The starting point for the decision
Sign-ups are growing, but first purchases remain low. Customers who have purchased are not returning, and a list of long-inactive customers is accumulating. If the person responsible cannot operate all of these campaigns at once, where should the first automation go?
Start with one journey that addresses the current business bottleneck and has customer eligibility, data, a meaningful offer, operational capacity, and measurement ready. The largest list or easiest template should not determine priority. If no candidate meets those conditions, narrow the audience, test the hypothesis manually, or defer automation.
A first-purchase objective does not require starting with purchase reminders. If new users have not experienced the service’s value, helping them complete an initial use may come first. Repeat purchase becomes a candidate when customers who recognize the value of their first purchase are missing the next opportunity. Reactivation becomes a candidate when previously satisfied customers have a new reason to return. The following framework and three organizational examples explain how to choose.
Count customers eligible for this message, not names on a list
Ask every candidate the same question: Who should receive this message now, and why? Compare the number of unique customers for whom you can answer that question. Do not use all historical sign-ups, installed devices, or past purchasers as the campaign audience.
Check the tool’s counting unit as well. OneSignal allows one user to have multiple device and channel Subscriptions. Segment screens show subscription and opt-out states by channel, and some audience sizes are estimates. A displayed number therefore does not automatically mean “distinct customers who will receive a message.” The Users[1] and Segments[2] documentation explains this distinction.
When defining the audience, resolve duplicate identities, verify eligibility for this message and channel-level reachability, then exclude customers who have already completed the goal or are receiving support. Separately reconcile the business’s marketing consent and withdrawal records with the tool’s technical subscription state. OneSignal email Subscriptions default to Subscribed when created unless configured otherwise; that label should not substitute for checking consent records. Subscriptions[3]
“No purchase event” also requires care. If you cannot distinguish a genuine non-purchaser from a missing event, eligibility for a first-purchase campaign is uncertain. Likewise, someone who continues purchasing through another channel should not be classified as inactive merely because app activity stopped. First narrow the scope to channels, products, and customers you can verify, rather than increasing send volume.
From customer data to campaigns ready to build — Review data and measurement, then define key campaigns and an implementation sequence around business goals.
Eliminate unworkable candidates, then choose one that addresses the bottleneck
The table below is an editorial framework for prioritization, not a model for calculating expected campaign revenue. Instead of adding scores to produce a ranking, defer candidates with uncertain eligibility or data. Among the remaining candidates, choose one that directly addresses the current bottleneck, is manageable for the team, and produces observable results.
| Decision factor | First purchase / initial activation | Repeat purchase | Reactivation |
|---|---|---|---|
| Actual audience and eligibility | Customers who signed up or showed interest but have not completed the defined first use or purchase | Customers whose first purchase was fulfilled successfully and for whom a relevant next purchase remains incomplete | Customers with demonstrated past value, no current activity, and a reason to return |
| Bottleneck to improve | Are instructions, the first step, or purchase-decision information missing? | Is the timing, product, or usage context of the next purchase unclear? | Has the previous reason for leaving been resolved, or has a newly relevant offer become available? |
| Required data | Sign-up, completion of core use, purchase status, and linked customer identity | Orders, fulfillment, cancellations, refunds, purchased products, and subsequent purchases | Past core use, current activity, interests, and the relevance of the return offer |
| Action the business controls | Make the first task easier to start or provide information needed for a purchase decision | Explain a genuinely useful next product or usage opportunity | Describe improved features, new schedules, or a concrete reason to use the service again |
| Cost of failure | Pressure customers who already used or purchased, or offer unnecessary discounts | Recommend further purchases to dissatisfied or returning customers, or discount before demand exists | Remind customers of unresolved problems or send irrelevant offers |
| Operational workload | Usage questions, account and access problems, and pre-purchase consultation | Product suitability, delivery, exchanges, refunds, and benefit questions | Reopened complaints, booking availability, and account recovery |
| Ability to measure effects | Can completion of the first value experience or first purchase be recorded? | Can additional purchases be assessed alongside cancellations, refunds, and costs? | Can resumed use or purchasing be verified beyond a return visit? |
When candidates are otherwise similar, consider the one requiring fewer assumptions before the one reaching more people. Clean data alone should not keep the business focused on a minor issue, however. If the important bottleneck is clear but automation is not ready, a small manual test and data improvements for that bottleneck can be the first task.
Three organizations can answer the same question differently
All organizations, audience counts, and situations below are design examples. They are not IXC customer cases, industry averages, or sample sizes established as sufficient for an experiment.
A learning service with many sign-ups: first value before first purchase
A hypothetical learning service has 5,000 sign-ups in the period under review. Among them, 1,200 customers have verified identity and messaging eligibility but have neither completed the trial task nor paid. Assume customer interviews identified the problem: “I do not know what to try first.” Meanwhile, the repeat-purchase audience is small, and the reasons past users left are unknown.
This team first chooses an activation journey that helps customers complete one trial task. Rather than offering a discount before customers can judge the paid product’s value, it provides a link to start the task and guidance for likely sticking points. Task completion is the primary goal; the first payment is a separate downstream metric. Increased activation is not interpreted as increased revenue.
Different conditions change the decision. If a login error prevents customers from starting, fixing the product comes before a campaign. If customers who finish the task still cannot explain the product’s value, review the value proposition and product experience rather than adding more messages.
A retailer that fulfills first orders well: repeat purchase in a narrow product category
A hypothetical retailer has 800 customers who received and used their first order successfully but have not purchased again. This count follows verification of messaging eligibility and exclusion of customers with unresolved delivery or return inquiries. Assume actual usage and purchase records, together with customer responses, establish a follow-up purchase opportunity for a specific product category. The overall inactive-customer list is larger, but old product codes and customer identities do not reconcile.
This team first chooses repeat-purchase guidance for that category. Timing follows the usage context established by this business and customers’ selected reminder timing, rather than an industry-wide rule such as “30 days after delivery.” The team also checks stock availability and whether the next product is appropriate.
Success is not determined by the occurrence of a second order alone. Define how cancellations and refunds will count, then assess incentive costs and additional support work. If the need for a follow-up purchase is hard to explain, deferring repeat-purchase automation and investigating the usage experience may be more appropriate.
A booking service with a new reason to return: selected customers rather than the entire inactive list
A hypothetical hobby-class booking service has 10,000 past users. For this review, it selects 300 customers who stopped booking after classes at their preferred time disappeared, have records of repeated attendance and relevant interests, have no current bookings, and meet messaging eligibility requirements. Assume classes at their preferred time have now resumed.
This team first chooses a reactivation journey for customers who match the new schedule. The reason is “you can attend at your preferred time again,” rather than “you have been away, so here is a coupon.” The goal is resumed bookings and attendance, not simply another visit.
Inactivity is not defined using one universal number of days since last access. Consider the value the customer last received, why the opportunity to participate stopped, and whether the customer currently has other bookings or usage. Do not start this campaign if class availability remains unreliable or past complaints remain unresolved.
The three examples choose activation, repeat purchase, and reactivation respectively. Their priorities differ because the problems they can change now and the evidence available to them differ, not because of list size.
Automation requires capacity for what happens after sending
Automating message delivery does not automatically resolve replies, exceptions, refund inquiries, or opt-out updates. When choosing the first journey, determine both “how many days will it take to build?” and “who will handle which responses?”
A workload calculation example: assume a batch of 500 recipients, an inquiry rate of 4%, and eight minutes to handle each inquiry. The estimated workload is 500 customers × 0.04 inquiries/customer × 8 minutes/inquiry = 160 minutes, or two hours and 40 minutes. This is workload under stated assumptions, not a forecast of the actual inquiry rate. Pre-send review, exception checks, status corrections, and result reviews require additional time. If the operator lacks that capacity, split batches or reduce the audience.
Consider conflicts between campaigns as well. OneSignal Push frequency capping is a push-only limit available on some paid plans; it does not cap total contact across email, SMS, and in-app messages. Capped pushes are not placed in a queue for automatic later delivery. As channel count grows, distinguish the business’s overall contact policy from what the tool actually controls. Push frequency capping[4]
Do not infer re-entry or opt-out behavior from feature names. OneSignal journeys triggered by Custom Events can allow concurrent entries by the same user and differ from segment-based re-entry settings. Opting out of one channel does not necessarily terminate every other channel in the journey. The first journey’s operator must be able to check repeated entries, channel-level subscription states, and the scope of withdrawal requests received by the business. Journey settings[5], Journeys overview[6]
Assign a person and authority to stop sending. Even when one person runs the operation, define a policy for coverage gaps, such as limiting new entries while that person cannot process inquiries or verify subscription states.
If these checks are difficult to manage, avoid connecting many channels and branches from the outset. Start with a verified audience, one objective, and a manageable path, then expand. Manual testing is not a way to bypass messaging eligibility; it tests whether an appropriate offer can be made to verified customers.
Messages connected to customer behavior — Use lifecycle stages and behavior to connect channels and timing, then refine journeys through experiments.
Define success, stopping conditions, and the observation period together
“Automated delivery worked correctly” and “customer behavior improved” are different forms of success. The former verifies operations; the latter verifies impact. Keep click-through rate as a supporting indicator that the path works, and choose completion of the behavior relevant to the journey as the primary metric.
To distinguish impact, consider whether you can maintain a comparison group that receives no message: a holdout. Customer.io’s Holdout tests[7] documentation distinguishes comparisons between message variants from comparisons with an unmessaged group, and recommends evaluating conversions rather than opens or clicks for holdouts. OneSignal also supports a message-free Split Branch path for comparison. Check the actual configuration, including whether re-entering customers are randomized again. Journey actions[8]
You do not need to finish a complex experiment design during prioritization. You should, however, be able to answer whether the target behavior can be recorded, a comparison group can be maintained, and observation can continue long enough to support a decision. If not, limit the first run’s objective to checking audience, offer, and operational feasibility rather than demonstrating impact.
Suppose the retailer above makes the following decisions. All periods and thresholds below are design examples, not recommended standards.
| Agreement before starting | This team’s example decision |
|---|---|
| Primary objective | The proportion of unique customers completing a second valid order within the observation period, out of all eligible customers assigned to each group. Do not change the denominator to successful message recipients; predefine how cancellations and refunds affect valid orders. |
| Comparison and success | Randomly assign eligible customers to messaged and unmessaged groups, aligning exposure to other campaigns as far as possible. Define in advance the minimum additional purchasing that justifies costs and workload, then assess uncertainty in the results. |
| Supporting metrics | Cancellations and refunds, incentive and delivery costs, support time per customer, opt-outs, and complaints |
| Observation period | Fourteen days per customer, assuming that covers this category’s purchase opportunity. If recruitment continues for 14 days, approximately 28 days are needed until the last assigned customer’s observation ends; finalization of cancellations and refunds and data-processing delays are additional. |
| Immediate stopping conditions | Stop follow-up sends until the cause is understood and controlled if messages reach incorrect customers, withdrawal requests are not reflected, or order/customer identity errors are confirmed. |
| Conditions for reducing volume | Under this team’s internal rule, temporarily stop new entries if at least 10 inquiries remain unresolved and there is insufficient capacity to handle them by the next business day. |
Set the observation period using the time customers need to have a chance to act, the time required to recruit the audience, and delays for cancellations, refunds, and events. Ending every campaign on the same date may leave late entrants too little time. Avoid extending the period arbitrarily until results look favorable.
Do not declare success merely because a small difference in orders appears. With a small sample or substantial uncertainty, the conclusion can remain “operationally feasible, but impact is not yet established.” Conversely, incorrect eligibility or excessive support workload can justify declining to expand even if some purchases occur. Customer.io’s documentation also explains that some results do not establish a clear winner.
The first journey should enable the next decision
Once you choose a candidate, describe it in one sentence: “This message can help these customers complete this currently blocked action, and our team can assess both the result and the side effects.” Any gap in that sentence indicates whether data improvement, customer verification, or operational preparation should come first.
If the audience is small and its status can be checked directly, existing tools and manual review may be enough to examine the first hypothesis. If customer identities and order data are scattered across systems, or responsibility for exclusions, support, and stopping is unclear, resolve those connections before increasing campaign count. The first journey only needs to be small and clear enough to inform the next one.
After choosing the first journey, use the OneSignal customer journey design workbook to specify entry, exclusion, exit, and verification conditions.
If you need help reviewing a first journey against your data and goals, consider IXC’s CRM consulting scope.
