The best ecommerce chatbot margin controls compare each eligible incentive against a contribution floor on the cart the shopper actually has. Calculate what the store would keep after a percent discount, a delivery subsidy or a bundle, then let the promotion engine make the final decision about eligibility, stacking and the applied amount. The chatbot should explain that outcome, not promise a saving from an earlier preview.
In a fictional €90 basket with €53 in variable costs, an 8% discount leaves €29.80 in contribution, while a €6 delivery subsidy leaves €31. Adding a €15 accessory that costs €7 to provide raises contribution to €45 before any extra promotion. Those figures show why the offer that sounds most generous isn’t necessarily the one that costs the store most.
1. Define contribution on the same basis for every offer
Start with the cart as it stands before the chatbot adds anything. If a sale price is already in effect, use that price, not the original list price. If the shopper already qualifies for free delivery, use zero customer-funded delivery in the baseline. Otherwise, you may count the same promotion twice or offer a benefit the customer already has.
Use a consistent definition: merchandise revenue after discounts, plus any delivery amount paid by the customer that you recognize as revenue, minus variable costs. Decide how to handle tax before setting thresholds. In the fictional examples here, amounts exclude tax collected for remittance. Include product cost, fulfillment, and any payment or other order-level fees that vary with the transaction under your costing model. If a fee changes when the payable amount changes, recalculate it rather than copying it from the original cart.
Delivery is where comparisons often go wrong. If your baseline counts the customer’s delivery payment as revenue and the carrier charge as a cost, waiving a €6 customer charge reduces contribution by €6 when the carrier charge is unchanged. Do not also deduct €6 as a new expense. If the offer instead requires a more expensive delivery service, account for that incremental carrier cost as well. Whatever convention you choose, use it for all three offers.
Set the contribution floor before approving the chatbot’s offer choices. A store might require an absolute euro amount per order, a percentage of net revenue, or both. State the denominator if you use a percentage and define whether equality passes. For this guide, the fictional policy is a minimum of €30 per order, with equality allowed. The decision is based on the full cart, including any ineligible lines that remain in it. If a required cost is missing, don’t treat it as zero; leave the incentive unconfirmed until you can evaluate it.
2. Work the three offers through one fictional cart
Take a €90 item with €53 in variable costs, including the store’s normal fulfillment cost. Its baseline contribution is €37. For this simplified comparison, there is no separate customer delivery payment in the starting figures. The free-delivery scenario adds a €6 net subsidy relative to that baseline. All other variable costs are held fixed unless the example specifies a change.
| Cart option | Net revenue | Variable costs | Contribution | Meets floor? |
|---|---|---|---|---|
| Original item, no added offer | €90.00 | €53.00 | €37.00 | Yes |
| Original item with 8% off | €82.80 | €53.00 | €29.80 | No |
| Original item with €6 delivery subsidy | €90.00 | €59.00 | €31.00 | Yes |
| Original item and €15 accessory, before extra promotion | €105.00 | €60.00 | €45.00 | Yes |
Eight percent of €90 is €7.20. That takes revenue from €90 to €82.80 and contribution from €37 to €29.80, missing the floor by €0.20. The delivery subsidy takes €6 out of contribution instead, leaving €31 and just €1 of headroom. The accessory adds €15 in revenue and €7 in variable costs, so it adds €8 in contribution and takes the combined cart to €45.
Notice what the bundle row does not show: a saving for the shopper. It compares a larger purchase at the stated prices with the single-item order. If you subsequently give €10 off the two-item bundle, revenue becomes €95, costs stay at €60 in this example, and contribution drops to €35. That is €2 below the original full-price single-item order, even though it clears the floor.
Use the comparison to decide which offers are worth checking, not which ones to announce as applied. On the single-item cart, free delivery is a candidate if checkout confirms the quoted subsidy and eligibility. The accessory is an option only if the shopper wants it and the cart reflects that choice. The 8% discount is out under this policy. A positive contribution alone does not make it acceptable.
3. Check what free delivery costs for this cart
Free delivery should be evaluated against the shopper’s current delivery option, not the store’s average parcel cost. Start by asking whether the shopper would otherwise pay for delivery, which service qualifies for the offer, and whether the destination and items are known well enough to price that service. If the method is not settled, describe the offer as conditional rather than attaching a confirmed zero to delivery.
Consider a fictional cookware cart with the same €37 starting contribution. With the €6 net subsidy in the comparison, it retains €31. If the actual cart and delivery method require an €8 net subsidy instead, it retains €29. The same offer wording now fails the €30 floor. If checkout already grants free standard delivery, presenting that same service as a new incentive gives the shopper no additional benefit. A paid upgrade is a separate proposal with its own economics.
Check the threshold against the subtotal the promotion engine actually uses. A customer may start above a free-delivery minimum and fall below it after a merchandise discount or after removing an item. Do not infer whether the engine tests the subtotal before or after discounts. Test that behavior in checkout. Also verify what happens to delivery if the cart splits into shipments or contains an item outside the qualifying service. A benefit on one shipment should not be described as free delivery for the entire order.
Give the shopper a specific status once the calculation is available. An illustrative checkout message might read: Standard delivery is free on this cart. Express delivery is not included. If a cart change makes the offer unavailable, state what changed if checkout supplies the reason. Offer an eligible pickup option or a different delivery method only if it is actually available. The most useful internal check is to reconcile the waived charge, any additional fulfillment expense and the contribution shown after checkout recalculates.
4. Apply percent discounts only to eligible revenue
A percentage tells you how to calculate a saving, not which items get it. Before the chatbot quotes an amount, the promotion engine needs to determine the eligible lines, the relevant prices and any cap or exclusion. Use the engine’s resulting discount to check contribution. Multiplying the headline rate by the whole cart can produce the wrong customer promise and the wrong margin decision.
For a fictional partially eligible cart, suppose the €90 total is made up of a €60 regular-price item and a €30 sale item, with €53 in total variable costs. If the engine applies 8% only to the regular-price item, the reduction is €4.80. Net revenue becomes €85.20 and contribution becomes €32.20. That clears the €30 floor. An 8% discount on every line would leave €29.80 and fail. A clear customer message is 8% off the eligible item, saving €4.80, not 8% off your order.
Recheck the cart when the shopper removes a line. The remaining item might still be eligible for a discount while no longer producing enough order contribution, or removing an excluded item might cause a minimum-spend requirement to fail. Rerun both checks. Use the promotion engine’s returned cents when line allocation and rounding matter; the chatbot’s estimate is not the amount that checkout must honor.
For the original €90 single-item cart, the €30 floor leaves at most €7 of headroom if variable costs stay fixed. That is an internal limit, not permission to invent a smaller coupon or turn an 8% offer into a different rate. Ask the engine about configured alternatives and compare their applied totals. If an approved alternative is available, explain the relevant difference, such as a smaller saving that applies to more items. If none passes, keep the original cart intact and tell the shopper that the requested discount cannot be applied to it.
5. Separate the value of the accessory from the bundle discount
A bundle changes what the shopper buys, so compare it with both the original cart and the same expanded cart without the bundle discount. The first comparison shows whether the upsell leaves more contribution than the original order. The second shows what the bundle promotion costs. Looking at only one can make a profitable accessory look like a cheap giveaway, or make a discounted bundle seem better than the original sale when it isn’t.
Take a fictional example of a €90 desk lamp and a compatible €15 accessory that costs €7 to provide. At their stated prices, the two items contribute €45 together, compared with €37 for the lamp alone. A €10 bundle discount leaves €35. It meets the €30 floor, but the store would be making €2 less contribution than on the lamp alone. At these fixed costs, a bundle discount of €8 would leave €37, equal to the lamp baseline. Decide whether your rule is simply to clear the floor or also to preserve the original order’s contribution.
Confirm that the €7 is the full incremental variable cost under your model. If the accessory needs separate packing, changes the delivery charge or adds a per-item fulfillment fee, the €45 figure is no longer the right starting point. Also test what happens if either component is removed. Let the promotion engine decide whether to remove the bundle price, reprice the remaining line or leave another eligible promotion in place. Don’t describe an accessory as free if checkout requires a higher total to obtain it.
Measure the bundle as its own offer. Track whether shoppers accept the added item and whether it appears on the paid order before interpreting the margin. A shopper who would have purchased both items at their stated prices gives up no extra revenue for the store until a discount is applied; in this example, the promotion costs €10 on that two-item order. That’s a different question from whether the bundle increased accessory sales.
6. Let the promotion engine decide what actually applies
Keep three separate states in the conversation: selection, accepted cart action and payment. A shopper might say Use the delivery offer in a voice interaction or make the same choice in text. That selects a candidate. It doesn’t mean the promotion engine accepted the request, the cart total changed or payment went through. Each confirmation should match the state the store can verify.
In a proposed workflow, send the requested offer and current cart to the promotion engine, then evaluate the resulting cart against the floor before confirming the incentive. If the floor is checked outside the engine, use that check to block an unprofitable offer, never to overrule an engine rejection or change the engine’s calculation. Where a preview is possible, check the preview before committing the change and verify the accepted cart afterward. A mock preview is illustrative; it isn’t proof that an offer is available or that a cart action succeeded.
Build the shopper response from the accepted cart. For example, Free standard delivery has been applied to this cart should follow confirmation that the charge was waived for that method. If the engine reports that an existing promotion cannot be combined with the requested one, keep the existing cart intact unless the shopper agrees to a replacement. Show the actual difference before asking them to decide. An offer with an attractive headline can still raise the total if it displaces a better one.
Revalidate after a quantity, item or delivery change, and check again at checkout if the offer can expire between selection and payment. If the totals change, explain the new outcome before the shopper pays. If the engine does not return a result, report that the offer is still unconfirmed rather than implying a saving. Treat a cart update as a cart update; only the store’s payment or order confirmation should support a statement that the purchase is complete.
7. Test the carts most likely to break the margin rule
Build a set of test carts with expected results in cents. For each, record the starting subtotal and contribution, requested offer, eligible lines, applied amount, final contribution and expected checkout message. Include a cart at exactly €30 and one at €29.99 under the fictional floor. The first should pass and the second should not, even if a draft conversation rounds both to about €30.
- Sale item: Start with an item already marked down. Verify that the baseline uses the sale price and that the engine determines whether the new offer stacks. If it rejects the offer, the shopper’s original sale price should remain intact.
- Partially eligible basket: Use the fictional €60 eligible line and €30 excluded line. Expect a €4.80 reduction, not €7.20. Remove an item and verify that eligibility, any minimum-spend threshold and the contribution floor are all recalculated.
- Offer about to expire: Select the incentive while it is valid, then try to apply it or move to checkout after expiry. Compare the response with the current checkout outcome, not the earlier quote. Check that any proposed replacement is evaluated independently.
Add a stacking test because discounts that pass independently can fail together. In the partially eligible example, €4.80 off leaves €32.20 in contribution. A €6 delivery subsidy alone leaves €31. If the engine allows both at those amounts and there are no other changes, the combined contribution is €26.20. Neither offer’s individual approval is enough to authorize the combination.
Finish each test by reading the shopper response alongside the accepted cart. Does the displayed saving match to the cent? Does the message distinguish an expired offer from one that was never eligible, when checkout identifies the reason? If no alternative passes, a plain explanation and the unchanged cart are better than an unverified promise. Repeat these checks whenever prices, variable costs or promotion rules change.
8. Measure accepted offers separately from paid orders
Track the sequence rather than giving the chatbot credit every time it mentions a promotion. Distinguish an offer shown, a shopper selection, a request sent to the promotion engine, an accepted or rejected cart action, and a paid order. A selection rate tells you which offer shoppers want to try. An acceptance rate tells you how often the request works on the cart submitted. Neither tells you whether the order was paid.
Use Google Analytics ecommerce events to observe relevant shopping steps, such as adding the accessory to the cart, beginning checkout and recording a purchase. Keep the internal decision record as well: the offer type, the engine outcome, the applied amount, the contribution calculation and the cart or order reference needed to reconcile them. An ecommerce event can show what was reported for a purchase; it should not stand in for the promotion decision or your cost calculation.
Compare the offers with more than a redemption count. Look at paid orders relative to the carts offered each incentive, contribution per paid order and contribution per cart offered. The last measure makes it harder to favor an offer that looks strong only among shoppers who complete checkout. Keep the comparison groups comparable: a bundle shown to shoppers who already have an accessory in mind is not directly comparable with a delivery offer shown only after a shipping charge appears. Without a suitable comparison, report an observed difference rather than claiming the incentive caused it.
Investigate the failures, too. Separate engine rejections from floor failures, and look for shopper-facing confirmations that do not match the next checkout total. For a bundle, check whether the accessory survives into the final paid order. For free delivery, reconcile the actual subsidy rather than the headline benefit. For a discount, compare the quoted amount with the applied amount. Those checks reveal which offer needs a rule change and which simply needs a clearer message.
9. Roll out a short offer list and audit the exceptions
Start with a short list of offers the store can cost and verify: one defined delivery offer, one configured percentage promotion and one clearly specified bundle. For each, document the products or services it covers, the variable costs required to evaluate it, the contribution floor and the fallback when it cannot be applied. The chatbot should draw from approved offers rather than negotiate new amounts. The promotion engine remains the final source for the applied amount, stacking and resulting subtotal.
Use a simple review routine before expanding the list. Reconcile a sample of accepted carts and paid orders against the engine outcome and the cost assumptions. Include orders close to the floor: in the fictional comparison, free delivery has only €1 of headroom. If the net subsidy rises from €6 to €7.01 and nothing else changes, contribution falls from €31 to €29.99. A small change in cost can make an apparently safe offer fail even though the promotion wording is identical.
Assign someone to update the inputs when a supplier cost, fulfillment charge, delivery rate or payment fee changes. Decide which offers should be paused if those inputs go missing. Review refusals as well as approvals. Frequent rejections may point to a chatbot suggesting an offer too early, stale eligibility information or a conflict with a promotion already on the cart. The fix isn’t necessarily a lower floor; it may be better timing or a more suitable offer.
When choosing what to show first, start with what the shopper is trying to resolve. If the issue is delivery cost, check the delivery offer before proposing an accessory. If the customer asks for a compatible add-on, show the bundle total rather than treating it as a substitute coupon. Keep the decision trace simple: what the shopper chose, what the engine applied, what contribution remained and whether the order was paid. That is enough to explain a decision and check it later without confusing a promised benefit with an actual sale.
Related guides
- Where in a shopping journey a chat can resolve cart hesitation
- Track a WooCommerce conversation from recommendation to purchase
