Models and Allocation
AI 1-to-1 Delivery uses predictive and prescriptive models to decide which Customers should receive which coupon. The Owner App does not expose formulas; it exposes the business inputs, generated campaigns, audiences, forecast, budget impact, and approval actions.
Use Cases
| Use Case |
|---|
| -> Rank existing category buyers and assign different coupon strengths by purchase behavior. |
| -> Select Customers with high probability of buying a category they have not purchased recently. |
| -> Size the generated audience from a budget or contribution before approving execution. |
How It Works
| Model / signal | Used for | What it predicts or ranks | How it affects delivery |
|---|---|---|---|
| Purchase Behavior Score (PBS) | Up-selling | Customer strength or affinity in a product/category group they already buy | Splits Customers into behavior tiers so incentives can be adjusted by propensity |
| Product recommender / cross-selling model | Cross-selling | Probability that a Customer will buy a category they have not bought recently | Selects the highest-propensity Customers for each non-purchase period |
| Coupon ranking | Coupon assignment | Relative strength of coupons in the same product/category group | Orders coupons so the platform can assign weaker or stronger incentives according to the goal |
| Coupon affinity and distribution rules | Prescriptive assignment | Which coupon should be assigned to which Customer segment | Matches coupon strength to Customer affinity and business rules |
| Forecast estimates | Forecast and budget sizing | Expected delivery, redemption, cost, incremental sales, and ROI | Lets the team review business impact before approving execution |
Coupon Distribution Logic
Coupons are grouped by product category or product attribute. Each group becomes an AI campaign, and the selected business goals determine how actions are created inside that campaign.
| Distribution dimension | Up-selling | Cross-selling |
|---|---|---|
| Customer pool | Customers who already buy the category | Customers who have not bought the category recently or never bought it |
| Model | PBS | Product recommender / cross-selling model |
| Coupons used | Up to 5 per category group | One best coupon per category group |
| Split logic | PBS tiers | Non-purchase recency periods |
| Customer selection | Percentile distribution by purchase behavior | Highest-probability Customers above the configured threshold |
| Restricted Audience | Optional plan-level restriction applies | Optional plan-level restriction applies |
Budget-Driven Allocation
AI 1-to-1 Delivery can work from a budget or contribution assigned to the Coupons Bank. Instead of starting with a fixed audience and calculating cost afterwards, the allocation works backwards: the budget helps decide how many Customers can receive coupons and how those Customers are distributed across goals, categories, and coupons.
At a high level, budget allocation follows this cascade:
- Business goal split — if the plan has more than one goal, budget is split between goals, such as up-selling and cross-selling.
- Affectation split — within a goal, budget is distributed across product/category affectations based on their business weight, such as sales weight in the selected period.
- Coupon split — within each affectation, Customers are distributed across coupons using the configured distribution percentages and coupon priority.
- Audience sizing — the platform estimates how many impacted Customers fit in the allocated budget, using expected delivery, redemption, redemption frequency, and average coupon cost.
The percentages configured for coupon distribution are Customer distribution percentages, not budget percentages. The platform keeps the intended Customer mix and uses the estimated average cost of that mix to size the audience.
Allocation Rules
| Rule | Behavior |
|---|---|
| Best-affinity Customers first | Customers are ranked by model affinity. If budget is limited, lower-affinity Customers are left out first. |
| Lower incentive for stronger affinity | In up-selling, the weakest coupon is assigned to the highest-affinity Customers, and stronger coupons are reserved for Customers who need more incentive. |
| Budget left over | If there are fewer eligible Customers than the budget can cover, all eligible Customers can be included and the remaining budget is shown as unused. |
| Control group | Control-group Customers are excluded from delivery and do not consume coupon budget. |
Budget allocation is an estimate. In the current model, Customers assigned to the same coupon are treated with an average expected cost for that coupon. Real cost can vary because Customers do not all redeem with the same probability or spend the same amount.
Configuration
Model thresholds and distribution defaults are configured at company level and can be overridden in the AI Plan advanced configuration when the UI exposes those controls.