The campaign is a useful management construct: it creates a deadline, a budget, a brief, and a moment when work goes live. However, it is a poor model of customer behavior.
Individuals do not become interested simply because a campaign launches on Monday. They enter and leave categories continuously. They compare options in fragments, moving across search, video, social feeds, physical stores, product pages, and peer conversations. Google’s research describes this dynamic as the “messy middle,” where individuals loop iteratively between exploration and evaluation prior to making a choice.1
Customers have already priced this behavior into their expectations: 83% expect to interact with someone immediately when they contact a company, and 85% expect consistent interactions across departments.2 Yet fewer than 10% of organizations have captured value from AI across end-to-end marketing workflows.3 The distance between those numbers represents the commercial opportunity.
A continuous engagement model starts from that reality. It treats every meaningful customer or market signal as an opportunity to make a better decision. Sometimes the correct decision is an outbound message. Sometimes it is a product recommendation, a service intervention, a sales handoff, or deliberate silence.
Customers do not wait for the campaign calendar
Campaigns are traditionally assembled from the inside out. The enterprise schedules a launch, a promotion, or a quarterly target. Teams then select an audience and distribute the message. Continuous engagement operates outside in. It begins with a change in the customer’s context, then asks what response delivers value in that exact moment.
The distinction is critical because relevance decays in hours, not weeks. Take one account at a mobile network operator, the single example this article carries from here on. A customer who compared two tariffs yesterday may need proof of coverage today. A customer whose line went down this morning does not need an upgrade offer. A customer already in a service conversation must not be treated like an anonymous site visitor.
Adobe’s research illustrates the persisting gap: 78% of customers want consistent brand experiences across digital and physical touchpoints, yet only 39% of organizations personalize the web experience while an individual is actively browsing, and only 31% update offers in real time.4 The calendar was built around the company’s operating rhythm, and the customer never agreed to it.
One account at a mobile network operator, April to June. The upper lane is what the brand sent, the lower lane is what the customer did.
A continuous engagement system is four jobs in order
Continuous engagement is four jobs that only work in sequence. The system has to notice the customer did something, remember what has already passed between the two of you, pick one response out of several that all look reasonable, and be allowed to pick none of them. Each job constrains the one after it, which is why a company that buys the third and never built the second simply sends more email.
Tuesday 09:14, the same account. One event, read four times.
- 01 / Notice A repeat visit, and the contract renews in 38 days. Passes on: a change worth deciding about.
- 02 / Remember Two of two emails already used this week. Last upgrade offer ignored. Passes on: one channel fewer.
- 03 / Choose Four candidate actions, ranked. The offer email is blocked, so the coverage proof wins. Passes on: one action, or none.
- 04 / Learn Ad served. Opened, ignored or complained about, the outcome is written back. Passes on: evidence for the next decision.
1. Notice that something changed
This is the signal layer. Signals originate from first-party behavior, billing and contract records, product and network usage, service conversations, consented CRM data, inventory shifts, competitor activity, and macro cultural trends. More data is not automatically better. The only useful signal is one that changes a business decision.
A disciplined system therefore begins with a strictly defined list of events that matter. For the operator in this article: opening the tariff comparison page, calling twice about the same bill, entering the last 60 days of a contract, going quiet in the first month after activation, or 5G going live at the registered address. Every event needs a reason to exist and an owner.
2. Remember what already happened
This is the memory layer. The system must retain records of what was previously displayed, what the customer completed, what they explicitly declined, which data permissions exist, and which channels remain appropriate. This is precisely where most enterprise personalization initiatives fail: 98% of marketing teams already using AI run into at least one data-related barrier, with siloed systems and poor data quality at the top of the list.5
Memory does not mean keeping everything forever. It means keeping the least context required to make a responsible next decision. Retention rules, consent tracking, access control, and suppression logic are product features, not legal paperwork bolted on at the end.
3. Choose one action, or none
This is the decision layer, and the part most systems skip. Standard marketing automation executes static rules well but cannot compare competing actions. One customer may qualify for several conflicting messages at once. The decision layer has to rank them by expected usefulness, brand priority, projected commercial value, how many messages the person has already had this week, and downside risk.
That calculation is the foundation of next-best-action marketing. The output is rarely as simple as “send email B.” It may be “show the coverage proof in paid social,” “route straight to service,” “wait until Thursday,” or “do nothing.” The right to send nothing is what separates continuous engagement from continuous interruption.
Tuesday 09:14. The customer has now opened the tariff comparison page three times in five days. Four things the system could do about it.
| Candidate action | May we? | Heard from us enough? | Is there a reason? | Decision |
|---|---|---|---|---|
| Email the upgrade offer | Yes, email consent | No room, 2 of 2 this week | Yes, three visits | BlockedThey have had their two emails this week. |
| Show the coverage proof in paid social | Yes, ads allowed | Costs no direct contact | Yes, three visits | SelectedReaches them without spending a message. |
| Hand to customer service | Yes | Room available | No, nothing is wrong | No reasonThe May fault was closed on 4 May. |
| Wait until Thursday | Nothing to ask. Waiting is not a contact. | HeldEligible again when the weekly limit resets. | ||
4. Do it, then learn from it
This is the delivery and learning layer. Execution should be channel-aware but not channel-owned. The identical decision context might shape a paid advertisement, a landing page, an email sequence, an in-app module, a retail prompt, or a customer service script. Every enacted response then feeds data back into the next decision cycle, including capturing negative evidence such as repeated ignores, cancellations, and filed complaints.
Adobe reports that 26% of organizations operate always-on customer journeys, and 20% maintain always-on retention journeys.4 The practical starting point is one journey with a defined outcome, a signal set small enough to manage, and enough volume to learn from.
03 / What a segment of one isA segment of one is a moment, not a profile
The phrase “segment of one” is frequently misunderstood. It does not require a permanent, perfectly accurate digital twin of every individual, and it does not mean building a campaign from scratch for every impression.
A segment of one is simply the combination of current context and available evidence captured at the exact moment of decision. The individual may still sit in several useful planning cohorts, but the action is chosen one decision at a time. The decision engine evaluates:
- Current intent signals and stage of exploration
- Recent cross-channel interactions and previous outcomes
- Tariffs, pricing, coverage at the registered address, and open service tickets
- Channel preferences, explicit permissions, and how many messages the customer has already had this week
- Brand rules, legal exclusions, and overarching commercial priorities
This changes creative strategy. Rather than building one finished advertisement per audience segment, teams develop a modular system of claims, proofs, offers, formats, and core brand components. The system automatically assembles the most relevant valid combination, executes it, and learns from the resulting performance.
The economics reward the discipline. Companies that grow faster drive 40 percent more of their revenue from personalization than their slower-growing peers.6 At the same time, 78% of marketers say they need more personalized content than they are able to produce.5
The same customer, recorded two ways, at 09:14 on the same Tuesday.
Written last year. Says the same thing today.
- Segment: price-conscious family
- Age band: 35–44
- Prefers: email
- Value tier: mid
- Motivation: reliable coverage
Still true next quarter. That is the appeal, and it is also the problem: nothing in this file knows they opened the tariff page an hour ago, or that they have already had two emails this week.
Built for one decision, then thrown away.
- Tariff page, 3 visits in 5 daysBehaviour
- Last upgrade offer ignoredSend log
- 5G live at their addressNetwork
- Email consent, no SMSConsent
- 2 of 2 emails used this weekDelivery log
Three of these five rows are false by Friday. That is the point, not a defect: every row names the feed it came from, so when one moves the next decision moves with it.
Personalization requires a clear value exchange
The best decision system in the world fails if the customer experiences it as surveillance. Customers have drawn the boundary clearly: 76% report frustration when interactions lack personalization,6 yet only 39% remain open to brands using AI to predict their emotional or mental state.7 Relevance is welcome, surveillance is not, and the line between them is drawn by the customer rather than by the brand.
Personalization has to make the experience easier, clearer, or more useful for the person on the other end. Where the only beneficiary is the advertiser, trust goes down.
That gives three design principles. Use consented data for a stated purpose. Show why a recommendation is relevant when the context calls for it. Give people real control over preferences, contact frequency, and opting out entirely.
Models need hard boundaries. Sensitive attributes, inferred vulnerabilities, regulated decisions, and high-impact claims either get stronger human review or stay out of the system. The team sets that policy before the system starts optimizing.
05 / How to startA practical first implementation
Select one journey where timing visibly dictates success. For the operator in this article that is contract renewal: activation, tariff research, and win-back work the same way, and all of them beat a broad brand program.
Define a highly focused signal list. Document exactly why each signal matters and specify what specific action it will influence.
Establish the decision rules. Rank the candidate actions, set a hard limit on how many messages one person can get in a week, and define exactly when a human must review the machine’s choice.
Measure both sides of the equation. Quantify customer value alongside commercial value, then feed both metrics directly into the next optimization cycle.
The objective is not to eliminate campaigns. Major launches and cultural moments retain their strategic value. The objective is to stop forcing every customer to wait for the next scheduled campaign before the brand can respond intelligently.
Footnotes and sources
- Google, Decoding Decisions: Marketing in the Messy Middle, replicated across more than 15 global markets.↩
- Salesforce, “State of the Connected Customer”, fifth edition.↩
- McKinsey & Company (2026), “Reinventing marketing workflows with agentic AI”.↩
- Adobe, 2025 AI and Digital Trends, based on 3,400 qualified business respondents and more than 8,000 consumers.↩
- Salesforce (2026), “State of Marketing, 10th Edition”, based on a survey of about 4,500 marketers.↩
- McKinsey & Company (2021), “Next in Personalization: The value of getting personalization right or wrong is multiplying”.↩
- Adobe (2026), “AI and Digital Trends”, consumer research.↩
