> ## Documentation Index > Fetch the complete documentation index at: https://firespark.cloud/docs/llms.txt > Use this file to discover all available pages before exploring further. # Menus A/B testing > Test what customers see per channel and measure impact before you roll out broadly. Menu performance is not only *what* you sell — it is **how** it is named, pictured, ordered, and offered on each surface. A hero burger might convert on your app but need a shorter title on an aggregator; a combo might belong above the fold on kiosk only. ## Why testing matters commercially | Without testing | With testing | | ------------------------------------------- | ------------------------------------------------------- | | Gut feel on naming and layout | Evidence per channel before national rollout | | Same menu everywhere “because it is easier” | Composed menus that still sync from the POS | | Fear of breaking kitchen logic | Tests on presentation and placement, not duplicate SKUs | Fire spark’s intelligence layer sits **on top of synced operational data**. You are not forking the catalog in Excel; you are experimenting on how the same items show up where customers actually order. ## What you can test * **Titles and descriptions** — Clarity vs appetite appeal per channel. * **Images and placement** — Which category order lifts attachment rate on web vs app. * **Assortment** — Subset on Uber Eats, full menu on owned channels ([menu composition](/docs/concepts/menus)). * **Modifiers and groups** — Drink choice framing, upsell order, default selections. * **Schedule** — Breakfast block vs all-day on selected surfaces. Tests respect **store, channel, fulfillment, and hours** — the same dimensions you already use to compose menus. ## How teams work in practice 1. **Hypothesis** — “Kiosk converts better when combos are the first category.” 2. **Compose** — Apply variant on kiosk channel only; POS master unchanged. 3. **Measure** — Compare mix and ticket against control via [consolidated analytics](/docs/case/consolidated-analytics). 4. **Scale or revert** — Roll winning layout to more stores; no re-integration. ## What you stop worrying about Kitchen tickets still map to real products and modifiers. When you 86 an item in the POS, it disappears from test and control alike. Operations does not learn a **second menu language** for experiments. Composition by channel, store, and schedule Modifier groups and catalog items Compare results across channels