A lower open rate can be a good thing. Bogdan Mihalache of Email Kong feels Klaviyo's engagement playbook is stale: Engaged 30/60/90. Keep open rates high. Keep unsubscribe under 0.1%. But Bogdan’s point is that a lot of these benchmarks don’t match how people actually use email in 2026. Your campaign is one small blip in a very noisy inbox. So if you only optimize for tidy engagement metrics, you'll end up sending to fewer and fewer people = less money. That’s why a broader could send mean: → Open rate goes down → Click rate goes down → Unsubscribes go up → Absolute revenue goes up For a Shopify brand, that trade-off can be worth it. The goal of email is not to have perfect-looking dashboards. It’s to reach enough real potential buyers to drive incremental revenue.

Meta doesn’t really want you picking audiences anymore. Instead, it wants you to focus on the creatives for your ideal buyers. Then, it'll go find buyers for you. I heard a great example recently from a hydration brand who were selling to the gym crowd. But when they tested a different creative, Meta found a stronger demand from new mothers looking for a quick sip while breastfeeding. So your creative is your input. Meta can only learn from the purchase signals you send back. If the signal is missing, duplicated, or poorly matched, the algorithm learns the wrong thing. And even if your signal is accurate, you still need to ask: What are we teaching Meta to optimize for? A low-margin sample pack? A discounted one-off order? A returning customer? Or the high-margin hero product you actually want to scale? You can't have AI do that for you - you need to decide what kind of purchase matters. Choose creative angles → choose conversion signals → feed Meta clean enough data to find more of those buyers.

I think a lot of brands miss the most relevant context in retention flows. If a customer pauses their dog food subscription... You send a 20% off to win them back. But what if they now own a cat? That discount is irrelevant. So, the Recharge trigger tells Klaviyo the subscription was paused or cancelled. Fine. But do we know what else was the customer browsing? That’s where better Shopify tracking is critical. If someone pauses dog food but spends the next week browsing cat food… ...your retention flow should react to that signal. At Littledata, this is the context we help push into Klaviyo profiles and events: → Browse behavior → Product metadata → Shopify Markets data → Customer metafields → Product category context Sometimes you need to help buyers switch to a relevant subscription. Context >>> Timing

Last week, a DTC founder asked me: “Why do you think AI is actually relevant to my brand?” I realised I spend so much time around AI evangelists on LinkedIn that I’d forgotten a lot of founders are still asking this question. I told him this: AI is not going to run your ecommerce brand. It’s not going to replace taste. Or creativity. Or the judgment involved in deciding who your product is for and how you position it. But it will accelerate a lot of the work around those decisions. For example: 1. Getting answers from your business data without looking at dashboards 2. Handling support workflows and refunds 3. Helping Meta and Google find audiences for your creative If you think about the last one, the old model was: Pick the audience → Then design the creative. But now those platforms want you to just focus on the creative and your message. ...and AI will find the audience for you. It's quite a change in how performance marketing works. So, the basics are still important - like understanding customers well, knowing what makes your product unique... Then use AI to move faster on the boring parts around that.

I’ve noticed something odd using Claude and ChatGPT. They will ignore explicit instructions. And if I ask why it ignored the instruction it'll say something like: “You’re right, I did ignore that.” Or worse: “I didn’t think it was relevant here.” That's because the model is weighting the billions of backend logic already fed into it than the specific instruction you gave it. I've noticed it happens even if I make the instruction more explicit. It shows that AI is brilliant when the cost of being wrong is low and the feedback loop is cheap. Because while you can give the model context based on your expertise... ...you can’t reliably force it to rank that context above everything else it has learned. That’s why I’m sceptical of fully autonomous systems in sensitive areas. Because “mostly follows instructions” is not good enough for some jobs.

In Q1, I built two AI-based projects at Littledata: 1. A GTM engine for sales prospecting 2. Monitor AI The GTM engine helped us enrich leads, understand their tech stack, and improved our prospecting overall. But it’s a generic SaaS problem. Besides, I'm not a career sales automation guy so I wasn't creating a unique advantage for ourselves. But the idea for Monitor AI came based on 10 years of analyzing Shopify data setups. I know what metrics matter and how to fix big data issues. So I could give it the rules, context, and edge cases to build against. As a result, Monitor AI is built based on my personal knowledge, has a longer shelf life and cannot be replicated in a weekend. I think this is where people get AI product development wrong. They expect AI to create the business logic. But AI is much better at building within rules that you supply than deciding what the rules should be. That still comes from the expert. So if you're building an AI product, ask yourself: “Are we uniquely qualified to build this?”

AI is very good at making more of what you already have. More email variants. More ad hooks. More landing page copy. More reports. That is useful. But I don’t think it makes your marketing strategy better. And this is where a lot of brands are getting slightly carried away. If your positioning is unclear, AI won’t magically sharpen it. If your brand feels generic, AI won’t suddenly make it distinctive. If your creative is not landing, AI can help you make 20 more versions of the same weak idea. Faster. The hard part in marketing is still understanding why people buy. What do they trust? What makes the product feel different? What are they comparing you against? What is happening in the market that changes how they make the decision? AI is still pretty poor at that. It can remix patterns. It can improve structure. It can iterate quickly. But it does not really understand the emotional connection between a customer and a brand. That still comes from talking to customers, watching behaviour, reading reviews, seeing what actually converts, and making judgment calls. So the smart use of AI is not “let’s automate marketing strategy.” It is: Automate the boring execution. Then spend the saved time on the thinking AI cannot do. At Littledata, we see the same pattern with data. AI can help you move faster. But if the signal underneath is wrong, you just get faster answers to the wrong question. Same with marketing. AI multiplies what is already there. So before asking it to produce more, I’d ask: Is the thing we’re multiplying actually worth scaling?

Sometimes the best sales proof is a customer coming back.

I think CPA has become a bit too comforting. It's easy to report, compare, and optimize against. But on Meta, the cheapest customer to acquire is usually the one who already knows you. So if you're asking Meta for the lowest CPA, you'll end up getting buyers who were already going to buy from you. But it won’t necessarily grow your brand. Instead, if you want to grow new customers, you need to pay more attention to CPM. Because it's closer to the cost of reaching the market. CPA is often just the cost of converting the warmest bit of it. That's why I think Meta’s new customer conversion goal is interesting. Of course, CPA will look worse. A genuinely new customer should cost more than someone already familiar with you. But at least Meta is learning from the right event: NEW customer purchase. That matters if you’re a healthcare or supplements brand where Meta has restricted targeting options. Your first-party Shopify data becomes one of the few clean ways to tell Meta what a new customer actually looks like. That way, a higher CPA can still be a right trade-off for a right new customer.

AI is exposing the real value of Shopify agencies. I was talking about it with Fred Cohen and Matt Parkes at Quickfire - Shopify Platinum Partner. They're seeing clients getting carried away with the chunk of code AI spits out. Matt gave a simple example: A client built a countdown timer with AI which counted down fine. But when it hit zero, it just stayed there 😀 No fallback message or next steps. AI can give you an answer but it doesn't know anything about your store, theme, integrations, checkout setup or even the commercial risk. And it definitely doesn’t care about the outcome when something breaks. Besides, ecommerce managers start with commerical problems than technical specifics: “Can we add urgency?” “Can we improve conversion?” “Can we launch this offer quickly?” Translating those business requirements into a technical problem is a huge part of the value. This is why good agencies still matter. Because they know what needs to be built, what can go wrong, and how to make it work for your specific Shopify setup.

Over the last month, I’ve been building an internal companion app at Littledata called Monitor AI. The inspiration for it came from the burglar alarm business model. You pay someone to tell you when something goes wrong. I think ecommerce data is similar - you rarely look at your tracking setups on a daily basis. Only when performance drops or an agency flags something, you look into it. That's what Monitor AI does - it scans marketing tools every day and looks for common issues: → Broken attribution → Unexpected changes in conversion rates → Data inconsistencies between platforms But the most interesting thing about AI is, it helps us convert rules into software quickly without creating new business logic: “If A breaks, check B.” “If B changes, investigate C.” “If this metric drops, compare it against these other signals.” It would have taken way too much time to build this previously. Now we can run hundreds of checks in seconds. That's why the best use cases for AI must satisfy two critieria: 1. You must deeply understand the problem 2. Your customers should need it.

Everyone complains about the rising cost of Meta ads. But few talk about signal quality they're feeding into Meta. If Meta can't attribute purchases to ads, it can't optimize properly either. That is why weak Event Match Quality scores show up as weaker performance. This can be simply fixed with Littledata 1️⃣ Connect Meta Ads in the Littledata app 2️⃣ : Add your Pixel ID and access token 3️⃣ Disable any duplicate tracking on the same pixel 4️⃣ Let Littledata apply the tracking automatically In Events Manager, you'll see cleaner server-side events with richer identity data attached, like hashed email, name, and country. That improves Event Match Quality. You can also split new and returning customer purchases. That matters because prospecting and retention should not be trained on the same event. At Littledata, upgrading your Meta signal quality is one of the biggest missed opportunities we see on Shopify. Watch this demo video to know how to do it for your brand.

I'm surprised how server-side tracking is still a disaster. You hear Google Tag Manager and a dozen apps, and a simple problem feels like a huge engineering project. We've been doing this for eight years, and the difference is actually very simple. ✔️ The classic way You can do server-side with generic Google Tag Manager (GTM) solutions. These are flexible if an agency works across BigCommerce, Magento, and Shopify. But for a dedicated Shopify brand, it requires huge technical setup and constant maintenance. If you want to pull in custom data, like subscription lifecycles, you need to do a lot more data engineering work. 💪 Deep expertise, not generic tools The reality is most brands spend their ad budget on two core channels: Google and Meta. So, there is no value in tracking a small channel like Reddit. Instead, it's in making those core channels perform a little bit better. At Littledata, our technical leadership allows us to go deeper. We update our tracking every month based on platform changes. 🎯 That enables true innovation For example, this year, we launched the ability to target new versus returning customers in your Meta and Google Ads. Why does that matter? You can now set a campaign to optimize only for net new customer acquisition. So although the CPA is higher for a new buyer you stop wasting ad budgets on retargeting people who were already going to buy via email or organic search. You move away from just tracking revenue to profiling and acquiring truly profitable customers. The question isn't whether to do server-side, but how deeply you go.

I'm seeing a lot of brands getting carried away with attribution. While it can correctly credit email, SMS, retargeting, branded search, or last click... It can't tell you what actually caused the sale. So, if you're optimizing for attribution you'll end up doing more of the same bottom-of-funnel activity. Marketing is about creating demand, not just capturing it: → What actually made someone aware of your brand? → What changed their mind? → What made them choose you over alternatives? → Would they have bought anyway? That’s why incrementality is more important. It gives you the answer to “did this campaign actually drive more purchases?” Tools like conversion lift try to answer that by comparing exposed vs non-exposed audiences. It’s not perfect but it’s closer to the real question. I’ve said this before: Attribution doesn’t need to be perfectly accurate. It needs to be directionally correct. So if you’re a marketing manager, spend time understanding what actually moved the needle.

70% of shoppers abandon their carts. But a lot of Klaviyo flows never reach the full audience. That is usually a tracking problem, not an email problem. So before rewriting your flows, ask yourself: "Can Klaviyo actually see enough of our abandoners to trigger the flow?" With Littledata, here's how you do it: Step 1: Connect Klaviyo in the Littledata Shopify app Step 2: Turn on the key flow events Step 3: Clone your existing flow Step 4: Switch the trigger to Littledata’s event Step 5: Add exclusions so people don’t enter both flows That’s it. That improves the signal for: → Abandoned Cart → Abandoned Checkout → Abandoned Browse Because the schema is backwards compatible, your templates and dynamic fields still work. For one of our customers, just one cloned flow generated £3,000 in extra monthly revenue. Across the three flows we power, that can be 20–40x the monthly cost of our software. Watch the demo to see exactly how this works.

We looked into a few multi-currency stores on Shopify. The numbers were way off. The default Google & YouTube app doesn’t handle currency conversion properly. Honestly, we didn’t realise it until we dug into it properly. So if you’re selling in multiple currencies but sending everything into one Google Ads account... your conversion values can be completely off. So a €100 order might come through as 100 SEK - that’s a 10x difference. And if you're using Smart Bidding, it means: → It overvalues some conversions → Undervalues others → And optimizes towards the wrong outcome Result? Your total revenue may look about right. But underneath, it's a mess! At Littledata, we natively support Shopify Markets, which means you can: → Convert currencies into a single ad account → Or send conversions into local currency accounts per market If you’re running Shopify Markets and using Google Ads, I'd check: Does your conversion value actually match Shopify? Because it affects how your budget gets spent.

I’ve seen dozens of Shopify brands tweak the same Klaviyo flow for months: New subject lines, incentives, email designs... But often the bigger issue is simpler: Klaviyo never saw enough of the right people in the first place. That is especially true in abandonment flows. Because these flows only work when the shopper can be linked back to a real profile. They browse. They add to cart. They even start checkout. But if the identity signal is weak, Klaviyo cannot do much with that behavior. That is why I think a lot of teams end up optimizing too far downstream. They rewrite the email when the real problem is who qualified to receive it at all. You do not get the biggest lift by endlessly rewriting the copy. You get it by making sure the platform can actually identify enough of the shoppers showing buying intent. Would you rather improve the message by 10% OR increase the number of real buyers who can receive it?

Yesterday, Pulse set the tone for exactly why we’re backing it this year. The Littledata growth team was out in full force. The day started properly with breakfast & coffee at The Ivy Tower Bridge. Stunning views. Lots of familiar faces and a few new ones. Thanks to Smarterships, Jurni, and Rivo. Then onto the Pulse Padel Tournament with Dataships, Klaviyo, and Global-e. I’m told Will Fisher and Jack Snellgrove couldn’t decide whether it was better to be seriously good at padel… …or disgustingly bad. Either way, they committed fully. Jack even came a close second. Later we headed to Clays with Shero Commerce for some virtual clay shooting. Apparently Jaz F.’s stance is a little casual...even for virtual shooting. But she still managed to beat both me and David, so maybe the rest of us should stop commenting. Then we closed out the night at the official Pulse drinks mixer with the wider ecommerce community. This is the bit of events I always value most because of the kind of conversations that are always better before everyone’s rushing between sessions.

AI shopping means less traffic but more qualified buyers. Paul Rogers gave a good example of how messy buying journeys are getting. He’d been looking at products for a back issue. So Meta started showing him ads for tennis machines → He started comparing them on ChatGPT → Searched Google for one specific product. Amazon was in the mix too. At some point, he bought it. AI won’t always show up as traffic... Which explains why brands are seeing 1-2% traffic from ChatGPT in Google Analytics. That's the issue Shopify brands are missing with AI - None of the shopper research/journey will show up on Google Analytics. Josh Duggan made the point well: "AI takes a lot of the comparison work that used to happen across Google Shopping, product pages, reviews, and tabs… …and moves it one layer earlier." That means you get more qualified traffic. At Littledata, that's the difficult task we're trying to answer: - Who saw the ad? - Who came to the site before? - Who compared products elsewhere? - Who came back and bought? That’s where Shopify customer profiles, account activity, server-side events, and clean order data matter.