Channel cost-aware decisioning

Many stacks fire every channel by default for any outcome the marketing team wants. A customer who has not opened the last three emails gets the email, the push, the SMS, and the retargeting ad against the same offer, because each channel ran independently and nobody asked which channel was most likely to convert them at the cheapest cost. The owned channels (email, push, in-app) are nearly free per send; the paid channels (display, social, search) cost real money per impression. Firing all of them in parallel for an outcome the cheap channels could achieve is exactly the waste the recipe is named after.

The cost-aware response is a decisioning layer that picks the cheapest channel likely to achieve the outcome, fires it, and reserves the more expensive channels for the moments the cheap ones either could not reach or did not work. The owned channel runs first; the paid channel fires only if the owned channel either had no eligibility (the customer is not subscribed, the push token is dead) or did not produce engagement within the window the recipe is willing to wait. The recipe is medium-high readiness because the cost model needs to be honest, the priority rules need to reflect actual conversion likelihood per channel, and the per-customer eligibility join is real engineering.

This recipe builds the cost-aware channel selection as the default, with the expensive channels as the fallback rather than the parallel.

A measurable reduction in cost per outcome through smarter channel selection, with the savings showing up at the per-outcome cost reconciliation. The metric is cost per outcome by channel mix: the cost to drive a unit of the target outcome, against the baseline of the all-channels-fire approach. The value shows up where the owned channels carry the load they should (the customer who is reachable on email and likely to convert there receives only the email), and where the paid channels are reserved for the customers the owned channels could not reach or did not work for. The cost savings concentrate where the audience is largely reachable on owned channels and thin where the audience is paid-channel-dependent.

The second outcome is the operational discipline of treating channels as a portfolio rather than as parallel firehoses. The team starts thinking about the channel mix per outcome rather than running every campaign in every channel, which is the cultural shift the recipe enables.

A decisioning service that takes the customer, the outcome, the cost table, and the eligibility status, returns the channel choice and the fallback channels, and integrates with the channel orchestration layer to fire the chosen channel. Composable stacks suit this because the decisioning lives in the activation layer and reads the warehouse for the customer's eligibility; packaged stacks may have channel-by-channel orchestration with no shared decisioning, and the build is the shared decisioning layer rather than the per-channel work. The capability that decides whether the recipe is durable is the cost table's accuracy: a table that under-counts the true cost of a channel (forgetting the platform's audience-match fee, the creative production cost, the integration overhead) produces channel mixes that look efficient at the surface and waste at the actual cost.

Work in this order. Eligibility belongs inside the decision, not after it.

1. Build an honest cost table. Per-send and per-impression costs plus creative production, integration and audience-match fees, with a maintenance cadence. A table that under-counts paid drifts the mix toward paid.
2. Define the outcome target and its latency requirement. How fast the customer has to be reached decides which channel tiers qualify.
3. Put the customer's channel preference in as a primary input. Preference and cost combine; cost never overrides a stated preference.
4. Check eligibility inside the decisioning, before a channel is chosen. A dead push token rules push out of the choice, so the fallback runs instead of a send that fails with nobody noticing.
5. Pick the cheapest eligible channel that meets the latency.
6. Route through a deterministic fallback when the model declines: a default channel for the outcome, or the next-cheapest in the priority order.

Latency-tiered channel selection covers steps 1, 2 and 5, channel eligibility check steps 3 and 4, and real-time decisioning with deterministic fallback step 6.