ecomcloser.com

Receipts and a calculator on a desk

How to set discount rules for an AI sales chat

Set ecommerce chat coupon rules in the store’s offer engine: specify which products qualify, which customers can redeem the offer, what the basket must contain, and how far the price can fall. Give each offer a stacking rule and an expiry. The AI sales chat can trigger an eligibility check and explain the result, but it should not make up a price or promise an exception the store has not approved.

For each offer, write down when the chat may present it, what the store must check before applying it, and what the shopper sees if the check fails. Recheck the rules when the basket changes and before checkout finalizes the price. Log both accepted and rejected attempts so you can tell whether the chat explained the offer correctly and whether checkout enforced it.

1. Define the offer before writing the chat prompt

Start with one offer a merchandiser could approve without reading a chat transcript. An illustrative rule is: “Take €10 off selected full-price products when the eligible merchandise subtotal is at least €100 before tax. One redemption per eligible customer. No combination with other merchandise promotions. Valid until the stated expiry.” This is a policy example, not a recommendation to discount every €100 basket. The offer still needs a margin check, and “selected” has to resolve to an actual list of qualifying products or variants.

Separate the reason to mention the offer from the conditions for redeeming it. A shopper asking “Can you do a better price?” may give the chat a reason to check for an offer. It does not establish that the shopper is eligible. Similarly, if you want to reserve an offer for shoppers buying from a specific collection, define the collection in the store. Don’t ask the assistant to infer membership from product names or decide that a similar item is close enough.

Give the rule an owner and a version. Record the approved amount or percentage, its scope, start and end times, any redemption limits, the calculation basis and the fallback message. Keep the shopper-facing wording alongside the rule. A useful review is to hand the same example basket to two colleagues and ask them to calculate the expected checkout price independently. If they disagree about which items count or what happens to an existing offer, settle that question before writing the chat copy.

2. Set eligibility at product, customer and basket level

Product: identify the lines that can be discounted

Decide whether the rule applies to whole products, particular variants or a maintained group of items. In an illustrative furniture assortment, an offer might cover a full-price oak chair but exclude a clearance finish from the same product family. The rule should check the actual variant in the cart. It should not assume that an entire product is eligible because the chat recommended one qualifying version. Spell out how the offer treats bundles, marked-down lines and items already covered by an automatic promotion.

Decide whether exclusions mean “no discount on this line” or “no offer on any part of this basket.” Those are different policies. If a shopper adds an excluded accessory to a qualifying purchase, the first policy leaves the accessory at its current price; the second rejects the entire coupon. Also specify what happens when a quantity changes. A discount approved for one unit should not silently apply to every additional unit unless the rule permits it.

Customer: define what counts as proof

If the offer is limited to a customer group, choose which store record establishes membership and when to check it. For an illustrative first-order offer, decide what qualifies as a previous order and whether guest shoppers can establish eligibility. If they cannot, the chat can explain what’s needed to check the offer instead of treating “I’m new here” as verification. Write a separate rule for one redemption per eligible customer. That is not the same as one use per conversation, because the same shopper can open another chat.

Basket: choose the subtotal and the moment to check it

Specify whether the minimum is based on all merchandise or only qualifying merchandise, and whether it is assessed before or after any permitted discounts. Define the treatment of tax and delivery rather than relying on the word “subtotal.” For example, a basket with €90 of eligible goods and a €15 excluded accessory reaches €105 in merchandise, but it fails a €100 eligible-merchandise threshold. The chat should be able to explain that distinction without claiming the basket total is wrong.

Check again after any change that can affect the decision: removing an item, switching a variant, changing quantity, identifying the customer or adding another offer. A coupon can be valid for the basket the shopper just saw and invalid for the basket they are about to buy. Explain what changed and show the current price before asking the shopper to proceed.

3. Calculate the largest discount the store can fund

Use a contribution margin floor, not a discount percentage chosen in isolation. In this explicitly illustrative sale, the starting price is €100 before tax, variable costs are €62, and the minimum contribution margin is 30% of the realized selling price before tax. For this first calculation, assume the €62 stays fixed when the price changes. The question is how much revenue must remain after discounting, not how much profit the store had at the original price.

Contribution margin = (realized selling price − variable costs) ÷ realized selling price. To keep that margin at or above 30%, the €62 in costs can make up at most 70% of the selling price. The minimum selling price is therefore €62 ÷ 0.70 = €88.571428… . Round the price up to €88.58, the first cent that meets the floor. The maximum discount on the €100 base is €100 − €88.58 = €11.42. Rounding the price to €88.57 would put the margin slightly below 30%, even if a display rounded the margin itself to 30%.

As a check, a €10 discount leaves €90 in revenue and €28 in contribution. Dividing €28 by €90 gives about 31.11%, so it clears the floor. A €12 discount leaves €88 in revenue and €26 in contribution. That gives about 29.55%, so it fails. Do not divide contribution by the original €100 price or by the €62 in costs. Neither is the denominator in this rule. The €11.42 is a ceiling under the stated inputs, not the amount the chat should automatically offer.

Recalculate when payment fees, fulfillment costs or store-funded delivery change. If a fee depends on the final price, use the fee at that price in the margin calculation rather than assuming the original €62 is still right. Keep the tax basis consistent between the revenue and cost inputs. For a basket offer, choose whether the policy protects each discounted line or the basket as a whole. If it protects individual lines, test how checkout allocates and rounds the discount across them. A basket that passes in total can still contain a line that fails its own floor.

4. Decide what happens when offers overlap

Write down which benefits can coexist. A no-stacking rule should say whether it covers entered coupon codes, automatic merchandise discounts and any delivery benefit the store funds. For an illustrative policy, a chat coupon might be incompatible with another merchandise promotion but compatible with a delivery offer, provided the resulting order still meets the margin rule. Compatibility and affordability are separate checks: an allowed combination does not automatically pass the price floor.

Next, decide what checkout does when a new offer conflicts with one already applied. It can keep the existing promotion and reject the new one, or it can offer to replace the first promotion under a defined rule. Don’t say the shopper will get the “better deal” unless you know how the comparison is made. If the offers affect different lines or change delivery charges, compare the resulting totals, not just the face value of the codes. Show the shopper what would be removed and what the new total would be before treating the replacement as accepted.

Using the earlier illustrative cost assumptions, suppose a €5 merchandise promotion would remain in place alongside a €10 chat coupon on the same €100 sale. A combined €15 discount leaves €85, below the €88.58 floor. Either offer could pass on its own, but the proposed combination cannot. Decide in advance whether the system rejects such a combination or allows a smaller discount under a separately approved rule. The chat should not improvise a new amount to make the calculation pass.

For combinations you do permit, specify the order of calculation and the base each offer uses. A percentage taken after a fixed reduction is not necessarily equivalent to applying it first. State when the basket minimum is tested in that sequence. Use the same calculation in the chat result and at checkout so the shopper does not have to reconcile two different prices.

5. Make expiry and redemption limits enforceable

Choose an exact expiry timestamp and time zone for a fixed campaign. “Ends Friday” is not enough to test a checkout attempt near midnight. If the offer expires a set time after an event, define the event precisely: for example, issuance rather than the first chat message. Specify whether opening a new conversation can produce a new offer. Otherwise a rule that sounds like a single deadline may behave like a window that keeps restarting.

Decide which event must happen before expiry. If checkout must validate the offer before the deadline, entering the code in chat earlier does not reserve the price. Test a shopper who applies the coupon just before the cutoff and returns to pay after it. If neither the offer nor the price is reserved, remove the discount at the relevant check and show the updated total. State the deadline plainly when presenting an offer rather than creating an impression of urgency without an enforceable cutoff.

Define when a redemption counts toward a per-customer limit: when the coupon is applied, when an order is created or at another event supported by the store’s process. Then decide how abandoned checkouts and failed payments affect that count. Test the same eligible customer in two tabs, not just two messages in one conversation. If the rule promises one redemption, the store needs to enforce the limit across attempts rather than relying on what the chat remembers.

6. Keep voice selection, cart acceptance and payment separate

If a shopper uses voice to choose an item or say which offer they want, treat that as a selection, not proof that the coupon was applied or an order was paid. Show the interpreted variant and quantity, particularly if names are similar or the spoken request is ambiguous. An accepted cart action is a separate event: the store confirms that it added or updated the intended item or applied the offer to that basket. Payment has its own status. Do not turn a successful cart update into a message saying the purchase is complete.

Write the chat messages around those events. An illustrative sequence might be: “You selected the oak finish”; “The €10 offer was applied to the qualifying items in your cart”; and, only after the store reports payment success, a payment confirmation. If the store rejects the offer, don’t use the second message. Explain the reason in shopper terms and show the next available action, such as reviewing eligible items or continuing at the current price. The shopper should not have to start another conversation just to check out without the coupon.

Treat mock previews as illustrations of wording and sequence, not evidence that a live cart was updated, a coupon was accepted or a payment was taken. Apply the same distinction when discussing Senoji Baldinė: attribute the workflow to the owner’s description, and use the public product listing only for facts it visibly confirms, such as the listed product name or displayed options. A listing does not verify how a spoken selection is handled, whether a discount passes checkout validation or whether payment succeeds. Those workflow steps need their own evidence.

7. Log the rule, the decision and the cart result

Design the audit record to answer a specific support question: “Why did this shopper see that price?” Record the offer identifier and rule version, the time of the check, the cart reference, and the product, customer and basket checks that mattered to the decision. Include the relevant subtotal and its basis, competing promotions, the requested discount, and the approved discount if there is one. If you enforce a margin floor, retain the inputs or an internal calculation reference that lets authorized staff reconstruct the decision.

Keep a machine-readable decision reason as well as the message shown to the shopper. “Eligible subtotal below minimum” is different from “expired,” “already redeemed,” or “would breach the margin floor.” The shopper can receive a plain explanation without seeing internal cost figures. Record whether the offer was merely checked, accepted into the cart or removed on a later check. Connect these events with a shared reference, and keep payment status distinct. A single label such as “used” makes it hard to work out whether checkout ever honored the offer.

For example, a useful audit trail could show that an offer passed at an eligible subtotal of €105, then failed after a cart edit reduced that subtotal to €95. It should also show that the cart reverted to the price without the coupon and what explanation the shopper received. Record the fact that a customer check passed or failed without copying unnecessary account details into the log. Store enough to troubleshoot the pricing decision; a full transcript is not a substitute for the rule inputs and outcome.

8. Test checkout with requests you expect to reject

Start with a qualifying basket and a customer who meets the rule. Note the expected line prices and total before running the test, then compare the chat message, cart and checkout. Confirm that the correct rule version and acceptance are logged. A friendly answer in chat is not a pass if the discount never appears, applies to the wrong line or changes the total by an unexpected amount.

  1. Ask for two coupons on one order. Apply the first and request the second. Check that the store rejects or replaces the second according to policy, and that the chat does not imply both were accepted.
  2. Try an expired coupon. Check the deadline in the configured time zone. Test both a fresh request after expiry and a previously accepted cart that reaches final validation after expiry.
  3. Ask for more off a sale SKU. Switch from a qualifying full-price variant to an excluded sale variant during the conversation. Verify that the original approval does not carry over.
  4. Edit the basket below the minimum. Remove a qualifying item while leaving an excluded item in the cart. Confirm that it is the eligible subtotal, not the apparent basket total, that governs the offer.
  5. Continue without the coupon. After a rejection, verify that the shopper sees the current total, can accept that price and can proceed without a discount or a duplicate cart action.

Add boundary tests at the exact minimum and one cent below it. For the illustrative €100 sale, check a discount of €11.42 and then €11.43 using the original cost assumptions. Include a case where the price is checked again after the shopper changes a cost-relevant input such as a funded delivery option. Review the log for every test, including denials. The outcome should be reproducible from the recorded inputs, not just explained by a plausible sentence in chat.

9. Measure rule behavior before expanding the offer

Keep separate counts for offers checked, accepted into a cart and redeemed under your chosen definition. A single shopper may retry a rejected coupon, so distinguish attempts from unique carts when reviewing results. Report rejection reasons and reversals after cart edits. A cluster of failures just below the minimum may prompt a review of how the chat explains which items count; it does not, by itself, show that the minimum should be lowered.

Compare the discount accepted in the cart with the discount on the resulting order where there is one. Review realized contribution on discounted orders using the same denominator and cost basis as the floor, and investigate exceptions such as an unexpected combination or a delivery cost omitted from the check. Track cases where an offer is declined and the shopper proceeds at the current price. That is a distinct outcome from a coupon redemption or a shopper leaving the checkout.

Use those checks to decide what to revise: the chat wording, the eligibility criteria or the underlying calculation. If you change a rule, create a new version and rerun the applicable success and rejection tests before expanding it to more products or customers. The standard for release is concrete: the store makes the pricing decision, the chat reports it accurately, and the log explains why the shopper saw that total.

Try the EcomCloser demo

Related guides

Sources