At 7:40 on a Tuesday morning, an assembly plant in Nagoya changes the daily kanban pull for a compact vehicle model from 420 instrument-panel brackets to 510, effective the following week. The plant scheduler sees the change immediately. The Tier-1 module supplier receives a revised schedule and adjusts its production plan. Three weeks later, a Tier-2 fastener supplier two tiers upstream is still working from the order pattern it received before the change, with no clear view of the pull signal that caused it.
This is a common shape of demand latency in a multi-tier automotive supply chain. The plant has a near-real-time view of takt time, line loading, model mix, and the parts required for the next production interval. Each upstream supplier sees a more filtered representation. By the time a change appears as a purchase order, it may be delayed, aggregated, or amplified. A supplier can then make a reasonable decision from incomplete information and still produce the wrong quantity.
The signal changes as it moves upstream
A plant pull schedule is not the same object as an upstream order. A kanban pull is tied to consumption or a planned consumption rate at a specific point in the assembly process. It may reflect the plant's current takt time, model sequence, and replenishment loop. An MRP order is usually a planning instruction generated from a bill of materials, inventory position, lot size, lead time, and planning horizon.
Both signals matter, but they answer different questions. The plant pull asks, "What does this line need next?" A supplier order often asks, "What quantity should I make or ship after applying our planning rules?" The second signal can be useful operationally while hiding the first signal's timing and context.
Consider the bracket change in the opening scenario. The plant increases its pull because the model sequence has changed. The Tier-1 supplier may see a revised call-off and increase module production. Its ERP system then converts that requirement into component orders. The Tier-2 fastener supplier receives those orders according to a batch size and release calendar. If the fastener supplier has a longer replenishment cycle, the original change can arrive after the relevant production window has already passed.
This is signal propagation delay. It is not only the time required to transmit a message. It includes the time spent batching demand, running planning jobs, waiting for approval, converting units of measure, and releasing an order through each tier. A one-day change at the plant can become a multi-day or multi-week change in an upstream planning system.
Latency and amplification create a different demand history
When each tier plans from its own order history, upstream demand often becomes more variable than the plant's actual consumption. This is the Bullwhip effect, but the mechanism is more specific than a general statement about volatility. Each tier is reacting to a delayed and transformed observation of the same underlying pull.
A supplier may see an unusually large order and interpret it as a sustained increase. It may add a buffer for uncertainty, round the quantity to a production lot, or order extra material because the next confirmed schedule is not visible. Another supplier may receive a smaller order because inventory at the intermediate tier temporarily absorbs the increase. Both decisions alter the shape of the demand signal moving upstream.
Order amplification can also occur when a plant reduces a pull. A Tier-1 supplier may have already released component orders based on the previous schedule. The Tier-2 supplier then sees a near-term order that no longer matches plant consumption. If it responds by cutting its own orders sharply, the next planning cycle may overcorrect in the other direction.
For forecasting, this means that an upstream order history is not a clean proxy for end demand. It contains the effects of inventory policies, release calendars, minimum order quantities, transport schedules, and local reactions to uncertainty. A model that reads only the latest purchase order may identify a pattern, but it may identify the planning process rather than the production requirement.
Translating a pull into tier-specific demand
The useful forecasting unit is not a single demand number copied to every supplier. Each tier needs a forecast that reflects its position in the bill of materials, its lead time, and the way requirements are released to it.
Lean Path reads multi-tier demand signals from assembly plant pull schedules and translates the plant signal into forward demand at each upstream tier. In practical terms, that means tracing how a change in plant consumption maps to the relevant assemblies, components, and parts. The result is a tier-specific view of expected demand rather than a duplicated plant schedule.
For example, a change in vehicle build mix may increase the requirement for a module but affect several component suppliers differently. One component may be consumed one-for-one. Another may have a different bill-of-material quantity. A third may be held in intermediate inventory and therefore require a later release. The forecast should preserve those distinctions.
The timing also matters. A Tier-1 supplier may need a forecast for the next production week. A Tier-2 supplier may need forward demand far enough ahead to cover its manufacturing lead time. A Tier-3 supplier may need an even earlier view because raw material procurement or process capacity constrains its response. Translating the signal means carrying the plant's change upstream with the timing appropriate to each tier, not simply sending the same number farther back.
Lean Path generates tier-specific demand forecasts weekly. It can work with standard ERP export formats and direct API connections, depending on what a supply network can provide. Exception detection flags deviations from standing order patterns, such as a change that is materially different from the release behavior normally associated with a part or supplier. That gives a planner a place to inspect the signal before it becomes an unexplained shortage or excess.
What the forecast can and cannot repair
Signal translation can improve the interpretation of demand already present in plant schedules and connected planning data. It can distinguish a plant pull change from the order amplification introduced by an intermediate tier. It can also expose a likely change earlier than a supplier would see it through a conventional order cycle.
This is not a problem of data availability alone. A complete schedule does not automatically produce a useful upstream forecast if part mappings are wrong, lead times are stale, or inventory positions are missing. Signal translation cannot manufacture capacity, correct an inaccurate bill of materials, remove a supplier's material constraint, or guarantee that a customer will keep a provisional schedule. It gives planning systems a better representation of demand, while execution still depends on the data and decisions surrounding that representation.
There are also cases where propagation is genuinely hard. ERP systems may use incompatible part identifiers or different unit conventions. EDI formats may omit the context needed to distinguish a model-mix change from a routine call-off. Some suppliers may be reluctant to share detailed inventory or production data, particularly when commercial relationships span several tiers. A useful implementation has to work with partial visibility, record its assumptions, and make uncertain mappings reviewable rather than hiding them behind a single forecast value.
From connection to an operating forecast
Our implementation sequence reflects that dependency. During Connect, which typically covers days 1 to 3, the team links the available ERP exports or API connections and identifies the relevant plant, supplier, part, and schedule fields. During Calibrate, in weeks 1 and 2, we examine the mappings, historical order behavior, timing conventions, and tier relationships. The first forecast is produced in week 3, followed by refinement toward full rollout around week 6.
For a small engineering team, this staged approach is important. We are not assuming that every supply network has a clean, shared data model waiting to be queried. The work is partly forecasting and partly reconstruction of how a demand signal is transformed between systems. A forecast that looks mathematically precise but cannot be traced to a plant pull, a part relationship, and a timing assumption is difficult for a planner to act on.
The practical test is whether a planner can answer three questions when a forecast changes: which plant signal moved, how that movement was translated through the bill of materials, and why the timing differs by tier. If the answer is visible, the upstream supplier has a basis for planning before the change arrives as a noisy order. If the answer is not visible, another forecasting layer may only add confidence to the same delayed signal.
In the Nagoya scenario, the relevant observation is not simply that the fastener order increased three weeks later. It is that the plant's change in takt-linked pull created an earlier forward requirement for the fastener, subject to the Tier-1 assembly relationship, intermediate inventory, and the Tier-2 lead time. That distinction is the difference between reading an order and reading the demand signal behind it.