Visit Website →
← Back to blog

App Clips and Research: Product Recommendation Pushes for E-commerce

October 1, 2026
App Clips and Research: Product Recommendation Pushes for E-commerce

A product recommendation push is a notification that suggests a specific item to a shopper based on their behavior, purchase history, or browsing pattern, and it works best for three jobs: recovering an abandoned cart, cross-selling after a purchase, and pulling lapsed shoppers back to the store. Done well, it lifts open rates and conversion without wrecking your permission rate. Start by tracking open rate, click-through rate, and disablement together, not opens alone.

TL;DR:

  • Complementary recommendations, anchored to recent purchases, outperform personalized single-item suggestions in open rates because they leverage contextual product pairings.
  • Recommendation pushes should match their trigger with the appropriate landing page, such as cart reminders landing on carts and browse abandoned items on their specific product pages.
  • Using dynamic timing systems like STEPS can reduce push permission disablement and increase active days by evaluating signals instead of fixed schedules.
  • Always verify product availability, loading integrity, and matching landing pages before sending, as irrelevant or out-of-stock recommendations rapidly erode trust.
  • StorePush bypasses permission barriers by delivering notifications directly to lock screens via iOS App Clips, reaching users without requiring email or app installs.
StorePush
storepush.io
Re-engage More Shoppers After They Leave
StorePush sends lock screen notifications through iOS App Clips, helping retailers reconnect with visitors without collecting emails or phone numbers.
Explore StorePush

Table of Contents

Recommendation models and how to pick what to recommend

Not every recommendation push should say the same thing. Three models cover most e-commerce use cases, and picking the wrong one for your data situation is the fastest way to send something irrelevant.

Complementary recommendations suggest an item that pairs with something the shopper already bought or has in their cart, like a phone case after a phone purchase. Personalized single-item recommendations predict the next item a specific shopper is likely to want based on their own browsing or purchase history, independent of any recent transaction. Popularity fallbacks show a best-seller or trending item when you have no usable signal for that shopper at all.

Here's what's interesting: complementary recommendations tend to outperform single-item personalized pushes in open-rate experiments, and the reason comes down to context. When a shopper just bought running shoes, a push suggesting moisture-wicking socks has two anchors instead of one: the category they already care about, and the specific product they already own. Research on complementary product recommendations found that this two-anchor effect produced statistically meaningful open-rate lifts over single-item personalization, because the purchased item itself acts as a relevance signal the shopper immediately recognizes.

The catch is that push notifications rarely have room to show more than one or two products. That constraint changes the design problem: instead of building a wide recommendation carousel like you would on a product page, you're ranking to find the single best complementary item or best personalized pick, and getting that ranking wrong is costly because there's no second option competing for attention.

Match the model to what data you actually have:

  • Rich purchase history: Use complementary recommendations anchored to the most recent order.
  • Browsing history but no purchase: Use personalized single-item recommendations based on viewed or wishlisted products.
  • First-time visitor or no tracked behavior: Use a popularity fallback, ideally scoped to the category they landed on.
  • Returning shopper with a stale cart: Anchor to the cart item itself rather than a broader personalization model.

A quick decision checklist before you build a campaign: do you know what they bought or viewed, is that data recent enough to be relevant, and does your catalog have a genuine complementary pair or does it lean toward standalone products? If the answer to any of these is no, drop down a tier rather than force a weak personalization guess. Our guide to push personalization walks through how this plays out across different store types.

Page context, triggers, and use cases: where to send recommendation pushes

The trigger determines both the message and the landing destination, and mismatching the two is one of the most common ways stores waste a send.

Four triggers cover most of the recommendation push calendar:

  • Cart abandonment: Fires when a shopper adds items and leaves without checking out, and should land directly on the cart or a pre-filled checkout page.
  • Browse abandonment: Fires after meaningful product views with no add-to-cart, and should land on the product detail page (PDP) they viewed, not a generic collection.
  • Post-purchase cross-sell: Fires shortly after an order confirms, and should land on a curated collection or the complementary product's own PDP.
  • Restock or price-drop: Fires when a previously viewed or wishlisted item becomes available or drops in price, and should land on that exact product.

The mismatch to avoid: sending a cart abandonment push that lands on your homepage instead of the cart. The shopper has to re-find what they already chose, and that extra friction kills the conversion you were trying to recover.

Device and channel also shape what you can actually send. Web push through the browser typically allows a short title, a brief body line, and one small image, often capped around 40 to 50 characters for the title before truncation on some browsers. Native mobile push through iOS or Android gives a bit more room and richer image support, but both channels punish long copy: if your message gets cut off mid-sentence, the recommendation loses its point before the shopper even taps it. Image assets should be simple and legible at thumbnail size since most operating systems compress or crop aggressively, and a blurry or cropped product shot undercuts the whole pitch. StorePush uses native iOS App Clips to deliver these pushes directly to a shopper's lock screen without requiring an app install, which sidesteps some of the browser-based image and character constraints that web push deals with.

Timing, cadence, and the whether/when decision for pushes

Most stores still run recommendation pushes on a fixed schedule: send one hour after cart abandonment, one at 24 hours, maybe one more at 72 hours. This works, but it treats every shopper the same regardless of whether they're actually likely to respond at that moment, and it's part of why permission-disablement rates creep up over time.

Newer approaches treat timing as a decision made jointly with the send decision itself, rather than a fixed interval applied to everyone. A self-triggered agentic system called STEPS decides both whether to send a push right now and when to reevaluate next, instead of firing on a rigid clock. In online A/B tests, this joint approach increased user active days slightly and cut push-permission disablement noticeably, which matters because a shopper who disables push is gone for every future campaign, not just this one. The system also used a lightweight filtering agent to cut the computational overhead of constantly re-evaluating timing, which is worth knowing if you're weighing build complexity against a fixed-schedule approach.

Timing, cadence, and the whether/when decision for pushes — overview diagram

You don't need a full agentic system to apply the underlying lesson: decide whether to send at all before you decide when, and let real signals adjust the clock instead of a static countdown.

Practical cadence rules to start with:

  1. Test three windows for cart abandonment: one hour, 24 hours, and 72 hours, and measure conversion and disablement separately for each.
  2. Cap sends per shopper per week so a browse-abandonment push doesn't stack on top of a cart-abandonment push on the same day.
  3. Build a cooling window after any push is dismissed without a click, delaying the next send by at least a day.
  4. Suppress sends around a recent purchase for 48 to 72 hours unless the message is a deliberate cross-sell tied to that exact order.
  5. Review disablement rate weekly, not just conversion, since a cadence that lifts short-term sales but spikes opt-outs is a net loss. Our timing and time zone guide covers how to localize these windows across a distributed customer base.

Message composition: templates, copywriting, and creative constraints

A recommendation push has almost no room for error. You typically get a title, one line of body copy, a small image, and a call to action, and every one of those elements has to earn its place.

The basic anatomy: an anchor (what triggered this message, like the item they viewed or bought), a recommended item line (the specific product and why it fits), and a CTA that tells the shopper exactly what happens when they tap. The image should show the recommended product clearly, cropped tightly, since most push surfaces render it at thumbnail size.

Illustration of recommendation push message structure

Recent research framing message composition as its own optimization layer, separate from targeting and timing, suggests judging copy against five criteria: relevance (does this match what the shopper actually did), clarity (can they understand the offer in one glance), actionability (is the next step obvious), novelty (does it avoid sounding like every other push they've dismissed), and persuasive appropriateness (does the urgency match the actual stakes, rather than manufacturing false scarcity).

A checklist worth running before you schedule any send:

  • Relevance check: Does the recommended item connect directly to a real action this shopper took?
  • Clarity check: Can someone understand the offer without opening the app or site first?
  • Actionability check: Is there one clear CTA, not two competing ones?
  • Novelty check: Does the copy avoid the exact phrasing used in the last three pushes to this shopper?
  • Tone check: Is the urgency proportional, so "still available" doesn't get inflated into "selling out fast" without evidence?

Pro Tip: Write the title as if the shopper only reads that one line and never sees the body copy, because on many devices, that's exactly what happens.

Three templates to start from:

Behavioral (browse abandonment): "Still thinking about the [Product Name]? It's waiting in your size." Landing page: the exact PDP viewed.

Complement (post-purchase): "Your [Product Name] pairs well with [Complementary Item]. Complete the set." Landing page: the complementary item's PDP or a two-item bundle page.

Best-seller fallback (no history): "Shoppers are loving [Best-Seller Name] right now." Landing page: the category collection featuring that item.

A/B test ideas worth running early: title length (short punchy versus slightly descriptive), CTA phrasing ("Shop now" versus "Complete your order"), and whether including a price in the copy increases or suppresses clicks. Our copywriting examples page has more variations by trigger type.

Fallback strategies and cold-user patterns

A recommendation push with no relevant data to work from is worse than no push at all, because an irrelevant suggestion reads as spam and often triggers a disablement. Building a fallback hierarchy before launch avoids that trap.

The safest order runs from specific to general:

  • User-specific: Recent purchase or browsing history, when it exists and is recent enough to trust.
  • Category-level: If you know what category they browsed but not a specific item, recommend a top performer in that category.
  • Most-popular: A general best-seller when you have no category signal at all.
  • Editorial pick: A manually chosen item, useful for new stores with too little data to generate any of the above reliably.

Vendor documentation on web-push recommendation features describes exactly this pattern, recommending user-specific suggestions when data exists and falling back to controlled popular-item suggestions when it doesn't, rather than leaving the field blank or guessing.

Before any send goes out, run three checks: confirm the product image actually loads and isn't broken, confirm the item is in stock (a recommendation for a sold-out product is an instant trust hit), and confirm the landing page works and matches what the notification promised. A controlled exploration strategy, occasionally testing a lesser-known item against your top fallback, can surface hidden winners in your catalog, but keep that exploration to a small percentage of sends so it doesn't drag down your baseline conversion.

Measurement and KPIs: what to track and how to evaluate lift

Open rate alone tells you almost nothing about whether a recommendation push is working, because a shopper can open a notification and bounce immediately if the recommendation missed. Track a fuller set from day one.

Primary performance metrics:

  • Open rate: How many recipients tap the notification, a useful top-of-funnel signal but not a success metric on its own.
  • Click-through rate to the landing page: Confirms the tap led somewhere relevant, not just a dismissal or accidental open.
  • Conversion rate: The share of pushes that end in a completed purchase, the metric that actually matters for revenue.
  • Attributed revenue: Total sales tied directly to the push, ideally tracked through a dashboard that separates it from unrelated organic traffic.
  • Average order value: Whether the recommended item, especially in complementary sends, increases basket size beyond the original item.

Just as important, and often ignored, is permission health. Track disablement rate (the share of recipients who turn off push after a given send), opt-out volume over time, and active days per user, since a campaign that boosts short-term conversion while quietly raising disablement is trading future revenue for a one-time bump. This is exactly the trade-off the STEPS research measured directly, watching active days and disablement rate alongside conversion rather than treating opens as the only signal that mattered.

For A/B tests, run at least two variants across a large enough cohort to detect a meaningful difference, use a consistent attribution window (24 to 72 hours after send is common for e-commerce), and always report disablement rate alongside conversion in the results, not as a footnote. A test that wins on opens but loses on disablement is not a win.

Recommendation pushes often rely on behavioral data, like browsing history or cart contents, and how you collect and reuse that data carries real compliance weight, not just a best-practice suggestion.

FTC guidance on misuse of information collected in confidential contexts makes clear that reusing information gathered in a confidential context for advertising purposes may require affirmative express consent, and that any disclosures tied to that use must be clear and conspicuous, not buried in fine print. Related FTC staff guidance on digital disclosures recommends placing disclosures near the claim itself, and where space is limited, as it often is in a push notification, making sure full details are available on the click-through landing page.

A practical checklist before launch:

  • Confirm consent basis: Know whether your push permission covers behavioral targeting or only general marketing, and separate the two if needed.
  • Place disclosures where they're seen: At opt-in, not buried in a privacy policy the shopper never opens.
  • Require separate opt-in for sensitive inferences: If a recommendation could reveal something sensitive about the shopper, treat that as a separate consent decision.
  • Log consent events: Keep a record of when and how permission was granted, tied to the specific use case.
  • Review targeting choices against disclosure scope: If your disclosure only mentions "order updates," using the same channel for product recommendations may need updated language.

Run this checklist past legal review before scaling any new targeting approach, since a strong conversion result doesn't offset a compliance gap. Our CCPA consent steps guide covers related state-level requirements worth checking alongside FTC guidance.

StorePush in practice: evidence and marketer playbook

Behavioral recommendation pushes, ones anchored to a specific cart or browsing action rather than a generic reminder, have been reported to produce higher open rates than generic re-engagement notifications, according to internal campaign data from some merchants. These are delivered through native iOS App Clips, allowing shoppers to receive and act on lock-screen notifications without installing an app or providing an email address, a setup that has shown improved opt-in rates compared to email-gated re-engagement flows according to available tracking.

A mini implementation checklist for a first rollout:

  • Connect your platform: The platform integrates with popular e-commerce platforms like Shopify, WooCommerce, BigCommerce, and custom storefronts.
  • Pick one trigger to start: Cart abandonment is the highest-volume starting point for most stores.
  • Set up your first A/B test: Compare a generic reminder against a behavioral, complement-anchored message.
  • Watch the dashboard for revenue attribution and CTR heatmaps to see which product placements and copy variants are actually converting.

Early experiments worth running: test the one-hour versus 24-hour send window against your own cart-abandonment baseline, and track disablement rate from week one rather than adding it later once a pattern has already set in. Details on the behavioral-push data referenced above are in our templates and results breakdown.

Balancing short-term conversion with long-term permission health

The mistake I see most often is optimizing a recommendation push program purely for this week's conversion number. Permission is a finite resource. Once a shopper turns off push, every future campaign, not just this one, loses that channel permanently.

Three rules of thumb worth adopting early: treat disablement rate as a KPI with the same weight as conversion, not a footnote you check once a quarter. Never let a fallback recommendation go out without a live inventory check, because a sold-out suggestion erodes trust faster than a slightly generic one. And build your cadence around signals, not a fixed clock, because the shoppers most likely to convert and the shoppers most likely to opt out are rarely on the same schedule. A related perspective on retention economics is worth a look in this customer retention strategy piece, which makes a similar case for treating the channel itself as an asset worth protecting.

— Lucas

How StorePush solves the same job

If you've read this far, you've probably noticed that most of the friction in recommendation pushes traces back to one problem: you need a shopper's permission and contact details before you can reach them at all, and most abandon before ever giving either. StorePush sidesteps that specific bottleneck. It delivers recommendation pushes straight to a shopper's lock screen using native iOS App Clips, with no email, no phone number, and no app install required, which opens up a channel to the vast majority of visitors who leave without buying and would otherwise be unreachable by any of the trigger and timing strategies covered above.

A simple test plan for your first rollout: pick a small cohort, run a behavioral cart-abandonment push against your current recovery method, and measure conversion lift alongside permission health, not just opens. The attribution dashboard tracks funnel performance and CTR by placement so you can see what's working without guessing.

StorePush offers a Free plan to start, a Pro plan at $50 per month, and a 5% usage commission on attributed recovered revenue. If you want a walkthrough first, you can book a demo to see how it fits your storefront.

Sources

FAQ

Can you give me an example of a product recommendation push?

A common example is a post-purchase cross-sell: after a shopper buys running shoes, a push might read, "Your new shoes pair well with moisture-wicking socks. Complete the set," linking directly to that product. This kind of complementary recommendation tends to outperform generic reminders because it anchors to something the shopper just bought, as shown in research on complementary product recommendations.

What are some good examples of push notifications for e-commerce?

Strong examples include cart-abandonment reminders that link directly to the cart, browse-abandonment pushes tied to the exact product viewed, restock alerts for wishlisted items, and post-purchase cross-sells suggesting a complementary product. The strongest ones share a trigger, a specific product, and a landing page that matches the message exactly.

Should I allow push notifications from stores I shop at?

That depends on how much you value quick alerts about restocks, price drops, or items you've already shown interest in versus the inconvenience of extra notifications. Stores are required under FTC guidance to seek affirmative consent and disclose clearly how your data will be used, so checking those disclosures before opting in is worth the minute it takes.

How do I turn off unwanted product recommendation push notifications?

You can typically disable push notifications from a specific store or app through your device's notification settings, or through an unsubscribe option in the notification itself if one is provided. If a store makes this difficult or doesn't disclose how to opt out, that itself may fall short of the clear-disclosure standard FTC guidance sets for marketers.