Push vs Pull Distribution Strategies: When Each Fails

In October the planning team pushed six weeks of room heaters to the northern depots on the strength of last winter's sales. By mid-December one depot had sold out while the depot two states away still held 40 percent of its allocation, and the only way to move it was a second inter-state transfer, a second e-way bill, a second freight bill and a sales team asking distributors to take stock they had not ordered. Every step had followed the plan; the plan had simply been built on a forecast.

Push and pull are not philosophies. They are two different triggers for moving stock: a forecast and an allocation plan in one case, a consumption signal in the other. Each fails in predictable conditions, and the cost of each failure can be estimated before you choose. This guide shows how to do that estimate, and where a hybrid with a clear hand-over point beats either pure model.

• What this guide covers: the trigger behind each model, the specific ways each one breaks, the arithmetic of coverage and safety stock, and the Indian tax steps that make a wrong push more expensive.
• What it adds: a worked network comparison in dollars and rupees, a pair of replenishment formulas, a festive-season incident trace and a checklist for choosing the hand-over point.

Forecast triggers and consumption triggers: what actually moves stock

A push network ships stock because a plan says demand will come; a pull network ships stock because demand has already consumed some. Everything else, from inventory levels to freight patterns, follows from which event creates the transfer order. Name the trigger precisely before arguing about which model you run.

• Allocation run: a central planner divides forecast quantities among depots or distributors for the coming period and releases transfer orders in bulk. Classic push.
• Distribution requirements plan: time-phased net requirements per location, built from forecasts, that release planned transfers on a schedule. Push with more structure.
• Reorder point or base stock: a depot replenishes when on-hand plus in-transit stock falls below a level set from recent consumption. Classic pull.
• Kanban or bin signal: an emptied container or a scanned card authorises a fixed replacement quantity. Pull at its simplest.
• Customer-sent consumption data: point-of-sale or warehouse withdrawal data from a retailer or distributor that drives the supplier's shipments. Pull extended across a company boundary.

Most real networks mix the two. A manufacturer may push finished goods from the plant to a central warehouse on a production plan, then let regional depots pull from that warehouse weekly. The point where the trigger changes is the decoupling point, and choosing it well matters more than choosing a label for the whole network.

Our article on forecasting distributor demand without treating stockouts as low demand covers how to build a clean forecast series. This guide assumes you have one and asks what should move stock once you do.

How push distribution fails, and the warning signs to watch

Push fails when the forecast is wrong at the level where stock is placed. A national forecast can be accurate within a few percent while individual depots miss by 30 percent or more, because errors at small locations do not cancel the way they do in aggregate. The visible symptom is stock in the wrong place, not too much stock overall.

• Location mismatch: one depot short and another overstocked on the same item in the same week, followed by rebalancing transfers.
• Distributor loading: primary sales into distributors run ahead of their secondary sales to retailers, and distributor stock days climb quarter after quarter.
• Expiry and obsolescence: short-life or seasonal goods pushed early sit until they are marked down, returned or written off.
• Freight doubling: stock travels twice, out on the allocation and back or sideways on the correction, with handling at every touch.
• Late signal: planners learn about a miss only at the next forecast cycle, often a month later, by which time the season has moved.

Push distribution to Indian depots in other states is a taxable supply between distinct persons under CGST Act Schedule I paragraph 2 read with Section 25(4), so every rebalancing transfer back to the central warehouse is a second supply needing a tax invoice and, above ₹50,000, a fresh e-way bill.

How pull distribution fails when lead times or demand misbehave

Pull fails when the replenishment signal cannot travel back and be answered within the time customers will wait. Long supply lead times, lumpy orders and sudden step changes in demand all break the assumption that recent consumption is a good guide to the next replenishment. The symptom is stockouts at locations that were following the rules.

• Long or variable lead time: a depot fed by a three-week transit has to cover three weeks of uncertainty, so its pull level rises towards what a push plan would have placed anyway.
• Promotions and launches: past consumption says nothing about a product that has not sold yet or a price cut that has not started.
• Upstream capacity: pull assumes the central warehouse or plant can answer each signal; if it is capacity-bound, signals queue and depots run dry.
• Order batching: depots that pull in large, infrequent batches send the upstream site a spiky signal, the familiar bullwhip in a pull costume.
• Signal latency: consumption data that arrives weekly in a file is a slow pull; by the time it lands, the shortage has already happened.

A useful screen is the ratio of supply lead time, from replenishment decision to stock on the shelf, to the time a customer will wait. When supply lead time is shorter than customer patience, pull works almost everywhere. When it is much longer, some stock has to be positioned ahead of demand, which means some push upstream of the decoupling point.

Replenishment arithmetic: coverage, safety stock and the decision rule

Both models can be written as formulas, and writing them down shows where each one carries its risk. Push carries risk in the forecast term; pull carries it in the lead-time term. The block below gives the standard forms with every variable defined, so a planner can plug in a depot's own numbers.

Pull: periodic-review base stock for depot i
  S_i = d_i x (R + L_i) + z x sigma_i x sqrt(R + L_i)
  order_i = S_i - (on_hand_i + in_transit_i)        # placed every R weeks

Push: allocation for depot i over plan period T
  A_i = F_i x (T + L_i) + SS_i - (on_hand_i + in_transit_i)
  sum(A_i) <= central_available                    # fair-share if short

Decoupling screen (per item family)
  P = decision-to-shelf supply lead time (weeks)
  D = time the customer will wait (weeks)
  if P <= D: pull from the source
  else:      push to the last stage where P_remaining <= D, pull after it

where
  d_i      average weekly consumption at depot i
  sigma_i  standard deviation of weekly consumption at depot i
  F_i      forecast weekly demand for depot i
  SS_i     safety stock for depot i under the push plan
  R        review interval, weeks
  L_i      replenishment lead time to depot i, weeks
  T        allocation plan period, weeks
  z        service factor, e.g. 1.65 for about 95% cycle service

Notice what each formula feeds on. The pull level depends on actual consumption and its variability, so it adapts automatically when demand shifts, but it grows with the square root of lead time. The push allocation depends on the forecast for each depot, so a biased forecast at one location places the wrong stock there, however good the national total looks.

Our guide to min-max and wave-demand replenishment inside the warehouse applies the same trigger logic one level down, from reserve storage to pick faces. The network decision in this article and the in-building decision there often use the same software but need different parameters.

Annual cost of each model on a four-depot network, US and India

On the illustrative four-depot networks below, weekly pull from a central warehouse costs less than monthly push in both markets, by about $66,000 a year in the US and about ₹21 lakh in India. Pull wins on inventory and corrections and loses on freight. Every figure is illustrative and assumes a one-week replenishment lead time.

Both networks have a central warehouse feeding four regional depots. The US case moves 4,000 units a week at a unit cost of $60 and a 25 percent annual carrying rate. The India case moves 6,000 units a week from Bhiwandi to depots in four other states, at ₹2,400 a unit and an 18 percent carrying rate. Push holds four weeks of cover; pull holds 2.7 weeks across depots and the central buffer.

Rebalancing uses $3.20 a unit in the US and ₹70 a unit in India for the second move and handling; lost margin uses $25 and ₹600 a unit. Stretch the replenishment lead time to four weeks, as on an import-fed lane, and the pull base stock rises from two weeks of demand to five, with safety stock up about 60 percent, while the freight premium remains. That is often enough to make push the cheaper choice for that leg.

Weekly pull replenishment from a central warehouse to four regional depots cut illustrative annual cost from $766,448 to $700,387 in the US, and from ₹2.80 crore to ₹2.59 crore in India, mainly because average inventory fell from four weeks of cover to 2.7 weeks while US freight per unit rose by about 24 percent.

Indian stock transfers: GST steps that make a wrong push expensive

In India, moving stock to a depot in another state is a taxed supply, so every push and every correction carries invoicing, valuation and e-way bill work. A US network moving stock between its own warehouses has no federal tax step for the transfer itself, which is why the cost of a wrong push differs between the two markets.

Section 25(4) of the CGST Act treats each registration of the same company as a distinct person, and Schedule I paragraph 2 makes supplies between distinct persons in the course of business a supply even without consideration. CGST Rule 28 values such transfers at open market value; where the receiving registration is eligible for full input tax credit, the invoice value is deemed to be that value.

• Transfer invoice: the sending registration issues a tax invoice for each inter-state stock transfer and pays the tax; the receiving depot claims the credit.
• Valuation option: where goods are for further supply as such, the first proviso to Rule 28 lets the supplier choose 90 percent of the price charged to unrelated customers.
• E-way bill: CGST Rule 138(1) requires Part A before movement above ₹50,000 consignment value, including movement for reasons other than supply.
• Return transfer: sending stock back to the central warehouse is a new supply in the opposite direction, with its own invoice and e-way bill.

Push into distributors adds another clock. If unsold stock comes back and the original sale is reversed by a credit note, Section 34(2) requires the note to be declared by 30 November after the financial year of the supply or by the annual return date, whichever is earlier. Since 1 October 2025, the proviso to 34(2) also allows the output-tax reduction only where a registered recipient has reversed the matching credit.

A festive-season push that had to be pulled back

The costliest push errors are seasonal, because the forecast is built on one data point per year and the window to correct is short. The trace below uses invented document numbers to show where a consumption check would have intervened.

1. 6 October, 10:00: allocation run AL-1006 releases transfer orders for 2,400 room heaters to the Lucknow depot and 2,400 to the Jaipur depot, sized on the previous winter's sales. Transfer invoices and e-way bills are generated at Bhiwandi.
2. 3 November, 18:30: week-four consumption shows Lucknow at 71 percent sell-through and Jaipur at 22 percent. The next allocation is not due until 1 December, so nothing changes.
3. 21 November, 09:15: Lucknow stocks out of the two fastest models. Distributors in Jaipur, pressed to take extra stock at month-end, hold 34 days of cover.
4. 2 December, 14:00: transfer order STO-55871 moves 900 units from Jaipur back to Bhiwandi for re-allocation, with a new tax invoice, a new e-way bill and a second freight charge, while Lucknow waits for the next truck.
5. With a weekly pull check: the 3 November consumption signal would have triggered a Lucknow replenishment from the central buffer and held Jaipur's second tranche, so no unit would have moved twice.

Fixed replenishment rules versus planner judgement in each model

Automate the arithmetic that follows directly from data, and keep forecast overrides, allocation under shortage and distributor pressure as named human decisions. The table lists the actual planning tasks and where software or machine learning helps without taking the decision.

Where a large retail customer sends forecasts and firm releases by EDI, those messages are a pull signal from outside your company. Our article on the X12 830 planning schedule explains how forecast and firm horizons in that transaction should feed your replenishment rules.

Choosing the hand-over point: a checklist for a hybrid network

Push to the last stage where supply lead time still exceeds customer patience, and pull from there to the customer. Work item family by item family, because seasonal and long-lead goods often need push upstream while fast-moving staples can be pulled all the way back to the plant.

Decoupling-point selection compares decision-to-shelf supply lead time with the time customers will wait: where lead time is at or under that wait, depots pull from the source; where it is longer, stock is pushed to the last stage that can answer in time, such as a one-week depot leg against a two-week promise.

Review the hand-over point twice a year, before the main season and after it, using the cost table above with your own figures. For related planning and distribution topics, browse the logistics and supply chain hub.

If you are redesigning depot replenishment or want consumption signals from distributors to drive transfers automatically, describe your network through our project request form and we will outline an approach for your item families.