When we evaluate an ecommerce campaign, we usually start with how visitors responded to it. Did they click the message? Sign up for email? Return to an abandoned cart? Complete a purchase? Those are reasonable things to measure, and they help us understand whether the experience is doing what we expected it to do.
What I think we need to question more often is whether the behavior we’re measuring would have happened without the campaign, and whether we’ve had a hand in creating that behavior in the first place. We can get very good at encouraging a particular action, build reporting around it, and then use the results to justify doing more of the same. That doesn’t necessarily mean we’ve made the experience better or created more value for the business.
It’s easy to see how this happens with discounts and abandoned carts, but I don’t think the problem is limited to promotional marketing. It can show up anywhere we’re using visitor behavior to decide what to put in front of someone, especially when we treat a response to our experience as proof that the experience was needed.
When the recovery campaign becomes part of the shopping behavior
Imagine someone adds a $200 product to their cart and leaves the site. They get an email reminding them about the item and, later, an offer for 15% off. They return, place the order, and the campaign receives credit for recovering the sale. I get why we’d consider that a success. The visitor purchased, the reminder or offer was part of the journey, and we can see the transaction in our reporting.
But there’s another possibility. Maybe the visitor has shopped with the brand before and knows that leaving something in their cart is a pretty reliable way to get a discount. In that case, the campaign may have encouraged the abandonment rather than solved it. We aren’t necessarily persuading someone who wouldn’t otherwise purchase. We could be giving them a reason to wait.
A 2026 Wunderkind report, The Battle for the BFCM Shopper, offers some useful context. In a survey of 300 U.S. consumers, 40% reported deliberately leaving items in carts hoping for a follow-up offer, 36% reported subscribing to email or text in exchange for a discount, and 20% reported using multiple email addresses or identities to take advantage of incentives. These are self-reported shopping behaviors, not evidence that a particular retailer’s campaigns caused them. Still, it’s hard to look at those findings and assume that every cart abandonment or new email signup means what we want it to mean.
To be fair, there’s nothing particularly irrational about the shopper’s behavior. If a retailer repeatedly offers a better price after someone leaves, why wouldn’t they learn to wait? The more interesting question is whether the retailer has accounted for that response when measuring the campaign. A report that counts orders placed after an incentive can look very healthy without telling us how many of those orders required the incentive.
Leaving a site doesn’t tell us why someone left
There’s a second issue with cart recovery that has less to do with shoppers looking for a deal and more to do with the assumptions we make about intent. We tend to interpret abandonment as a moment of hesitation or an obstacle to purchase. Sometimes it is. Someone might be surprised by shipping costs, unsure about the product, or comparing the price with another retailer. In those cases, an intervention might help, although the right response isn’t necessarily a discount.
Other times, life just gets in the way. Maybe their dog threw up and they have a mess to clean up. Maybe they need to leave to take their kids to practice. Life happens, and a discount isn’t going to prevent any of that. The visitor could have every intention of buying later, but from our perspective they look very much like someone who was about to leave for good.
That’s one of the limitations of behavioral data. We can see that a visitor added something to a cart and didn’t purchase in that session. We can’t see why. The same goes for many of the signals we use for personalization. A repeat product view might mean serious consideration, uncertainty about the price, or simply that someone forgot which item they looked at yesterday. One action can have several plausible explanations, and those explanations don’t all call for the same experience.
This is where intent becomes useful, provided we’re honest about what we can and can’t infer. Looking at a broader pattern of behavior may help us distinguish someone exploring options from someone preparing to purchase, but even a good model isn’t going to tell us that a visitor’s dog just threw up. We should be using those signals to make better decisions about when to help, not pretending we can eliminate uncertainty about why people behave the way they do.
The campaign can win while the business gives away margin
Going back to that $200 order, a 15% discount reduces the selling price to $170. If the visitor would have paid full price without the offer, we’ve given up $30 without creating an additional order. Let’s say the product cost the retailer $120 before other variable expenses. That leaves $80 in gross profit at full price compared with $50 after the discount, a reduction of 37.5% in gross profit dollars for that order.
Of course, the calculation changes if the visitor wouldn’t have purchased without the incentive. In that case, the discounted order may contribute profit we wouldn’t otherwise have received. That’s why I wouldn’t argue that recovery discounts are bad or that we should remove them across the board. I’d want to understand where the offer actually makes a difference and where we’re simply reducing the value of a purchase that was already likely to happen.
It’s also why the usual campaign metrics can be misleading. A recovery platform may report attributed revenue, a promotional team may celebrate conversion, and a loyalty or email team may point to increased signups. Each may be accurately describing what happened inside its own program. But none of those numbers, on its own, establishes the incremental impact on the business. Revenue per visitor provides a more complete view than a campaign click or conversion rate, although when discounts are involved I’d also want to look at margin and whether purchases were simply shifted from one day to another.
At scale, that distinction matters. If we routinely give incentives to visitors who would have come back anyway, small amounts of unnecessary discounting can add up quickly. The risk isn’t just that one campaign is measured incorrectly. It’s that we see a seemingly successful result and invest more budget, more placements, and more development effort into expanding a tactic whose actual contribution we haven’t established.
We’re making similar assumptions with the rest of the onsite experience
Discounting is an obvious example because the cost is easy to calculate, but consider how we measure many of the other experiences we put on a website. A visitor clicks a recommendation module and we count the interaction. Someone opens a shopping assistant and we look at the engagement rate. We show an email signup prompt, capture an address, and count that as acquisition. All of those metrics have a purpose. What they don’t necessarily tell us is whether the visitor was better off because the experience was there.
Imagine someone arriving directly on a product detail page because they saw an item in Google Shopping. They want to check the price, look at a few photos, and find out when it will arrive. Instead, they get a welcome offer before they’ve had a chance to look at the product. Some people will sign up, particularly when there’s money involved, but that doesn’t establish that interrupting them was the right decision. We still need to understand what happened to the purchase journey, including the visitors who dismissed the offer and the ones who might have purchased without it.
The same logic applies to personalization. We can use behavior to identify visitors who are likely to interact with a certain message and then get better at showing that message to those visitors. That’s useful if the message helps them do something they needed to do. But if we’re only improving the interaction rate, we’re optimizing our ability to get attention, which isn’t quite the same as improving the shopping experience.
There’s an organizational side to this, too. One team wants email addresses, another wants warranty attachment, another wants credit applications, and marketing has several promotions to communicate. Each objective may make sense individually, but when we put all of those messages on the same pages, they start competing with one another and with the reason the visitor came to the site. I’ve compared it to Times Square before. The answer isn’t necessarily to remove every message. It’s to be more deliberate about which messages are relevant to which visitors, and when they deserve attention.
Personalization should change the decision, not just improve the targeting
One of the things I like about personalization is that we don’t have to keep treating every visitor as though they need the same experience. We can use what people are doing now, what they’ve done previously, and what they explicitly tell us to help make a reasonable decision about what might be useful. Someone browsing a broad category may need help narrowing the choices, while someone returning to a specific product and checking delivery details may be much closer to making a decision. Those are different situations, even if both visitors end up leaving without a purchase.
But there’s a trap here. If our objective is simply to increase response to a promotional offer, personalization can make the original problem worse. Suppose we identify a group of visitors who regularly abandon carts and respond well to discounts. We can target those visitors more precisely and potentially improve the campaign’s conversion rate. What we haven’t established is whether those particular visitors needed an incentive, or whether they’re the people most likely to take one when it’s offered.
I’d rather distinguish between visitors who are likely to respond to an experience and visitors whose outcome is likely to change because of that experience. It’s a subtle difference, but an important one. Someone can be highly likely to redeem a discount and equally likely to purchase without it. Someone else may be unlikely to click a promotional message but would benefit from a clearer shipping explanation or a better way to compare products.
That doesn’t mean we need to build a unique experience for every visitor or divide our traffic into dozens of tiny segments. In fact, that can create more complexity than value. I’d start with broader groups whose needs are meaningfully different, test whether those differences justify a different experience, and get more specific only when the evidence supports it. I’ve written more about how this can work without relying entirely on known customer profiles in Personalization for Anonymous Visitors.
What I’d want to learn before expanding the campaign
For a recovery campaign, I’d start by testing the value of the intervention itself rather than jumping straight into different discount amounts or creative. Randomly hold back the incentive from a comparable group of eligible visitors and follow both groups for long enough to capture purchases that happen later. Then compare more than the orders attributed to the message. I’d want to see total purchase rate, revenue per visitor, average order value, discount cost, and contribution margin. If visitors in the no-incentive group still return and purchase, that’s important information, even if those sales never show up in the recovery campaign’s reporting.
I’d also want to understand whether the results differ by visitor need. Perhaps the incentive helps people who have shown repeated price sensitivity but makes little difference for visitors who were already close to purchasing. Or perhaps a group that appears to be struggling responds better to product information than to a discount. Those are hypotheses to test, not assumptions I’d bake into the targeting rules from the start. In particular, we need to define the audiences before treatment where possible; comparing people who clicked an offer with people who didn’t can tell us about engagement, but it doesn’t establish the offer’s incremental effect.
There’s also a longer-term question that a short campaign report won’t answer: are our interventions changing how people shop next time? If visitors learn that leaving gets them an incentive, a campaign could produce an immediate lift while making full-price purchases less likely in the future. That would take a different measurement window and potentially a different experiment to understand. It’s more work than looking at recovered revenue, but it gets us closer to the actual business decision.
The same principle extends beyond discounts. Before expanding a signup prompt, recommendation, or shopping assistant, I’d want to know whether it improves the broader journey compared with not showing it. Sometimes the answer will be yes, sometimes only for a particular audience, and sometimes the feature may generate engagement without providing enough value to justify its presence. Interaction is still worth measuring, but it belongs alongside outcomes rather than standing in for them. I cover that distinction in more detail in my personalization metrics guide.
Sometimes the better experience is to do less
I don’t think retailers need to stop using discounts, pop-ups, reminders, or promotional messaging. There are situations where each of those can help, and getting rid of them indiscriminately would be another version of the same one-size-fits-all thinking. The point is to stop assuming that because a visitor responds, the intervention was necessary, or because a campaign reports revenue, it created all of that revenue.
For me, the useful shift is to start with what the visitor appears to need, recognize how much we still don’t know, and then measure whether changing the experience helped. Sometimes that will mean offering a discount or giving someone more guidance. Sometimes it will mean letting them finish what they came to do without another message getting in the way. And sometimes they’ll leave because they have to take the kids somewhere, clean up after the dog, or do something that has nothing to do with us.
We have more data and better tools than we’ve ever had for influencing onsite behavior. That’s valuable, but it also makes it easier to become very good at producing the signals we like to see in our dashboards. I’d rather know that we’re helping visitors make progress and creating incremental value for the business, even if the result is fewer clicks, fewer messages, or fewer campaign-attributed conversions.
Research referenced: Wunderkind, The Battle for the BFCM Shopper (2026), page 3; survey of 300 U.S. consumers. The percentages describe self-reported behavior and do not establish that retailer incentives caused those behaviors.