In-App Marketing Strategies: The Canon Nobody Told You About

I believe in-app marketing strategies have a writing problem. Search the phrase, and you get a wall of vendor blog posts recycling the same five product engagement examples under different headlines.

In my experience, the people who think the most about messaging users inside a product rarely call themselves in-app marketers. They may call themselves growth practitioners, behavioral designers, mobile economists, or experimentation engineers.

But that’s the thing. Growth loops, behavioral psychology, experimental science, and lifecycle CRM each provide a different part of the answer, and none of them alone is sufficient. The 2026 in-app marketer’s job is to handle all four simultaneously while external pressures keep rewriting the rules.

In this piece, I unify the thinking that serious practitioners in the field have published under other names. I’ve organized it by the questions I get asked most, answered with the actual frameworks rather than the vendor blog versions. The canon exists, but it has never lived on a page titled “in-app marketing strategies.”

What even is in-app marketing?

I spent the first few years of my career thinking in-app marketing was a category of tools: tooltips, modals, checklists. It’s not. It’s the operational layer where four older disciplines collide.

The in-app messaging practitioner who studies only one of those traditions will hit a ceiling that looks like a tactics problem but is actually a strategy gap.

Brian Balfour, Casey Winters, Kevin Kwok, and Andrew Chen provide the growth-loops framework; Nir Eyal and B.J. Fogg provide the behavioral psychology framework. Ron Kohavi and the experimentation culture at Booking.com provide the rigorous testing/experimentation science framework. Andy Carvell, Eric Seufert, and Phiture supply the lifecycle CRM and mobile economics framework.

With this in mind, my working definition of in-app marketing is this:

The practice of designing, targeting, and measuring behavioral interventions delivered inside a live product, sitting at the intersection of growth loop engineering, behavioral psychology, experimentation science, and lifecycle CRM.

An in-app message earns the “marketing” label when it is grounded in a defined loop, triggered by a behavioral signal, measured against a hypothesis with a clean holdout group, and calibrated to the user’s lifecycle position. Anything that falls short on one of those four criteria might still be worth doing, but calling it in-app marketing will set the wrong expectations for what it can deliver.

For example, a growth team that lives in loops but has no experimentation discipline ships messages whose lift numbers are mostly noise. A CRM team with perfect lifecycle logic but no loop thinking optimizes the lubricant while the engine stalls.

The implication of this definition on staffing is real. The best in-app marketers I’ve worked with have a primary tradition and working knowledge of the other three disciplines. The ones who struggle are usually deep experts in one field who treat the others as implementation details.

How should I think about in-app messages strategically?

This is the question I wish someone had answered for me before I spent years running campaigns that moved metrics I didn’t fully understand. The strategic framing matters because in-app messages are not inherently good or bad; they’re load-bearing or decorative, depending on the underlying loop.

What follows are the questions and arguments that have shaped my understanding of in-app marketing strategies today.

What are they actually for? Loops vs. funnels

If you’ve been in marketing long enough, you’ve likely come across Dave McClure’s 2007 AARRR framework. This framework describes a linear journey from acquisition to revenue. For many marketers like me, it was the most important framework for identifying which user should receive which message.

Not anymore.

The most useful reframing for in-app marketing in the last decade came from Brian Balfour, Casey Winters, Kevin Kwok, and Andrew Chen in “Growth Loops are the New Funnels.”

They describe loops as self-reinforcing cycles in which the output of one user’s action becomes the input that brings in or retains the next.

The way I see it is this: If you believe in funnels, every in-app message is a nudge down a stage. If, however, you believe in loops, every in-app message is a load-bearing trigger inside an engine, a piece of behavior the product depends on to keep growing.

Elena Verna’s Racecar model sharpens this further: loops are the engine, messages are the lubricant, one-off tactics are the turbo boost, and funnels are just the fuel. In-app messages are lubricants. If the loop underneath isn’t real, no amount of clever messaging will make a product grow.

The practical question, then, is: how do you know whether your loop is real before you start building messages on top of it?

For me, the answer has always been in the behavioral data: which actions predict retention, which features the users who renew actually use, and which moments reliably precede drop-off. That’s the prerequisite work that makes every subsequent message decision defensible, and it’s why I won’t write a message brief before I’ve spent time in the analytics.

Andrew Chen’s Power User Curve takes it even further. His approach is to plot users by the number of days active in the last 28 days. Power users get expansion and advocacy prompts, mid-curve users get habit-reinforcement flows, and new users get activation guidance, with lifecycle stage driving the logic rather than signup date.

Who should they target? The activation moment, found by regression, not gut.

Casey Winters’ framework for defining customer activation moments formalizes what most teams treat as intuition. An activation moment is the behavioral threshold beyond which a new user is statistically likely to retain. You find it by regressing early behaviors against long-term retention, not by guessing based on what feels like a meaningful first action.

Discovering what activation looks like for your product is one of the trickiest tasks you may ever undertake.

For example, you may have heard the aha moment stories of Facebook and Slack. According to Chamath Palihapitiya, the Facebook growth team noticed that new users who added 7 friends in 10 days were more likely to become engaged, long-term users. Likewise, Slack’s Stewart Butterfield observed that teams that sent up to 2,000 messages were more likely to keep using the product.

But none of those numbers were plucked from thin air. Both products had their teams study their users and define their activation threshold using the data. The same applies to you.

Invest the time to know your real activation and long-term retention threshold. Armed with this data, you can create targeted messages that fire at different points in the user journey and drive users toward activation and retention.

The tooltip-vs-broken-product debate

Intuit’s content design guide states the purist position plainly: “Tooltips shouldn’t be essential to completing a task. If a tooltip is needed to work around a broken experience, it isn’t the best solution.”

Des Traynor and the Intercom design tradition repeat this argument across their published writing on product design, and at the consumer-product level, they are broadly right.

The vendor-tooling framing, however, runs in the opposite direction. The in-app messaging layer is a permeable interface over an unchanged product, deployable without engineering work, which is precisely what makes it attractive to growth and CS teams who can’t get priority in the engineering queue.

If you ask me, the design-purist argument is correct for consumer products with low surface area. But it’s wrong for multi-tenant B2B SaaS with role-based permissions, complex integrations, and admin surfaces, where some explanation is structurally unavoidable and engineering bandwidth to fix every UX rough edge does not exist.

Even then, the answer is rarely straightforward.

For example, in Casey Winters’ Pinterest case study, he found that the team hid group boards and messaging from new users until they had activated. His conclusion was simple: If the user can’t yet handle a feature’s complexity, the right response is to remove it from their path, not to write a tooltip explaining it.

To ensure I make the right decisions about whether to use tooltips or implement product fixes, my default is to watch our users’ session replays. If a tooltip is doing the work the product should be doing, it’s a band-aid. If it’s surfacing a genuinely non-obvious workflow to the right user at the right moment, it’s useful.

Does the same playbook work on mobile and B2B?

Cross-pollinating consumer mobile playbooks into B2B SaaS is the single most common tactical mistake I see. The structural differences are real enough to require entirely separate strategic frames; borrowing the wrong one produces misfires that look like execution problems but are actually category errors.

I’ll summarize these structural differences across six (6) dimensions:

Dimension	Consumer mobile	B2B SaaS
Decision maker	Individual user	Buying committee (admin, finance, champion)
Habit model	Variable rewards, streaks, social proof	Job-to-be-done completion; habits are a risk (churn from job loss)
Identity targeting	Post-ATT: probabilistic, cohort-based	Deterministic: known user ID, role, account
Push channel	Primary re-engagement channel; ~54% iOS opt-in	Supplementary: email and in-app dominate
Sales motion	Self-serve or app-store transactional	PLG/SLG hybrid; PQL routing to sales
Regulatory exposure	App store policies + platform privacy rules	FTC dark-pattern enforcement + enterprise procurement compliance

The clear differences between the two sides have led to a major difference in the use of in-app messaging. On the Remerge podcast, Andy Carvell argues that in-app messaging on mobile is “still in its infancy in terms of the amount of impact it can drive.” That framing is a useful manifesto for consumer mobile teams whose default channel is push.

Now, there’s one transferable principle across both contexts: loop thinking. Others, however, are non-transferable: habit-formation mechanics (fit consumer, misfit B2B) and post-ATT probabilistic targeting (fit mobile, breaks in identity-resolved B2B email and in-app programs).

What’s making in-app messaging harder than it used to be?

The honest answer is that there are more issues than I expected when I started in this field.

Push permission is structurally constrained, the FTC is actively enforcing dark-pattern rules, AI agents are being oversold, and the PLG/SLG division that everyone referenced in 2022 has already dissolved into a hybrid motion most teams aren’t fully designed for. I keep a short list of external forces I review before any strategy update, and these are the five that changed the most coming into 2026.

Are push notifications dead?

I’ll say push notifications are dying for one structural reason: Apple and Google decided that user attention belongs to them, not to the apps on their platforms.

iOS 13 introduced explicit opt-in permission in 2020; Android 13 followed in 2022; and iOS 26, released in September 2025, added Apple Intelligence priority ranking, which actively demotes generic notifications. The platforms built the channel, own the distribution layer, and are now enforcing that ownership in software.

The second cause is simpler: apps collectively flooded the channel until users started tuning it out. Airship’s 2026 benchmark covers 681 billion notifications sent to three billion users across 15 verticals, which works out to roughly 200 notifications per user per year before you count messaging apps.

At that volume, denying permission is a rational user response, not an anomaly. Even then, users who opt in to receive your push notifications are often choked by notifications from several other products. So, the space to win here is limited.

That said, iOS opt-in still sits at approximately 54%, and Android sits at around 97% after initially crashing to 67%. While these numbers may vary across product verticals, they show that push notifications aren’t dead yet, but they’re not as easy as they used to be either.

If you decide to pursue this channel, be sure to benchmark your product by Airship’s vertical numbers, not its platform numbers.

Are my growth tactics about to become illegal? The $2.5B Amazon ruling

Dark-pattern enforcement stopped being purely theoretical or ethical in September 2025 when the FTC’s action against Amazon ended in a $2.5 billion settlement, the largest FTC rule-violation penalty ever recorded, over the Prime sign-up flow and the internally named “Iliad Flow” cancellation maze. The FTC complaint is a case study every product and growth team should read before their next conversion optimization sprint.

Thanks to this ruling, some practices previously viewed as viable conversion rate tactics are now legally considered deceptive by the FTC. Harry Brignull’s taxonomy at deceptive.design identifies some of these: fake countdown timers on upgrade prompts, friction modals on the downgrade path, pre-checked add-ons, and close buttons sized to discourage clicking.

Of course, this doesn’t mean that every growth tactic is now illegal. Instead, it means you must be a lot more conscious of your choices and approach to growth.

If you’re like me and you wish to stay on the legal and ethical side of things, I recommend Nir Eyal’s Regret Test. Published in Behavioral Scientist, it involves asking one question: “If users knew everything I know about the design decision I just made, would they regret taking that action?”

If the honest answer is yes, the design is an ethical disaster and a regulatory liability, not a feature.

Why in-app messaging is actually more valuable now, not less

The introduction of Apple’s App Tracking Transparency (ATT) feature has made external user acquisition a lot more expensive.

As Eric Seufert argues, though, the loss of deterministic attribution post-ATT has actually made everything inside the product boundary more valuable. I agree.

You see, combined with the explicit opt-in permissions, everything outside the app is now completely or partially out of reach. Your app, however, is a channel within your control where you also have the lowest CAC. And it is more valuable than ever.

It’s no surprise, then, that teams are investing more and more in in-app marketing using tools like paywalls, onboarding flows, and first-session experience. If you’re not yet doing so, you’re missing out on one of your best growth channels.

Is PLG dead now? The hybrid motion

If you were around the product marketing space between 2020 and 2023, you must have watched the rise of the puritan PLG model in real time. But things have changed plenty since then. PLG isn’t dead, but it’s hardly PLG versus Sales anymore.

Kyle Poyar at Growth Unhinged and Elena Verna’s “Five PLG tactics every sales-led company should do” both describe the same convergence: PLG to land, sales to expand, and a shared in-app surface that serves both motions simultaneously.

The implication for your in-app marketing strategy is that your surfaces now have to serve two distinct user segments. The self-serve individual trying to activate a free trial needs different guidance from the enterprise account executive who already understands the product and just needs to configure it. A modal that delights one may create friction for the other.

This means your messaging strategy must have a branch for “this user just hit the product-qualified threshold.” It also means that PQL routing, in-app meeting booking, and sales-assist nudges are first-class in-app primitives in 2026, not edge cases.

Read more on cross-functional product communication that serves both motions.

What about AI agents in onboarding? Gartner’s 40%-will-fail prediction

Every onboarding vendor in 2026 ships something labeled “agentic” or “AI-personalized.” Gartner predicts that over 40% of enterprise apps will embed AI agents by the end of 2026. Gartner also predicts over 40% of agentic AI projects will fail by 2027, citing cost overruns, security gaps, and underestimated change management burden.

Personally, I agree that LLMs are great at adaptive in-app copy. This makes them excellent for rephrasing a tooltip for a technical vs. non-technical persona, or summarizing a help document inline when a user needs it. However, I don’t see them as reliable enough for autonomous guidance on non-trivial workflows where errors have real downstream consequences.

Still, I understand why you may want to provide some form of in-product agentic support. At Userpilot, we decided to take a slightly different approach to the agentic question. Here’s how our CEO, Yazan Sehwail, describes it:

“We see Userpilot as becoming the infrastructure that powers your product usage data for that sort of system. As teams start deploying their own AI agents, those agents are gonna tap on our existing infrastructure that will be powering all of the usage and all the product data, and that’s extremely powerful.”

Essentially, if you choose to lean on agents to provide a more dynamic, hyper-personalized onboarding and support system, Userpilot can help you track what the agent is doing inside your product. Agents are great, but you may want to hold back from giving them unchecked control until they mature even further.

What do serious people actually disagree about?

Most debates in in-app marketing happen at the tactics level, which is where I used to spend most of my energy too.

I’ve since realized, though, that the two disagreements that actually matter are both strategic: whether habit-forming design is inherently manipulative, and whether anyone can honestly claim their in-app program drives incremental value rather than just measuring correlation. I should warn that I have strong positions on both.

Is habit-forming design manipulation? Hooked, Indistractable, and the Regret Test

Nir Eyal’s Hook Model (2014) is the canonical behavioral framework for consumer in-app marketing: trigger, action, variable reward, investment. It’s foundational for consumer products, structurally unfit for most B2B, and unavoidable as a reference because of how widely it has shaped how product teams think about engagement over the past decade+.

Eyal’s 2019 follow-up, Indistractable, pivoted to teaching users to defend against those same mechanics. Tristan Harris and the Center for Humane Technology describe this as a structural cop-out. Placing the burden of attention defense on one individual against billion-dollar engineering teams is not a solution to the problem the first book helped create.

Eyal’s final refinement, the Regret Test, which I mentioned earlier, asks whether users would regret an action if they knew everything the designer knows. That’s a usable ethical screen in practice, even if it doesn’t resolve Harris’s structural critique about the asymmetry of the relationship between platform and user.

So, is habit-forming design manipulative?

Ethically, yes. More importantly, it’s unnecessary. B2B SaaS products simply do not need to rely on the Hook Model; they are utilitarian by design. Importing consumer habit-formation mechanics into a B2B tool invites both ethical exposure and structural mismatch.

The user who opens your analytics platform is trying to complete a job, not receive a dopamine reinforcement loop.

Can in-app marketing even be measured honestly?

Per-message, yes; program-level, mostly no.

Per-message measurement is tractable with the right setup: Ron Kohavi’s Trustworthy Online Controlled Experiments sets the standard with a clean split, a defined primary metric, and a measurement window longer than one week.

I put this to the test earlier in my career when two colleagues insisted that in-app messages didn’t move the needle. So, I ran an A/B test on our next feature launch, splitting users so half saw a single contextual tooltip at the first encounter and half saw nothing.

By the end of week four, feature adoption for the message group was 200% higher than the control group.

On a program level, though, measurement is a whole different problem.

Global holdout groups, where 5-10% of users receive zero in-app messaging for a quarter or more, are the only way to know whether your entire messaging program is adding value rather than just shifting timing. But there are three genuine reasons why most teams skip this:

• Message-to-message contamination means per-message A/B tests don’t capture cumulative fatigue effects
• Novelty effects mean 1-week wins often reverse by week four as users habituate
• Most revenue teams won’t agree to leave 5-10% of users untreated for months

Booking.com once tested this out by running over 1,000 concurrent tests. Their published results found that the vast majority were flat or negative.

Real wins have never been about hyper-optimizing button colors, but about strategy. And, while I’m certain of the effect of in-app marketing at the product level, it’s difficult to prove without a global holdout group.

How do I do this without bullshitting myself?

This is one question I wish someone had answered when I was building my first in-app flow. I’ve watched enough lift percentages get presented to leadership without a holdout group to know the field has a self-serving stats problem, and I’ve been guilty of it too early in my career.

Here’s how I now think about the numbers worth citing and the execution decisions that follow from them.

Which numbers can I actually trust?

Six numbers survive a methodology check and are worth putting in a deck or an internal brief.

• Push opt-in: ~54% iOS / ~97% Android median (Airship 2026, 681B notifications, 3B users, 15 verticals). Vertical variance is more important than platform variance for benchmarking your product.
• Activation rate: 30% median, 36% average, B2B enterprise 33%; median time-to-activation approximately 10 days (Lenny Rachitsky, 500+ SaaS products, independently surveyed). Revalidate your own activation threshold quarterly; the benchmark drifts as your user population changes.
• Trial-to-paid conversion: 2.5x higher with a defined activation event versus without one (same Rachitsky source). This is the single highest-ROI argument for doing activation work seriously.
• Feature adoption ceiling: 6.4% of features generate 80% of click volume; 15.6% for best-in-class (Pendo, 6,800 customers; measured as click-volume share, not unique-user usage, which matters for interpretation).
• Product tour abandonment: 67% before completion, 84% past eight steps: steepest at steps three and four (Gartner 2024 SaaS Adoption Report, directional rather than exact; use as design constraint, not a precision figure).
• Userpilot SaaS Product Metrics Benchmarks (547 customers, 7 verticals): the activation and engagement numbers triangulate well against Rachitsky’s independent survey, which increases confidence in both sources.

There are widely recycled numbers I don’t pay attention to. For instance, “70% of users churn after onboarding” and “90% churn from poor onboarding.”

My personal rule of thumb when deciding which numbers work best for benchmarking is this: if I can’t trace a number to a methodology, a sample size, and a measurement window in the same citation, I don’t use it.

The volume of vendor content that says “studies show” without a footnote is not a body of evidence. It’s opinion dressed as data, and it will get you embarrassed in front of anyone who actually reads the source.

How do I know if my messages actually worked?

If you’ve read this far, you’ll recognize my answer to this question: use a global holdout group, typically 5-10% of users who receive zero in-app messages for a defined window. That’s the gold standard.

Without this, you can describe what happened to users who received a message, but you cannot separate that from what would have happened to them anyway. When you’re ready, Blaze’s global control group documentation is the most operationally clear reference for setting it up.

However, there are two measurement failure modes to design around before running any tests.

The first is the novelty effect: a message that lifts goal completion in week one often reverses by week four as users habituate to it, which is why Kohavi explicitly warns against shipping on a 1-week result. The second is message-to-message contamination: when users in your A/B test for message X are simultaneously receiving messages Y and Z, the test measures the interaction of all three, not the effect of one.

The 200% adoption lift I described in the previous section held up because I ran it right: a clean control group, a four-week measurement window, and analysis of the full user funnel rather than just the immediate goal metric.

That’s not the industry norm for in-app A/B testing. But it should be, because the alternative is a growing pile of “lift percentages” that describe correlation at best and p-hacking at worst.

What should I actually do on Monday? The 10-point synthesis

Here are the ten decisions that follow from everything above, translated into next-week actions.

1. Define the growth loop this message feeds before building it. If you can’t name the loop, you can’t measure whether the message strengthened it.
2. Find your activation moment by regression, not gut. Brainstorm candidate behaviors, regress them against 90-day retention, and run experiments to lift the rate. Time-bound it at roughly 10 days per Rachitsky’s median, and revalidate quarterly as your product and user population evolve.
3. Before adding a message, try removing the complexity it explains. This may involve reworking the feature to make it simpler, or hiding it until the user is ready with progressive disclosure, as in Winters’ Pinterest example.
4. Segment by lifecycle position using behavioral signals, not signup date. The L7/L28 split tells you which band each user sits in. Userpilot’s segmentation engine runs this in real time from event data without a developer ticket.
5. Hold a global control group before making program-level claims. Without 5-10% of users seeing zero messages for a measurable window, every program-level lift number is attribution. Skip this if you only want to measure correlation.
6. Run cross-channel engagement as one program, not three teams. Email, in-app, push, and SMS either share suppression rules or they compete for the same user’s attention and frequency budget. Userpilot Workflows orchestrates this from one canvas rather than requiring you to reconcile three separate tools after the fact.
7. Treat regulatory and platform constraints as design inputs, not compliance footnotes. Dark-pattern enforcement, push permissions collapse, and Apple Intelligence priority ranking are real structural constraints on what channels can do. Design around them from the start.
8. Design for the hybrid PLG/SLG motion from the start. PQL routing, in-app meeting booking, and sales-assist nudges are first-class in-app marketing primitives in 2026. A pure self-serve in-app stack is underbuilt for most B2B products; building PQL branching into your message logic from the start avoids retrofitting it later.
9. Stay out of the Hook Model for utilitarian products. Variable rewards and habit-formation mechanics are a structural mismatch for B2B jobs-to-be-done. Always apply the Regret Test: What will the user do if they know what I know?
10. Be honest about AI’s current ceiling. LLMs are good at adaptive copy and unreliable for autonomous workflow guidance. If you must infuse AI, build the measurement layer first with Userpilot’s Agent Analytics. Understand what users actually need from your AI features before expanding into autonomous guidance, because the failed implementation stories from 2025 all started with overclaiming in 2024.

The synthesis

I don’t believe in-app marketing has a bad-practice problem. It has a missing-canon problem: the thinking that makes it work has been developed by Balfour on loops, by Winters on activation, by Kohavi on measurement, by Carvell on mobile, and by Seufert on post-ATT economics. It was just never assembled under the label most teams search for.

Userpilot sits at the intersection of all four traditions. Flows and spotlights serve the behavioral design layer; Product Analytics and event tracking serve the experimentation layer; Lifecycle segmentation and Workflows serve the CRM layer; and Lia plus Agent Analytics serve the growth loop’s emerging AI dimension. The platform you use for in-app marketing should cover all four, because any of the three you skip will keep breaking on you.

The vendor blog posts that prompted this article aren’t wrong for recommending tooltips. They’re wrong because they treat in-app marketing as a tactics conversation when the real work is strategic. Get the loop right, find the activation moment by regression, hold a measurement standard you’d defend publicly, and the tactics follow.

Book a demo to see how Userpilot operationalizes your in-app marketing strategy in a single platform.