Try it sometime, with your kids or to break the ice in a meeting: lay a rope on the floor and try pushing one end so it travels in a straight line. It buckles, it piles up, it goes anywhere except where you want. Now pull the same end. The rope tightens, follows your hand, moves cleanly. Simple as that… and yet it's exactly the debate that has split push and pull flow for the last forty years in our supply chains.
MRP calculates a plan at a given point in time - bills of material, lead times, safety stocks - and pushes it into the system: "here's what to produce, in this order, on these dates." It's neat, rational, and it has the enormous merit of having structured an entire profession since the 1970s. The problem, every planner knows it, is that tomorrow's reality won't tell the same story as today's plan. An urgent order comes in, a machine breaks down, a supplier ships late… and you have to push the rope again, and again, with a bit more slack piling up along the way each time. That's the bullwhip effect: a tiny wobble at the far end of the chain becomes a tidal wave three tiers upstream.
Pull flow works the other way round, which is exactly why it works at all: real consumption downstream pulls replenishment from upstream. Toyota proved it decades ago with kanban - you only make what's been consumed, you only replace what's disappeared from stock. The rope stays taut at all times because it's pulled by real demand, not by a forecast. DDMRP applies the same principle through the net flow equation: you don't steer on projected inventory - a hypothesis, a fiction - but on what's actually on hand, on order, and consumed, today.
So why not extend pull flow across the entire chain, from the end customer all the way back to the iron mine? Because a rope has a length. And the longer it gets, the less efficient it becomes to pull on it from a single end: the signal stretches, dilutes, and arrives so delayed it barely means anything by the time it reaches the other end. Ask an aircraft manufacturer to pull its raw titanium demand straight off an A320 delivery: the cumulative lead time runs into years. No physical kanban survives that.
This is where block-and-tackle comes in - the pulley system that crane operators and sailors use to multiply a force and change its direction without ever "pushing" it. A pulley pushes nothing: it relays a pull, redirects it, and above all absorbs local shocks before passing them on further, smoothed out. That's exactly what decoupling points do in a Demand Driven model.
But there's a second family of pulleys, just as useful: the ones we call tensioners - the idler pulley on a belt, a bike chain, a conveyor. Their job isn't to redirect the pull but to regulate it continuously: as the available length of rope varies, the tensioner moves forward or back on its own, keeping tension constant - never too tight, never too slack. That is precisely what a DDMRP buffer does: a regulation loop that continuously absorbs variations in consumption and lead time, rising or falling through its zones - red, yellow, green - to keep the right tension on the flow, without anyone having to intervene at every jolt.
A decoupling point, with its buffer, plays the role of the pulley: it relays the demand's pulling signal upstream, but it never brutally transmits the variability it just absorbed. The red zone soaks up the randomness, the yellow zone covers demand over the decoupled lead time, the green zone regulates order frequency and size. The result: each link genuinely pulls on its own short, manageable stretch of rope - its decoupled lead time - without having to carry the cumulative lead time of the whole chain, and without passing every twitch of the end customer straight through to its own supplier. The pull has been multiplied by relaying it, exactly the way a block-and-tackle multiplies the effort of a single arm to lift a load no hand could pull directly.
This is what the DDOM does at the scale of an end-to-end chain: position the right pulleys (the decoupling points), in the right places, with the right buffers - stock, time, or capacity depending on the case - so that every segment stays pullable, and the pull propagates without ever degenerating into blind push. On stretches too long, too uncertain, or too capacity-bound to be steered directly - an aerospace order book, a component running at the pace of a bottleneck - you keep a scheduling logic, a Drum-Buffer-Rope cadence, but always synchronized to the bottleneck's actual consumption rhythm, never to a plan pushed down from the top.
That's why I'm wary of the binary "pull versus push" debate, and just as wary of software promises claiming to steer everything through a single optimal plan. Common sense on the ground says no real value chain is a single rope, pulled taut in one piece from a tier-5 supplier to the end customer. It's a system of segments connected by well-placed pulleys, each one able to absorb its share of variability without passing it on unchanged to the next link.
Demand Driven doesn't reject MRP or Lean: it orchestrates them, installs the right pulleys at the right points in the flow, and accepts that on certain segments you keep scheduling - but never blind pushing. You just have to accept letting go of the rope where it needs to be pulled, and fitting a block-and-tackle - with its tensioner - where it's too long to be pulled in one single motion.