Distribution

Shopee
Lean the daily split toward campaigns doing well right now — on clicks, sales, ROAS or profit — instead of the forecast model alone.

By default a group splits the budget from its learned forecast model alone. Distribution can lean that split toward campaigns doing well right now — useful when the model is still young, or when the market moved faster than the model did.

It lives in the Strategy box on Global settings.

How to use it

  1. Pick Model only (default) or Follow recent results.
  2. If following, finish the sentence: Shift budget toward campaigns with the best signal in the window. The signal starts on your objective's own metric — Profit, GMV, or Direct orders — marked Suggested in the menu.
  3. Pick how far to lean toward recent results: Small, Medium (default) or Large.
  4. Save.

Signals

The menu groups signals by how fast their result is known.

GroupSignalsWindows
ActivityImpressions, Clicks, Unpaid ordersLast 3 hours, Last 24 hours, Last 7 days
SalesGMV, Units sold, Direct orders, Direct GMVLast 24 hours or Last 7 days
Profit & ROASPlatform ROAS, Profit, PA-ROASLast 24 hours or Last 7 days

Sales and profit results take time to come in, so those signals read 24 hours or 7 days only. Last 24 hours is the default window.

  • Direct orders and GMV count the advertised product itself, not halo sales.
  • Unpaid orders is a leading signal: it can change when buyers pay.
  • PA-ROAS is True ROAS.

How far to lean

Each step caps how much of the model's expected gain you give up to follow recent results. The gain is what the model's own split was expected to add over today's.

LeanGives up at most
Small10% of the expected gain
Medium (default)30%
Large60%

If the model expects its split to add ฿1,000 of profit, Large still has to add at least ฿400. Leaning changes which campaign gets the money, not how much the group spends.

Activation is blocked while a group holds a preview-only window, or a basis and window that cannot run together. The row says so — "Needs a live distribution basis" — rather than hiding the control.

Reading the effect

In the Allocation panel, hover a campaign. The tooltip separates:

  • Model-only — what the objective and constraint alone would have given it.
  • Preference wants — what the recent-results signal alone would have given it.
  • Proposed — the number the solve actually landed on, between the two.

If those first two are far apart, your lean is doing real work. If they are nearly identical, the signal agrees with the model and you can leave it on Model only.