Name the friction before selecting the benefit
A customer returns to a product page and leaves again. The team can offer guidance, recommend a compatible item, provide a delivery benefit, introduce a discount or wait. Those actions solve different problems. Browsing alone does not tell us that price is the obstacle.
Use evidence in proportion to what it reveals. A direct question about delivery cost supports a delivery-cost hypothesis. Repeated compatibility questions suggest an information need. Treat inferred motivations as hypotheses, and avoid turning a customer score into a confident story about the person.
Keep value, purchase probability and responsiveness separate
Recency, purchase frequency, observed spend, category preference and service history help describe a relationship. They can inform candidate actions. They do not establish what will happen because of a discount. A customer likely to buy may purchase without an incentive; someone unlikely to buy may remain uninterested after receiving one.
The economic question concerns the difference between the proposed policy and a relevant alternative. That difference needs an evaluation design. Use predicted scores as inputs to a policy, and test the policy’s effect rather than describing propensity itself as incremental lift.
Build a small menu of meaningful alternatives
Consider product guidance, a relevant recommendation, early access, a shipping benefit and a price incentive. For each, specify the need it addresses, who qualifies, its validity period and its cost. Waiting should remain available when there is no useful reason to contact the customer.
Count operational costs as well as the visible offer. Free delivery can create a subsidy, early access can consume limited stock, and a recommendation can increase returns if compatibility is poor. Maintain a consistent treatment of product cost, fulfilment, payment fees, refunds and benefit funding.
Work through the margin arithmetic
The following example is fictional and uses generic currency units. Each policy has 1,000 randomly assigned eligible customers. All receive the same message and timing; only the benefit differs. The comparison therefore evaluates the benefit policy among contacted customers, not the effect of sending a message at all.
Assume a price of 100 and variable product and ordinary fulfilment costs of 60 per order. Without a benefit, 100 orders contribute 4,000. A 20% discount with 130 orders contributes 2,600, because each order contributes only 20. A delivery subsidy of 5 with 115 orders contributes 4,025, because each order contributes 35.
The discount policy produces more orders but 1,400 less contribution in this example. The delivery policy’s 25-unit advantage is small and could be outweighed by uncertainty or omitted operating costs. These are arithmetic illustrations, not campaign forecasts or evidence that one benefit always wins.
| Benefit policy | Orders | Contribution per order | Total contribution |
|---|---|---|---|
| No benefit | 100 | 40 | 4,000 |
| 20% discount | 130 | 20 | 2,600 |
| Delivery subsidy of 5 | 115 | 35 | 4,025 |
Apply eligibility before allocating the offer
Define the qualifying customer state, channel permission, recent contact limits and product availability. Exclude already-completed purchases where the offer is no longer relevant. Decide how open service issues affect eligibility and who can override a rule under a documented policy.
Make funding and limits explicit: maximum benefit cost, redemption conditions, expiry and any order-value threshold. A coupon should not promise a product that is unavailable. Verify the actual price and benefit applied at checkout; a marketing rule and a commerce rule can drift apart.
Compare outcomes per assigned customer
Use a consistent observation window and keep all assigned customers in the primary analysis, including those who never opened the message. Comparing only redeemers selects people after treatment and can make an ineffective offer look attractive. Review net revenue, contribution, returns and unwanted-contact indicators alongside purchase rate.
Choose the minimum commercially meaningful improvement and plan the required sample before launch. Preserve a treatment and a relevant control, and report uncertainty. If several benefits are compared, account for the extra comparisons. A small observed lead is a reason to investigate, not an automatic instruction to scale.
Turn the result into an operating policy
Record which customer states qualified, what benefit was offered, the decision version, actual redemption cost and the resulting order. In Binoban, customer attributes and audiences can provide the selection context, while the configured journey and commerce integration determine execution. Confirm available controls before promising automated optimization.
The owner should be able to answer two questions: why was this benefit useful for this customer, and what additional contribution did the policy create? Keeping both questions visible makes incentive design a repeatable commercial decision.
Take the next step