User Guide · Deliver

Choose the data record behind the message.

Each POI list or data feed used by the project becomes a subject in Which record. Define eligible candidates, rank or rotate them, then verify the winner with realistic request inputs.

DecisioningPOI · feeds · computed fields

How subjects appear

Delivery detects structured tokens in all project creatives. A POI token can create a panel named for that list; a sports token can create a panel for its selected league/feed. Configure every detected subject used by an enabled creative.

Build the candidate rule

Candidate setLoad the relevant list or feed scope.
FilterKeep records that meet required conditions.
ExcludeRemove explicitly unsuitable records.
ComputeDerive reusable fields such as distance.
ScoreRank candidates when the mode uses score.
PickChoose one record according to mode.

Pick modes

Mode	Behavior	Score
Best	Chooses the eligible record with the largest score.	Required for meaningful ranking.
First eligible	Uses the first record in deterministic candidate order.	Ignored/hidden.
Rotate all eligible	Cycles through the full eligible set.	Ignored/hidden.
Random top-N	Randomly chooses from the highest-scoring N candidates.	Used to build the top set.

Nearest POI with a recipe

Apply the Nearest recipe to use listener and POI coordinates. The scoring strategy is equivalent to:

Score expression

-distance_km(user.lat, user.lon, field.lat, field.lon)

The nearest location has the smallest distance; negating distance makes it the largest score, which fits Best.

Confirm that every POI has valid coordinates.
Apply Nearest.
Select Best.
Use Pick on Map to set a simulated listener location.
Evaluate and inspect the winning candidate.
Repeat near boundaries between POIs.

Configure with AI

Describe the record strategy in business language. The agent can propose filters, score, computed fields, and pick settings. For example, it can create a deterministic nearest rule and a human-readable km_away computed field.

Apply to rule only after review.Read every proposed expression, field, and default. Simulate the resulting rule; the natural-language instruction is not what Delivery executes.

Pick on Map

Use Pick on Map to populate request latitude and longitude without manually typing coordinates. After evaluation, inspect:

the number of source candidates;
which records passed filters;
scores and computed values;
the record or top-N set marked as selected;
the fields that will be inserted into the creative.

Random top-N

Set N only after scoring is stable. A larger N increases variety but can include a materially weaker candidate. Simulate enough requests to confirm all selected candidates are acceptable.

Rules for sports and other feeds

Use the same candidate pipeline, but filter on business time and status before scoring. A closest or highest score is irrelevant when an event is cancelled, outside campaign scope, or stale.

No eligible record

Design the creative so it remains valid when no record wins. Do not leave a sentence that requires an unavailable store, event, or value. Use a default creative or approved generic wording.

Test failure semantics.A zero-candidate result is different from a feed service error. Operators should investigate unexpected empty panels or missing history instead of assuming the campaign has no data.

Rule QA checklist

Only campaign-eligible records enter the final set.
Field types match each expression.
Best, First, Rotate, or Random top-N matches the business goal.
Map and boundary simulations choose acceptable records.
Ties and missing coordinates have deterministic behavior.
No-match copy remains complete and safe.