Abandoned carts: how to identify the causes and build a recovery process

An abandoned cart and an abandoned checkout are not the same

Adding to cart can be a way to save a product or to calculate the total price. Starting checkout indicates a different step, but it doesn't guarantee firm purchase intent either. Define the stage analysed and the interval in which you check whether an order appeared.

An event can be fired several times by the same person. Don't simply divide the raw number of purchases by the raw number of add-to-carts and then call the result the rate of people who complete. Use a consistent population: sessions, users or identifiable carts, with explicit limits.

Check what the platform can observe

Shopify documents recovering abandoned checkouts. Check the features, conditions and data available in your store. A tool can't automatically contact every anonymous visitor, and you shouldn't assume that all abandonments are eligible for messages.

In behaviour analysis, GA4 ecommerce events can describe the buying steps if implemented correctly. The technical documentation defines the events; commercial interpretation still requires a well-chosen path and denominator.

First check the obstacles that prevent a real order

The total cost appears too late

If shipping or an important fee becomes visible only at the end, the customer can reconsider the offer. Test the process as a new buyer and note the moment you learn the full amount. Don't assume the information on a separate page is easy enough to find.

Clarify thresholds and conditions without contradictory messages. “Free delivery” must have the relevant context when there are exceptions. A message that attracts the click through a wrong impression can produce carts and then abandonment, without the problem being the lack of a voucher.

The choice of variant isn't clear

Size, colour or compatibility can remain unclear until the cart. Check the name and image of the variant at each stage. If the user can't confirm what they chose, they may postpone or give up.

A change of quantity or variant must update the price and availability coherently. Also test going back from checkout to the product. Losing the selection or the data entered can turn a simple correction into a reason for abandonment.

The form and payment create errors

Try representative addresses and payment methods. Check the error messages, data retention and the return from the processor's page. Don't test only from an administrator's account, on a fast device and with data already saved.

Don't ask for fields without a clear operational reason. Explain information that may seem unusual and keep errors next to the affected field. If your Shopify store needs work, describe the problem through reproducible steps, not just the impression that checkout is complicated.

The customer isn't ready yet

Some products require measurements, another person's agreement or comparisons. In these cases, a buying guide can be more useful than pressure to buy right away. Abandonment doesn't automatically prove a faulty experience.

Gather real questions from support and use them to improve the page. Don't attribute precise intentions to each visitor from a single event. The data shows observed behaviour, and the explanation must be verified through other evidence.

Build recovery around relevant help

The first message can clarify

A useful message can remind the person of the product and offer a simple way to continue, together with the information that was missing. Avoid promises of reservation if the stock isn't actually reserved. Check the link, the variant and the price before automating.

The tone must be natural, without intrusive claims about every move the person made. Offer help and an option to stop the communication, according to the applicable rules. Don't assume that having an email address allows any commercial message.

The second message must bring something else

Identical repetition may not answer any new objection. You can clarify delivery, the size guide or usage, if the information is relevant. The choice of content must take the product into account, not just the time elapsed.

Don't invent urgency, limited stock or expiries. If there is a real promotion, the terms and the period must be correct in the message and on the site. Trust lost through a false claim can cost more than a recovered order.

The discount is an economic hypothesis

An incentive can help some people, but it can reduce contribution and can be used by people who would have bought anyway. Define eligibility, stacking and the limit. Check that the automation stops messages after purchase and doesn't send several contradictory offers.

Retargeting ads from Meta Ads or Google Ads must be coordinated with the emails. Don't evaluate each channel as if it were the customer's only interaction.

Three hypothetical examples so you don't overestimate recovery

Hypothetical example 1: the measured path

In a defined 30-day cohort you have 500 sessions with an add-to-cart. Of these, 300 start checkout and 120 include an order each. The cart–checkout transition is 300 / 500 = 60%, and checkout–order is 120 / 300 = 40%.

The final rate on the cart cohort is 24%. The 380 sessions without an order in the observed path are not 380 certain lost buyers. Some people may come back later, and others may not have had firm intent.

Hypothetical example 2: orders attributed to the message

You send eligible messages to 200 checkouts and the system attributes 20 orders, of which 18 are retained. The reported recovery rate on this denominator is 10%, and the commercially adjusted one is 9%.

These percentages don't automatically measure the incremental effect. Without a suitable comparison, we don't know how many people would have bought without the message. Use the label “attributed” and don't turn the figure into revenue created with certainty by the automation.

Hypothetical example 3: contribution after the voucher

An order has net revenue of 250 RON and variable costs of 170 RON: a contribution of 80 RON before marketing. A voucher of 25 RON reduces the contribution to 55 RON, assuming unchanged costs. If you allocate 20 RON of recovery cost, 35 RON remains before fixed costs.

For 18 retained orders, the remaining contribution is 18 × 35 = 630 RON in this model. Don't compare only revenue with the value of the vouchers. Include the real costs and the allocation convention for the tool or management.

A process that links data to interventions

Choose a few cases and reconstruct the path without exposing personal data unnecessarily. Record the device, the stage, the error and the conditions. A reproducible description helps the fix more than an aggregate percentage without context.

Keep a list of changes and check after implementation. A form fix can also affect measurement, and a new app can introduce another message. Don't assume that solving one problem automatically leaves the rest of the path unchanged.

A plan for the first four weeks

In the first week, define the stages and verify the data. In the second, test buying on the relevant devices and payment methods and fix the blockages. In the third, prepare a short sequence of messages, with conditions and exclusions verified. In the fourth, analyse retained orders and contribution.

Don't launch a long sequence just because the tool allows it. Each message must have a role and a stopping condition. If the volume is small, keep conclusions cautious and use observations from support to prioritise improvements.

For a coordinated analysis of recovery and the buying experience, contact BlueAds. Start the conversation with the stages and examples of blockages, then move on to automation.

Frequently asked questions

Final checklist

• The abandonment stage is defined.
• The numerator and denominator are compatible.
• Checkout has been tested on mobile.
• The total cost is clear.
• Messages have conditions and exclusions.
• Incentives are evaluated through contribution.
• Attributed orders are not confused with the causal effect.