending-platform documentation demonstrates that inclusion and exclusion can be separate controls. That is a useful operating model, not proof that finer segmentation always produces more sales. For a small training practice, the rule needs to justify the maintenance it creates.
Privacy features undermine simple open-based assumptions, while privacy research raises wider questions about how email addresses are used and shared. The practical response here is to use relevant, explainable evidence and avoid collecting detail merely because a tool permits it.
Our baseline is to start with message purpose, preserve restrictions and test a small number of meaningful states. More complex rules can be built when the actual campaign needs them; complexity is not the goal.
Medium confidence in the applied strategy; high confidence that recorded opens are not equivalent to verified human reading.
No segmentation uplift percentage, customer experiment or blanket compliance claim. No audience was imported or messaged.
Audience-rule example with explicit overlap
Count exclusions once
Assume 200 test contacts match the topic. Thirty already bought this offer and twenty are restricted; five people appear in both exclusion groups. The excluded union is 45, leaving 155 candidates. That remaining number still needs the other eligibility and data-quality checks.
30 + 20 − 5 = 45 distinct exclusions. 200 − 45 = 155 candidates. These are assumed inputs, not a live audience or forecast.
Subtracting 30 and 20 without accounting for their overlap gives 150, which is incorrect for these assumptions. Different overlap produces a different result.
Reconcile recipient-level rules and deduplicate. A larger number is not automatically a better audience.
Assumptions worth checking
A tag means the person can receive marketing.
A topic or relationship label does not supply missing eligibility evidence or override a restriction.
More segments mean more relevant marketing.
Each rule needs a purpose and maintenance. Unclear, stale groups can make messages less appropriate.
Repeated opens identify the hottest buyers.
Recorded opens can be affected by privacy technology and are not proof of reading or buying intent.
An experienced marketer can infer buying intent from an engagement score alone.
Experience cannot restore context that the underlying events never captured. Check explicit preferences and relevant buyer actions before treating a score as intent.
Your next 30 days
Days 1–7
Define one invitation
Write the message purpose, relevance evidence and eligibility conditions. Inventory the inputs you actually have.
Ready for the next step: The rule is understandable without reference to a tool’s interface.
Days 8–14
Model states and exclusions
Define offer-specific buyer states, overlap and restrictions. Create representative test records.
Ready for the next step: Every test case has an expected include/exclude/review result.
Days 15–21
Verify implementation
Check the exact controls and late-change behaviour in a safe environment; complete required configuration or build.
Ready for the next step: The final audience matches the defined rule without activating production sends.
Days 22–30
Review and maintain
After separately authorised campaign execution, inspect outcomes and inappropriate inclusions; update or retire the rule.
Ready for the next step: The segment has an owner, evidence and next review date, without invented attribution.
Questions before you proceed
What is the difference between a tag and a segment?
A tag is a stored label. A segment is a selection rule or resulting group. Neither replaces the evidence needed for the particular message or overrides an opt-out.
Should I exclude every previous customer?
No. Use the specific offer and purpose. Someone who bought this programme differs from someone who bought a different course and may find this one relevant.
Can I build segments from opens alone?
That is a weak basis for inferring interest. Privacy features can affect open records. Prefer explicit preferences and relevant verified actions, with their limits understood.
How many segments should I create?
Start with the few rules that change a useful message or exclusion. Add more when the decision requires them, not to make the dashboard look sophisticated.
Does Gbeya already support every rule in this guide?
Gbeya includes particular source-based audience groups. Verify the controls against your intended rule; any additional condition needs configuration or development before the campaign relies on it.
What if someone buys after I prepare the audience?
Define whether that purchase should remove them and verify the late-change behaviour before relying on it. The same care applies to new restrictions.
Sources and further reading
- ICO: Electronic mail marketing rules
- Mailbox-provider sender requirements for personal accounts
- Email-platform contact import and subscription-status documentation
- Mail privacy protection and background content loading
- Email-platform audience inclusion and exclusion controls
- Why You've Got Mail: Evaluating Inbox Privacy Implications of Email Marketing Practices in Online Apps and Services
Prepared by Gbeya · Sources reviewed 30 October 2026.