Starting from the decision
Most campaigns begin with an occasion: a sporting event, a new game, a promotional period, a month with a gap in the calendar. The audience is then selected and the message written.
That sequence produces communications that exist because the calendar had a slot.
A better sequence begins with a decision the campaign is meant to change. What do we want a specific group of customers to do differently, why would they, and what would tell us whether they did?
Working this way, several campaigns that would otherwise have been sent are not, because the honest answer to the second question is that the customers would have done the same thing anyway. That is a saving in bonus cost, in contact frequency and in the base's tolerance for being contacted.
Constructing a campaign
The components, in order.
Objective. A specific behavioural change in a specific group, stated in a form that could be measured.
Audience. The rules selecting who qualifies, built from the segmentation described in the previous lessons. This should be narrow enough that the message is relevant to everyone receiving it.
Suppression. Applied after selection and before anything sends: self-excluded, restricted, risk-flagged, opted out, market-restricted, frequency-capped. This must be a system control rather than a checklist step, because a manual process will eventually be skipped.
Holdout. A randomly selected proportion of the qualifying audience that receives nothing, without which the campaign cannot be measured.
Offer, where there is one, constructed and costed as the next lesson describes.
Message, in the channel appropriate to it.
Timing, including both the send moment and any sequence.
Measurement plan, defined before the send rather than assembled afterwards, specifying what will be compared and over what period.
The discipline that matters most is that the measurement plan and the holdout are decided in advance. A campaign measured retrospectively will be measured by whatever numbers look supportive.
Channel selection
Each channel has a character, and matching the message to it is most of the work.
Email carries detail, is inexpensive, and has low urgency. It suits offer explanations, statements, product introductions and anything a customer may want to return to. Its constraint is deliverability: a sender reputation damaged by complaints and poor list hygiene degrades every subsequent send, and recovering it is slow.
Push notification is immediate, brief and intrusive. It suits genuine time-sensitivity, such as an event a customer follows beginning shortly. It is the channel most easily overused, and the response to overuse is permanent disabling rather than reduced attention, which removes the channel entirely.
SMS is expensive, has very high open rates and carries an implication of importance. Using it for routine promotion trains customers to ignore a channel they would otherwise treat as significant.
On-site and in-app messaging reaches customers already engaged, which suits contextual relevance and does not suit interruption.
Outbound telephone contact is reserved for high-value account management and for safer gambling interaction, and it is the only channel where a customer's tone conveys information.
The principle is that the channel should follow the message. Operators that select by cost send promotional SMS and detailed offer terms by push, both of which perform badly for entirely predictable reasons.
Messaging under advertising codes
Gambling marketing is governed by the requirements set out in the Law and Compliance course, and writing within them without producing unreadable copy is a genuine skill.
The constraints that bear on CRM messaging.
No strong appeal to minors, assessed on likely appeal rather than intent, which affects imagery, personalities and style.
No suggestion that gambling solves problems or is a route out of financial difficulty.
No implication of guaranteed or reliable winning.
Accurate offer description with material terms presented clearly.
Responsible gambling messaging in the required form and prominence.
Age statements where mandated.
Market-specific requirements, which differ and which must be applied per recipient rather than per campaign.
The recurring practical difficulty is material terms. A wagering requirement, a game contribution rate and a maximum stake limit are material, and presenting them in a linked document is disclosure without communication. They belong in the message, at the point the customer decides.
This is resisted because it makes offers look less attractive, which is precisely the reason it is required. An offer whose appeal depends on its conditions being invisible is an offer with a problem, and the complaint volume it generates downstream, as the Customer Service course established, usually exceeds the response it gained.
The practical approach is to write the offer in plain terms first, including its conditions, and then assess whether it is still attractive. Offers that fail that test should be redesigned rather than obscured.
Timing
Timing affects response substantially and carries specific considerations here.
Send-time optimisation, delivering to each customer when they are most likely to engage, is effective and available in most platforms.
It also requires constraint, because the times an optimiser identifies as most effective are not neutral. Late-night contact reaches a population that skews towards customers warranting more care. Contact timed to follow a losing session reaches people at a moment protective practice says to leave alone. An optimiser rewarded for response will find both, because both work.
The constraints worth applying: quiet hours during which no promotional contact sends regardless of predicted response; no contact following a significant loss, which requires the system to know; and no contact during an active session, which interrupts rather than engages.
Beyond individual timing, sequence timing matters. A dormancy sequence that fires a week after the customer went quiet has missed the moment. A welcome sequence delivering its third message before the customer has returned for a second session is talking to itself.
Frequency
Covered in the previous lesson and worth restating as a campaign discipline.
A customer may qualify for several campaigns simultaneously. Without governance they receive all of them.
Caps limit total contact per period. Priority rules determine which sends when several qualify. Channel caps apply separately, since tolerance differs by channel. Recovery periods prevent clustering.
The evidence that frequency is wrong is usually available and rarely examined: rising opt-out rates, falling open rates on a stable base, and complaint rates trending upward. Each indicates that the base is being contacted more than it wants, and each precedes the point at which the channel stops working.
Operators that implement governance generally see volume fall and effectiveness rise, which is the signature of a base that was over-contacted.
Testing
Campaign testing in this sector faces the constraints described in the Operations Strategy and Product Innovation courses, and several things remain testable.
Subject lines, creative and copy, which are straightforward and cheap.
Channel, comparing the same message delivered differently.
Timing, within the constraints above.
Offer construction, where regulation permits differential terms, which it does not everywhere.
Sequence design, comparing programme variants.
Frequency, which is among the most valuable tests available and among the least run, because it requires deliberately contacting some customers less.
The statistical cautions from the Operations Strategy course apply with full force. Revenue skew means a result carried by a small number of customers is not a result. Test duration must be long enough for novelty to decay. And the outcome measured should be contribution over a meaningful period rather than immediate response.
What good messaging looks like
To close, the characteristics of CRM communication that works.
It is relevant, meaning the recipient can see why they received it.
It is honest, including about conditions, which builds the trust that makes future messages worth opening.
It is timely, connected to something happening rather than to a calendar slot.
It is restrained, arriving infrequently enough that it is noticed.
It is useful even without the offer, since messages whose entire value is a bonus train customers to ignore anything without one.
And it is appropriate to the recipient's situation, which requires knowing what that situation is and having the discipline not to send when the answer is that they should be left alone.
That last characteristic is the one that distinguishes CRM in this sector from CRM generally, and it is the subject the next lessons develop.
Deliverability
An operational discipline that determines whether any of the above reaches anyone, and one that CRM functions frequently discover only when it fails.
Sender reputation governs whether email reaches inboxes or spam folders. It is built over time and damaged by complaint rates, by sending to invalid addresses, and by volume spikes that look like compromise. Once damaged it recovers slowly, and the recovery period affects every campaign.
List hygiene means removing addresses that bounce, that have never engaged, and that are likely to be inactive. Operators reluctant to remove non-engaging addresses, because a larger list feels better, damage deliverability for the customers who do engage.
Authentication through the standard sender verification protocols is a technical requirement and is occasionally missing at operators whose email is set up by whoever was available.
Complaint rates are the fastest route to reputation damage, and in this sector they are elevated because customers who receive gambling marketing they did not expect complain rather than unsubscribing.
Engagement segmentation sends more frequently to engaged recipients and less to unengaged ones, which improves deliverability for everyone.
Push and SMS have equivalents. Push permissions once revoked are rarely restored. SMS shortcode reputation and carrier filtering behave similarly.
The general point is that channel access is an asset that can be spent. An operator that sends aggressively to a large unengaged list will find, over a year or two, that its messages stop arriving, and the recovery costs more than the volume gained.
Campaign operations
A brief note on execution, since the failures here are mundane and expensive.
Audience selection should be reproducible, meaning the rules are documented and would produce the same list if run again. Ad hoc selections assembled by whoever built the campaign cannot be audited or repeated.
Suppressions must be system-enforced rather than applied as a manual step, since any manual step in a repeated process will eventually be missed.
Approval should include compliance for anything novel, and should not be required for material assembled from pre-approved components, which is how a compliance function stays fast enough to be used.
Test sends should go to real devices, in the market's language, rendered as customers will see them.
Records should retain what was sent, to whom, when and with what terms, because defending a historical campaign requires knowing what it actually said.
Post-send monitoring should catch delivery failures, unexpected complaint spikes and errors quickly, since a campaign with a mistake in it is considerably cheaper to address in the first hour.
Personalisation in messaging
The extension of the segmentation work into content, and one where the effort frequently exceeds the return.
Basic personalisation, using a customer's name and referencing their known preferences, is cheap and modestly effective.
Content personalisation, showing different games, sports or offers based on behaviour, is more valuable and requires the content to actually differ rather than the same message with a different image.
Dynamic assembly, where a message is constructed from components selected per recipient, allows genuine relevance at scale and requires infrastructure many operators lack.
Triggered relevance, where the message responds to something the customer did, generally outperforms all of the above, because timing and context carry more weight than content variation.
The caution is that personalisation can be conspicuous in a way that is unwelcome. A message demonstrating detailed knowledge of a customer's behaviour reads as surveillance rather than as service, and the line is easily crossed. Referencing a game someone played is helpful; referencing how much they lost on it is not.
The specific consideration in this sector is that personalisation drawing on risk-relevant behaviour must not be used to increase engagement, as the Product Innovation course argued. A system that has learned which customers respond to which prompts at which moments has learned something that should be constrained rather than exploited, and the constraint has to be designed in rather than assumed.
The practical guidance is to personalise on preference rather than on intensity, to keep the reference light, and to check what the system is actually using to make its selections.
The always-on shift
A structural change in how CRM operates, worth naming because it affects everything in this lesson.
The traditional model is a campaign calendar: a schedule of sends planned in advance, each with an audience and an occasion. It is manageable, it is visible, and it means most communications exist because a slot needed filling.
The alternative is an always-on model built from triggers. Communications fire when something happens to a customer rather than when the calendar reaches a date, and the calendar is reserved for genuinely time-bound events.
The advantages are substantial. Relevance improves, because the message relates to something that just occurred. Volume falls, because customers only receive things when there is a reason. Measurement improves, because each trigger is a discrete programme with its own holdout rather than a one-off send. And the function's work shifts from producing campaigns to designing and refining programmes, which compounds rather than resetting each month.
The requirements are a trigger infrastructure capable of acting on behavioural events promptly, a governance layer preventing accumulation, and a discipline about what genuinely warrants a communication.
The transition is difficult for a specific organisational reason. A campaign calendar produces visible output that looks like work. A well-designed always-on programme produces fewer sends and better results, and a function measured on campaign volume will resist moving to it.
That is a measurement problem rather than a strategic one, and it is addressed in the final lesson of this course.
Common campaign failures
A short catalogue, drawn from the sections above.
Campaign exists because the calendar had a slot, with no decision it was meant to change.
Audience too broad, so the message is relevant to a minority of recipients and irritating to the rest.
No holdout, so the result cannot be interpreted.
Material terms in a linked document, producing the dispute volume described in the Customer Service course.
Channel selected by cost, producing promotional SMS and detailed terms by push notification.
Send-time optimisation unconstrained, delivering at hours and moments the operator should avoid.
No frequency governance, so customers qualifying for several campaigns receive all of them.
Suppression applied manually, which will eventually be missed.
Success measured on open and click rates, which describe engagement with the message rather than any change in behaviour.
Deliverability neglected until it fails, at which point recovery is slow.
Each of these is straightforward to correct and most persist because nobody has looked at the programme as a whole rather than campaign by campaign. An annual review of the full communication set a customer receives across a year, read as that customer would experience it, surfaces most of them in an afternoon.
