Switchback tests

A switchback test is how BudLogix measures a price move it is not yet sure of. Where a size group's permission reads test first (see How the pricing engine decides), the new prices go out on alternate weeks instead of all at once, and the weeks are compared.

How a test runs

• A whole size group, one direction. A test takes the size group's Test-first raises, or its Test-first cuts, or its best sellers at a 10% move. The whole group moves together. Holding back half of the same shelf does not work: buyers switch from the moved half to the held half, and the move reads twice as strong as it is. Raises and cuts are tested separately for the same reason.
• Started from Approve and push. Confirming a push on Approve and push schedules the size group's test from the next Monday. The Test-first rows are the grid's Decision column, so a row you set to Test first joins and one you set to Approve or Skip does not.
• Four weeks, Monday to Sunday. Week 1 at the new prices, week 2 back at today's, week 3 new, week 4 today's.
• Nothing is pushed for you. The day before each week starts, that week's prices appear on Price events as a draft named "Switchback: Flower 3.5g raises, week 2 of 4 (today's prices)". You review and push it like any other event. It has no held-back group: the alternate weeks are the comparison. A week's draft you did not push is deleted when the next week's is drafted, so a late push cannot undo the next week.
• The nearby market as a control. Each day of the test BudLogix reads what the stores near you charge for the same size. If the market cut its prices during one of your "new" weeks, the read counts your move against the market's, not against nothing.
• One test per size group at a time. A second test on the same size group in the same weeks would read each other's buyers.

How a test is read

After the fourth week closes, each product's two "new" weeks are compared with its two "today" weeks, adjusted for how busy the rest of the store was each week. A product's week is left out when it was off the shelf, or mostly sold on a promotion, for more than two days of it. A product with no counted week on one side reads nothing, and neither does one whose price never actually moved (a week's draft that was never pushed).

The test also reads how many of the tested products' lost (or won) buyers stayed in the store: what the rest of the size group, and the sizes beside it, picked up or gave up. A Pre-roll 1g test watches infused pre-rolls, multipacks and flower too.

A test where fewer than half its products, or fewer than three, could be read was interrupted, usually by a stock-out, and nothing it read is used.

What a read test changes

• The share kept in the store replaces the demand model's low-confidence figure for that size group (the 64% flower borrowed until a test measured it). Raises and cuts share one figure until both have been tested.
• Each tested product's price sensitivity joins the reading from its own price history, weighted by how sure each is. A product whose price had never moved before has a reading of its own for the first time.
• The next Sunday's refit carries both into the size group's permission and the profit-best prices, so a size group that was test first can become allowed or refused on what its own shelf did.