TikTok Shop inventory planning gets difficult when demand, cancellations, refunds, creator videos and supplier lead times move at different speeds. A reorder point copied from a generic spreadsheet can look precise while using the wrong demand window.
This model turns paid orders into mature net demand, adds supplier variability, and keeps creator campaigns separate from the baseline. It is designed for day-level planning where decisions need to be reproducible.
What a reorder point needs to answer
The reorder point is the inventory level that should trigger a purchase order before the next delivery can arrive.
Use this working definition:
Reorder point =
expected net demand during lead time and review period
+ safety stock
Do not use gross GMV or order count as the demand input. Use units that still represent real demand after cancellations and refunds have had time to mature.
Define mature net demand
A TikTok Shop order can change state after it is created. Cancellations and refunds reduce the units that should drive replenishment. Late refunds can make last week look stronger than it was.
For a cohort of orders, calculate:
Mature net units =
paid units
- canceled units
- refunded units
Use a maturity window that matches the SKU's behavior. A practical starting point is 21 days for ordinary delivery and refund timing, then compare that policy with a 30-day window for slower markets.
The maturity window is a planning assumption. It is not a claim that no refund can happen after day 21.
Avoid counting returns twice
Refunded units can become sellable inventory after inspection. Do not subtract the refund from demand and then add the same physical return back to stock again.
Choose one of these treatments and document it:
- Use net demand after refunds, then add only confirmed or scheduled sellable returns that are not already in on-hand inventory.
- Use gross shipped demand, then model all returns as future supply. This requires reliable return timing.
The first approach is usually easier for weekly planning.
Calculate available-to-promise inventory
The reorder trigger should use inventory the team can actually allocate:
Available to promise =
sellable on hand
+ confirmed inbound units
- reserved or unfulfilled units
+ scheduled sellable returns not already in on hand
Leave supplier promises out until they have a confirmed ship date. A purchase order without a confirmed date is a plan, not inbound inventory.
Reorder when:
available to promise <= reorder point
Add supplier variability
The protection period is:
Protection period = supplier lead time + review period
If the team reviews stock every seven days and the supplier takes ten days, the protection period is seventeen days.
One reproducible safety-stock formula is:
Safety stock =
z
× sqrt(
protection period × daily demand variance
+ average daily demand squared × lead-time variance
)
For a 95% service-level target, use z = 1.65. Treat that result as an operating buffer, not a guarantee against stockouts.
Worked example
Assume a SKU has these inputs:
- Mature net demand: 40 units per day
- Daily demand standard deviation: 10 units
- Supplier lead time: 10 days
- Review period: 7 days
- Supplier lead time is treated as stable for this first calculation
- Service-level factor: 1.65
- Sellable on hand: 500 units
- Confirmed inbound: 150 units
- Reserved orders: 50 units
Protection-period demand:
40 × (10 + 7) = 680 units
Safety stock:
1.65 × sqrt(17 × 10²) ≈ 68 units
Reorder point:
680 + 68 = 748 units
Available to promise:
500 + 150 - 50 = 600 units
The SKU is below its reorder point. The replenishment gap before MOQ and campaign adjustments is:
748 - 600 = 148 units
The buyer should then add expected demand during the next review cycle, apply supplier MOQ, and check whether cash and storage can support the order.
Keep creator spikes separate
A creator video can produce three strong days without changing next quarter's baseline. Fold that event permanently into daily demand and the model over-orders for months.
Use two demand layers:
- Baseline: matured, non-campaign net demand
- Event demand: expected units from a specific creator, livestream or paid campaign
Calculate the campaign stock requirement as:
Incremental campaign units =
expected campaign units
- already committed campaign inventory
Add that requirement only for the event window. If the campaign becomes a repeatable sales channel, recalculate the baseline after several completed events.
SQL starting point
This query assumes one row per SKU and day after orders have been reduced to paid, canceled and refunded units.
WITH daily AS (
SELECT
sku_id,
order_date,
paid_units - canceled_units - refunded_units AS net_units
FROM sku_daily_order_events
WHERE order_date >= CURRENT_DATE - INTERVAL '90 days'
AND order_date < CURRENT_DATE - INTERVAL '21 days'
),
params AS (
SELECT
10::numeric AS lead_time_days,
7::numeric AS review_period_days,
1.65::numeric AS z_score
),
velocity AS (
SELECT
d.sku_id,
AVG(d.net_units) AS avg_daily_units,
STDDEV_SAMP(d.net_units) AS daily_sd
FROM daily d
GROUP BY d.sku_id
),
model AS (
SELECT
v.sku_id,
v.avg_daily_units,
v.daily_sd,
p.lead_time_days,
p.review_period_days,
v.avg_daily_units * (p.lead_time_days + p.review_period_days)
AS protection_demand,
p.z_score * SQRT(
(p.lead_time_days + p.review_period_days) * POWER(v.daily_sd, 2)
) AS safety_stock
FROM velocity v
CROSS JOIN params p
)
SELECT
m.sku_id,
ROUND(m.protection_demand + m.safety_stock, 0) AS reorder_point,
ROUND(m.protection_demand, 0) AS protection_demand,
ROUND(m.safety_stock, 0) AS safety_stock,
i.on_hand,
i.confirmed_inbound,
i.reserved_units,
i.on_hand + i.confirmed_inbound - i.reserved_units AS available_to_promise
FROM model m
JOIN latest_inventory_snapshots i USING (sku_id)
ORDER BY reorder_point DESC;
The query excludes immature cohorts. A later reconciliation can add refunds that arrive after the 21-day cutoff.
Failure modes to test before trusting the model
Cancellations recorded after shipment
Inventory may have left the warehouse even when the order is canceled. Reconcile the status event with physical stock movement.
Refunds without returns
A refund does not always mean a unit returns to sellable stock. Separate financial refunds from physical return receipts.
Bundle and component demand
One bundle order can consume several component SKUs. Convert bundle demand into component demand before calculating component reorder points.
Multi-warehouse inventory
An item may be available nationally but not in the market that needs it. Calculate reorder points by the inventory pool that can serve the SKU's real orders.
Supplier MOQ
The model produces a gap, not a purchase order. Supplier MOQ, cash flow, storage, shelf life and freight timing can change the final quantity.
Campaign overlap
Two creators releasing videos in the same week should be modeled together when their audiences overlap. Separate event plans can double count the same demand.
A weekly operating checklist
- Freeze the demand window and maturity cutoff.
- Remove canceled and matured refunded units from demand.
- Combine physical on-hand, confirmed inbound and reserved inventory.
- Recalculate reorder points by SKU and service-level tier.
- Add confirmed campaign demand separately.
- Apply MOQ, cash and storage constraints.
- Record the assumptions used for every purchase decision.
- Reconcile late refunds and late returns against the previous forecast.
The useful output is a decision that another operator can reproduce next week with the same inputs.
Related resources
For product selection inputs, see the TikTok Shop product research guide.
For the gross-margin and net-margin inputs that affect service-level decisions, use the free ecommerce profit margin calculator.
For fulfillment choices that change lead-time variance, read the TikTok Shop logistics guide.
I work on Starway Global, a TikTok Shop analytics and operations platform. This article is educational and does not guarantee inventory, revenue or ranking outcomes.
AI disclosure: Fully Autonomous.
Top comments (0)