Most ecommerce stores collect far more customer data than they use. A mid-sized Shopify brand will typically hold order history, email engagement, on-site behaviour, support tickets, product reviews and ad platform signals, spread across five or six systems, and make almost no decisions with any of it. The data is not the problem. The absence of a question is.
This article covers what customer data actually is, where it lives, which parts of it you can trust, and how to turn it into decisions that change what your business does. Some of what follows contradicts advice that was standard five years ago, because the technology underneath it has moved.
What counts as customer data in ecommerce
Customer data is any information your business holds about the people who buy from you or could buy from you. The useful distinction is not personal versus anonymous. It is who gave it to you, and on what terms.
Zero-party data
Zero-party data is what a customer tells you on purpose. Quiz answers, size and fit preferences, a birthday given at signup, a "how did you hear about us" response at checkout, a preference centre selection about how often they want to hear from you.
Zero-party data is volunteered, so it is accurate and it lasts. Its weakness is volume. You only get it if you ask, and you only get an answer if the exchange is worth the customer's time.
First-party data
First-party data is what your own systems observe. Orders, average order value, products viewed, sessions, email opens and clicks, refunds, return reasons, support conversations, subscription status.
You collect first-party data on your own properties, which means you control it and you keep it. Every durable advantage in ecommerce measurement now rests on this category.
Third-party data
Third-party data is bought from brokers or inferred by advertising platforms watching behaviour across sites you do not own. It powered a decade of retargeting. It has become substantially less reliable, and it is worth understanding exactly why before you build anything on top of it.
The cookie situation, stated plainly
Advice written in 2020 assumed third-party cookies would disappear on a fixed date. That is not what happened, and the reality is more awkward than either version of the story.
Firefox turned on default blocking of third-party cookies in 2019, and Safari's Intelligent Tracking Prevention has blocked them by default since 2020. Chrome went the other way. Google abandoned forced deprecation in July 2024, and in April 2025 confirmed it would not roll out a standalone consent prompt, leaving users to manage the setting inside Chrome's privacy and security settings. Most of the major Privacy Sandbox APIs, the proposed replacement, were retired in October 2025.
So third-party cookies still function in Chrome. That is not a reprieve. A large share of your traffic already arrives with cross-site tracking switched off, Chrome users increasingly choose stricter settings, and ad blockers strip what remains. Cookie-based audiences shrink quietly rather than breaking loudly, which is worse, because nobody notices until performance has already drifted.
The practical consequence is simple. Your first-party and zero-party data is the asset. An email address, hashed and passed back to the ad platforms, is doing work that a third-party cookie used to do for free.
Where your customer data actually lives
The data is rarely missing. It is scattered.
A typical Australian ecommerce stack holds orders and customer records in Shopify, behavioural events in GA4, email and SMS engagement in Klaviyo, conversion claims inside Meta and Google Ads, conversations in a helpdesk, sentiment in a reviews app, and acquisition signal in a post-purchase survey. Each system reports its own version of the truth, and none of them agree.
The email address is what stitches the picture together. It appears in the order, the email profile, the support ticket and the review, which makes it the practical join key for almost any analysis worth doing.
Identity is where most reporting quietly breaks
One person buys as a guest in March, creates an account in June with a different email, and buys again from their work address in September. Your reporting now shows three customers, each with one order, and your repeat purchase rate is wrong.
Guest checkout, multiple email addresses, subscription orders recorded separately from one-off orders, and wholesale orders sitting in the same customer table as retail all corrupt the same numbers. Before you trust a single lifetime value figure, find out how many duplicate customer records you are carrying. In most stores the answer is uncomfortable.
Which data you can trust, and which you cannot
Not all of your data is equally honest, and knowing the difference is most of the skill.
Trust your order data. Shopify knows what was sold, to whom, for how much, and what came back. It is the closest thing you have to ground truth.
Treat platform-reported performance with suspicion. Meta and Google both claim the same sale. Add up the conversions each platform reports and you will frequently find you sold more than you actually did. Platform numbers are useful for optimising within a platform and close to useless for deciding how to split budget between them.
Understand that GA4 models what it cannot see. Sessions, attribution and conversion paths in GA4 are partly modelled, because a meaningful share of visitors cannot be tracked end to end. That does not make GA4 useless. It makes it directional. Read the trend, not the third decimal place.
Ask the customer. A post-purchase survey asking how someone heard about you is imperfect, but it is the only source that captures word of mouth, podcasts, in-store conversations and the friend who sent a screenshot. It routinely contradicts the ad platforms, and it is often right.
The four questions worth asking
Customer data earns its keep by answering questions that change a decision. These four cover most of the value.
Who buys again, and how soon?
Repeat purchase rate is the most under-examined number in Australian ecommerce. It tells you whether you have built a brand or a very expensive customer acquisition machine.
Two methods do the heavy lifting. RFM segmentation groups customers by how recently they bought, how often they buy and how much they spend, which surfaces your best customers and the ones quietly slipping away. Cohort analysis groups customers by the month of their first order and tracks what each group spends over the following year, which shows whether the customers you acquired last quarter are worth more or less than the ones you acquired a year ago.
Cohort data makes for uncomfortable reading in most brands. It is also the fastest way to find out whether a channel is buying you customers or buying you transactions.
What is a customer actually worth?
Lifetime value is worth calculating properly or not at all. Use contribution margin, not revenue: strip out cost of goods, shipping, payment fees, returns and discounts, then compare what remains against what you paid to acquire the customer.
The useful output is not a single LTV figure. It is a payback window. If it takes nine months to recover acquisition cost and your supplier wants paying in thirty days, that is a cash flow problem dressed up as a growth strategy.
What do people want that you are not showing them?
Internal site search is the most honest data source in your store. Customers type exactly what they want, in their own words, and a zero-results search is a customer telling you they came to buy something you did not offer, or did not name the way they name it.
Return reasons do similar work. A concentration of "too small" against one supplier is a sizing problem, a product page problem, or both, and no amount of paid media will fix it.
Where does the journey break?
Track the drop between add to cart, checkout started and order placed, and split it by device. Mobile and desktop behave differently enough that a blended figure hides the actual issue. Watch what happens around your free shipping threshold. Watch whether people who use search convert at a different rate to people who browse the navigation.
Turning insight into decisions
Segmentation that changes what you do
A segment is only worth building if it changes an action. "Engaged subscribers" is not a segment, it is a feeling.
Segments that earn their place look like this. One-time buyers, thirty to sixty days after purchase, who have not returned. High-value customers who have never used a discount code, and who therefore should not be sent one. Customers who have only ever bought on sale, whose behaviour you trained and can retrain. Buyers of a consumable product approaching the point where they would run out.
Personalisation, and its limits
Personalisation works when it reduces effort. Recommending a complementary product, remembering a size, sorting a collection so the customer sees the styles they actually wear, timing a replenishment reminder to the week someone runs out.
Personalisation fails when it demonstrates surveillance. Showing a customer the item they bought yesterday, guessing gender from a single browse, or greeting someone by name in a subject line while getting everything else wrong. The rule that holds up: if the customer would be uncomfortable learning how you knew, do not use it.
Pricing and discounting
Regional pricing is normal and expected. Australian shoppers understand that an AUD store and a USD store carry different prices, and Shopify Markets handles this without controversy.
Individualised pricing, where two customers see different prices for the same product based on their behaviour, is a different proposition. It is commercially fragile. Customers compare notes, screenshots travel, and the damage to trust lasts far longer than the margin you captured.
Discounting deserves the same scrutiny. Every discount teaches a customer something. Run cohort analysis on discount-acquired customers and you will usually find they repeat less, return more, and never pay full price again. That is not an argument against discounting. It is an argument for knowing what it costs you beyond the margin on the order.
Service and post-purchase
Data improves customer service in an unglamorous way: it puts order history, delivery status and previous conversations in front of the agent before they reply. Resolution time drops because nobody has to ask the customer to explain themselves twice.
Delivery exceptions are the highest-leverage use of data after the sale. A customer told about a delay before they notice it stays a customer.
Merchandising and buying
Demand data belongs in the buying conversation. Search terms with no results, waitlist signups, size sell-through and sustained out-of-stock demand all say something concrete about what to order next. Most brands hold this data and never put it in front of the person placing the order.
How to earn data instead of harvesting it
The brands with the best customer data are the ones who asked for it and gave something back.
A fit quiz that produces a genuinely better recommendation earns size and preference data. A preference centre that lets people choose how often they hear from you reduces unsubscribes and tells you who wants what. A post-purchase survey asking how someone found you fills the gap the ad platforms cannot honestly fill.
Progressive profiling works better than a long form. Ask for an email at signup. Ask for a birthday when there is a reason to. Ask for preferences after someone has bought once and has a reason to trust you.
Explain why, in one sentence, in plain language. "So we can stop sending you menswear" earns more consent than a paragraph of boilerplate. And collect less. Data you have no use for is storage cost and risk, not an asset.
Where this usually goes wrong
Across the ecommerce brands Growth Huntr works with, the same failures recur.
- Collecting data with no question attached, then building a dashboard nobody opens.
- Building segments that nobody ever sends anything to.
- Believing platform-reported ROAS, then wondering why revenue does not match.
- Carrying duplicate customer records, which quietly understates repeat purchase rate and lifetime value.
- Personalising on stale data, so the site recommends a product the customer already owns.
- Measuring everything and changing nothing.
How to tell whether it is working
Judge the data by the decisions, not the dashboards.
Watch repeat purchase rate, contribution margin by acquisition cohort, payback period against acquisition cost, revenue per email recipient, and the share of revenue coming from customers you can identify. Those five will tell you more than a fifty-tile report.
Then ask the harder question. How many decisions did the business change last quarter because of something the data said? If the answer is none, the reporting is decoration.
Start with one question you cannot currently answer and care about the answer to. Find where the data for it lives. Answer it. Then do the next one. That sequence produces more value in a quarter than a year of dashboard building, and if you would rather work through it with someone who has done it before, we are always happy to have that conversation.
.svg)
.svg)