Toyota stops asking factory stations to guess how much to build and copies how a supermarket shelf decides instead

问题

Every stage of a production line has to forecast how much the next stage will need, and forecasts are wrong in both directions

背景

换别人会怎么做

Improve forecasting accuracy — better demand models, more data, more careful scheduling. It fails because forecasting inherently carries error, and on a multi-stage line those errors compound as they propagate: a small miss at an early station cascades into a much larger mismatch several stations downstream, no matter how good any single forecast is.

他们看到了什么

Watching American supermarkets, Ohno saw that a shelf restocks only after a customer has actually taken an item, never on a schedule built from a guess about future demand — nobody is predicting shopper behavior days ahead, the store simply reacts to what just happened. The insight was that the entire cascading-forecast-error problem could be eliminated by inverting the flow of information: instead of each station pushing output forward based on its own prediction, each station should pull only what the next station has just actually consumed.

那一手

Ohno ported that exact logic onto the factory floor starting in Toyota's main plant machine shop in 1953: each production step was treated as a 'customer' of the step before it, allowed to pull parts only as it actually consumed them, signaled by a physical card — the kanban — that traveled back down the line as a trigger to produce more. Every station now reacted to real, one-step-away consumption instead of a schedule built from a forecast of downstream demand.

为什么管用

Push production requires every station to forecast the next station's future needs, and any error compounds as it moves down the line, since each station's own miss adds to whatever error already existed upstream. Replacing that forecast with a kanban pull signal means each station reacts only to real, already-observed consumption one step away, not a prediction — there's no forecast to be wrong about at any individual handoff, because the system only responds to what has already happened rather than guessing what will happen next. That eliminates compounding error entirely, which is why Toyota could run with dramatically less work-in-process inventory than push-based competitors while surfacing supply-demand mismatches immediately at the point they actually occurred, instead of after inventory had piled up several stations downstream.

值了多少

The pull system, alongside jidoka (automation that stops on defect), became one of the two founding pillars of the Toyota Production System, letting Toyota run with dramatically less work-in-process inventory than push-based competitors while surfacing supply-demand mismatches immediately at the point they occurred instead of after inventory had piled up several stations downstream.

什么时候会失灵

The system only works when production can respond with the low latency a pull signal requires — a process with long lead times or high setup costs between stations can't react fast enough without risking stockouts or line stoppages while waiting on the next batch. It also depends on relatively stable overall demand even once stage-to-stage forecasting is removed; a pull system handles normal variation well but can struggle against sudden, large demand spikes that exceed its built-in buffer, since there's no forecast-driven safety stock accumulated in advance. And it requires disciplined, consistent adherence to the pull signal at every stage — if any station reverts to producing on its own schedule instead of strictly on kanban triggers, the compounding-error problem the whole system was built to eliminate can creep back in at that one point and cascade forward again.

后来呢

Ohno formalized the method in his 1978 book Toyota Production System: Beyond Large-Scale Production, and kanban spread from Toyota's factory floor to become a standard term and technique across global manufacturing and, decades later, software development and knowledge work, cited as the origin of 'just-in-time' and 'pull' as general alternatives to forecast-driven push production.

资料来源

1. [1]What is The Toyota Production System?Management And Accounting Web (academic article summaries, Univ. of South Florida), 2010maaw.info
2. [2]Supermarket — TPS EncyclopediaArt of Lean (Art Smalley, former Toyota TPS trainer), 2020artoflean.com

收下它

同一路聪明

• 2010 · American ExpressAmex didn't advertise its card on the biggest shopping weekend of the year — it invented a whole new day inside it
• 1962 · AvisAvis told customers outright it was only #2, and the confession made everything else it said believable
• 2025 · AXA France (with Publicis Conseil)AXA covered domestic violence like a house fire by changing three words in a contract millions of people already had