I run a small store and, like most people who sell things online, I got hooked on discount codes early. First-order-off, seasonal sales, abandoned-cart coupons. They work, in the narrow sense that they get people to click “buy.” What took me longer to notice was a specific group of repeat customers hiding in my own numbers: people who had bought from me five or six times and had never once paid full price.
That’s not the same as someone who used one welcome coupon and then converted to a normal customer. This is a customer who waits. They open the emails, they add things to a cart, and they check out only when a code shows up. I want to walk through how to actually find these customers in your own data, what you can reasonably measure about them, and where the analysis runs out and you’re left making a judgment call.
Defining the segment before you analyze it
You can’t study “discount-only customers” as a vague feeling. You need a rule you can run against your order data. The simplest version: pull every customer with two or more orders, then check what share of their orders used a discount code or fell during a sitewide sale. If it’s 100%, or close to it, they go in the segment.
This is basically an RFM cut (recency, frequency, monetary) with a fourth column bolted on for discount usage. If you’re running Shopify, WooCommerce, or almost any subscription platform, this data already exists in your orders table or export. The hard part isn’t the query. It’s deciding your threshold. Do you count a customer who’s bought 10 times and used a discount 9 of those times as “discount-only”? I don’t have a universal answer for that. I set my own cutoff at 80% or higher, because in my business that’s roughly where the behavior pattern starts looking consistent rather than coincidental. Your threshold should come from looking at your own distribution, not from copying mine.
What retention actually looks like for this group
Once you’ve got the segment, the next question is whether they stick around. This is where cohort analysis earns its keep. A cohort, here, just means “everyone who made their first purchase in a given month,” tracked forward to see what percentage are still buying in month two, three, six, and so on.
Build two cohort retention curves side by side: one for customers whose first order was full price, one for customers whose first order used a discount. Plot the percentage of each group still active at each interval. What you’re looking for isn’t a single number, it’s the shape of the curve. Does the discount cohort’s retention line drop off faster after the first purchase? Does it plateau at a lower level? Or does it actually track close to the full-price cohort, meaning the discount got them in the door but didn’t change their long-term behavior?
I want to be direct about something here: I’m not going to tell you what your curves will show, because I don’t know your business, and anyone who claims discount buyers always churn faster is generalizing past what a cohort chart can prove. What the chart can tell you is descriptive fact about your own customers. It can’t tell you why, and it can’t promise the pattern holds for next year’s cohort.
Customer lifetime value needs a margin correction
CLV gets thrown around loosely, so let’s be specific. The simplest version is average order value times average purchase frequency times average customer lifespan. That formula treats every dollar of revenue the same. For a discount-only segment, that’s the wrong lens, because the dollars aren’t the same. A $40 order at 25% off nets you different margin than a $40 order at full price.
If you want an honest CLV comparison between segments, run it on gross margin dollars, not revenue. Take each order, subtract cost of goods and the discount amount, and use that adjusted figure in your CLV calculation. It’s more work than pulling a revenue report, but without it you can end up concluding a segment is valuable when it’s actually just generating volume at a thin or even negative margin. I’ve had product lines where the “loyal” discount buyers were, on a margin basis, barely worth shipping to.
The question you actually want answered
Underneath all of this sits one real question: if you stopped discounting to these customers, would they still buy, or would they leave? No amount of historical analysis answers that on its own, because you’re asking about a world that hasn’t happened yet. That’s what A/B testing is for.
The test itself is straightforward to describe: take a slice of your discount-only segment, split it randomly into two groups, offer one group their usual discount and the other group full price (or a smaller discount) on their next purchase opportunity, then compare conversion and revisit rates. What’s not straightforward is running it responsibly on a small list. If your discount-only segment is 200 people, splitting it in half and looking for a meaningful difference in purchase rate is asking a lot of a small sample. A difference of a few percentage points between the two groups could easily be noise rather than a real effect. Before you act on a test like this, it’s worth checking whether the gap you saw is bigger than what you’d expect from random chance alone, given your sample size. If you’re not sure how to check that, that’s a sign to treat the result as a hint, not a verdict.
There’s also a real cost to running this test: some of the people in your no-discount group may leave and not come back, and you generally won’t get a clean answer about why. That’s a decision you have to be willing to make going in, not something the test will resolve for you afterward.
What the numbers can’t decide for you
Here’s where I want to be honest about the limits. Even a clean cohort curve, a properly margin-adjusted CLV, and a well-run A/B test only tell you what happened, under the specific conditions you tested, with the specific customers you had. They don’t tell you what will happen if you change your discount strategy across the board, they don’t tell you what your competitors will do in response, and they definitely don’t guarantee any particular outcome for revenue or growth. I’ve seen people treat a single test result like a permanent rule for their business. It isn’t. Customer behavior shifts with the season, with the economy, with what else is in their inbox that week.
What the analysis does give you is something more modest and more useful: a clearer picture of who’s actually buying from you and on what terms, instead of a gut feeling. That’s worth having even when it doesn’t hand you a clean answer.
Where I’d actually start
If you’re doing this for the first time, don’t try to build all of it at once. Pull your order history, tag customers by discount usage, and just look at the cohort retention curves first. That alone usually tells you whether this is a real pattern in your business or something you noticed in two or three customers and generalized too fast. From there, adjust CLV for margin before you draw any conclusions about segment value. Only run the A/B test once you have a segment big enough that a real difference wouldn’t get lost in noise, and go in accepting that some customers in the full-price group may not come back.
None of this is a shortcut to a decision. It’s a way to make the decision with your eyes open instead of guessing.
If you want more of this kind of breakdown, plain explanations of the metrics and methods behind actually running a business on your own numbers, you can find the rest of what we’ve covered on the home page.
Get new guides and videos first — join the Telegram channel.