Table one: receiving and purchase evidence

Capture supplier, item description, package, lot or batch when available, invoice, receiving date, location, quantity, and source document. The SI Receipt duplicate-record control helps prevent duplicated evidence from inflating the apparent quantity received.

Table two: ingredient-to-menu mappings

Map the purchased ingredient to recipes, modifiers, combos, dayparts, service locations, and temporary substitutions. Include effective dates. A current recipe is not necessarily the recipe used during the exposure window.

Keep system and endpoint ownership visible through ServingIntel hardware planning, especially when records are split across local stations or offline workflows.

Table three: order and service history

Use order timestamp, location, channel, item, modifier, quantity, void, refund, and fulfillment state. Then join only where keys and effective dates support the relationship. The 86 The POS ingredient-hold workflow explains how to preserve the menu state while the analysis runs.

Return ranges, not false certainty

  • Confirmed: records directly match the affected identifier and window.
  • Possible: menu and timing match, but lot evidence is incomplete.
  • Ruled out: reliable evidence excludes the affected identifier or period.
  • Unknown: a missing key or record prevents classification.

Record assumptions, source timestamps, query version, reviewer, and unresolved gaps. Track current operating context through ServingIntel News & Insights.

Prebuild the test and rehearse it quarterly

  1. Choose a fictional ingredient, lot, location, and date window.
  2. Run the three-table join and time the first usable result.
  3. Review unmatched records and duplicated keys.
  4. Assign owners to close the highest-impact gaps.
  5. Rerun after corrections and retain the evidence.

Escalate missing integrations or inaccessible records through ServingIntel support resources.

The bottom line: a useful recall query bounds exposure with traceable evidence and clearly labels what remains unknown; it does not turn incomplete data into a confident claim.