Discounts work best when you know who actually needs one.
Most ecommerce email campaigns make the discount decision at the campaign level. A brand decides to run 20% off, builds an audience and sends the same offer to everyone. Some customers may genuinely need the incentive to purchase. Others may have bought anyway.
A better approach starts by understanding how customers have responded to price in the past. At a basic level, that means separating them into discount buyers, full-price buyers and neutral or unknown customers, then using that context to decide whether an incentive is needed and how strong it should be.
For customers with little purchase history, the answer will not always be obvious. That does not mean they should automatically receive a discount. It means the brand should use the information it has and be selective about when an incentive is likely to change the outcome.
Discounts are easy to use, but expensive to get wrong
Discounts are simple to understand, easy to communicate and can give customers an immediate reason to purchase.
But every discount has a cost.
If a customer would have purchased at full price, giving them 20% off may not have created a sale, it may simply have reduced the margin on a sale that was already likely to happen. Frequent promotions can also train customers to wait for the next offer and affect how they perceive regular prices. Shopify’s guidance on promotional pricing points to lower margins and reduced perceived value as risks of frequent discounting.
This is why the useful question is not simply whether a discount increased conversion, it is whether the discount changed customer behaviour enough to justify what the brand gave away.
One campaign-level discount can make two different mistakes
Imagine a campaign offering 20% off to 50,000 customers. Some regularly purchase at full price, others consistently buy during promotions.
For the second group, the discount may be the reason they purchase, for the first it may be unnecessary.
Discount depth creates the same problem. A price-sensitive group might have converted at 10%, making 20% more than necessary. For another group, 10% may not be enough to change their decision.
This is why promotion performance cannot be judged only by the revenue generated during a sale. McKinsey recommends evaluating promotions partly by customers’ willingness to purchase without them and the additional transactions and margin they actually create.
Instead of asking “What discount should this campaign have?”, the better question is “Which discount segment is this customer in, and what offer makes sense for that segment?”
Start with the customer’s discount segment
You do not need dozens of segments to make discounting more intelligent, these three are a useful starting point:
Discount buyers consistently tend to purchase when an incentive is available. Their orders may cluster around sales, promotional campaigns or discounted products.
Full-price buyers have demonstrated that they are willing to purchase without an incentive. They can still receive offers, but there should be a reason for doing so.
Neutral or unknown customers do not give you enough evidence either way. New subscribers, first-time buyers or customers with limited purchase history may fall here. A lack of data should not automatically be interpreted as price sensitivity.
These segments should not be permanent: customer behaviour changes, and price sensitivity can also vary by category.
McKinsey describes a similar approach, where customers are grouped by characteristics such as discount sensitivity, purchase frequency and product preferences, and promotions are targeted accordingly.
Price sensitivity shows up in purchase behaviour
Using a discount once does not necessarily make someone a discount buyer. They may simply have used a code that was available when they were already planning to purchase.
What matters is the pattern:
- Do their orders consistently happen during promotions?
- Do they purchase at full price between sales?
- Have previous incentives brought them back?
- Do they pay full price in some categories but wait for discounts in others?
That last distinction matters because price sensitivity does not always apply to the customer as a whole. Someone might replenish one product at full price every month while only buying another category during promotions.
The purpose of the segment is not to permanently label someone, it is to give the brand better context for the next discount decision.
A discount should solve a problem
There are situations where an incentive makes sense.
A genuinely lapsed customer may need a stronger reason to return. Someone considering a category for the first time may need an incentive to try it or a shopper who repeatedly browses without converting may be hesitating because of price.
In each case, the discount is trying to change behaviour that might not happen otherwise. This is different from sending an offer simply because a promotional campaign is scheduled.
A discount earns its place when there is a reasonable case that it changes the outcome.
Some customers do not need the incentive
Consider a customer who replenishes the same skincare product every six weeks and has purchased it several times at full price.
Sending them 20% off at week five may produce a conversion, but that does not mean the discount caused it. They may simply have made the purchase they were already about to make.
The same applies to loyal full-price customers or someone who continues to show strong purchase intent.
That does not mean these customers should never receive promotions, an offer could be used to encourage category expansion, increase basket size or reward loyalty. The point is that the incentive should have a job to do.
Discount depth is a decision too
Deciding that a segment needs an incentive leaves another question: how much?
Suppose a group of discount-sensitive customers historically converts with 10% off. Giving them 20% may simply make every resulting order less profitable. The opposite is also possible. An incentive that is too small may not change behaviour at all.
McKinsey makes a similar point about markdowns: one-size-fits-all strategies can result in reductions that are deeper than necessary in some cases and too weak in others.
The decision therefore should not stop at discount or no discount. The depth should also reflect the segment and situation.
AI makes discount segmentation easier to operate
None of this is impossible without AI. A marketer can identify promotion-heavy buyers, separate customers with a strong full-price history and build campaigns for each group.
The problem is maintaining that logic as the customer base grows.
Customers move between segments and price sensitivity can vary across categories. New customers arrive without enough history to classify confidently, and most importantly: more segments also mean more audiences, campaign rules and decisions to maintain.
AI can continuously evaluate purchase history, promotional behaviour, category preferences and engagement to help determine which discount segment a customer currently belongs to.
Campaigns can then adapt around those segments. Full-price buyers might receive the product or message without an incentive. Discount buyers might receive an offer based on their previous behaviour and neutral customers can be handled selectively until there is enough evidence to understand their price sensitivity.
The underlying idea is still segmentation. AI makes it easier to keep those segments current and use them across more campaigns without manually maintaining every rule.
As we discussed in Email Personalization for Ecommerce, personalization is not only about changing what appears inside an email. The offer a segment receives can be personalized too.
A practical discount decision table
| Discount segment | What we know | Likely behaviour without discount | Right offer | What the default costs |
|---|---|---|---|---|
| Full-price buyer | Regularly buys without promotions | May purchase anyway | No discount unless there is another objective | Unnecessary margin loss |
| Discount buyer | Purchases cluster around incentives | Less likely to buy at full price | Discount based on previous response | Too little may not work; too much loses margin |
| Neutral / unknown | Not enough history | Uncertain | Test selectively | Training customers to expect promotions |
| Category-specific discount buyer | Price sensitivity varies by category | Depends on category | Discount where sensitivity exists | Applying one rule across different behaviours |
| Full-price replenishment buyer | Inside normal reorder window | Likely to purchase without help | Reminder or product-led message | Paying for an already likely purchase |
| Lapsed discount buyer | Promotion-sensitive and outside normal purchase window | Less likely to return | Targeted reactivation offer | Discounting deeper or broader than necessary |
The exact segments will differ by brand. What matters is that the decision starts with what the brand knows about the customer’s relationship with price, rather than whatever code happens to be attached to the campaign.
The goal is to discount where it changes the outcome
Discounting is not inherently good or bad. A well-placed incentive can bring back a lapsed customer, encourage category trial or convert someone who genuinely needs a stronger reason to purchase.
The problem is giving the same incentive to everyone because it is easier to run the campaign that way.
Better discounting means understanding how different segments respond to price, deciding whether they need an incentive and choosing the right depth. AI makes that easier to operate at scale, but the principle remains the same.
The goal isn’t to discount less. It’s to discount where the discount actually changes the outcome.
Nexie is a retention marketing agent for e-commerce brands. It reads your customer data across engagement, recency, category interest and lifetime value together, surfaces the audiences worth campaigning to, and builds those campaigns for your team to review and approve, inside the guardrails and marketing policies you set. See your segments →