The most expensive assumption in analytics is that a solved bottleneck stays solved.

Theory of Constraints taught this on factory floors decades ago. Fix the slowest work center and the constraint moves. Some other part of the system becomes the new limit, usually where nobody is looking, because everyone is still watching the old fix hold.

A business runs into one binding constraint at a time, with less visibility than a factory floor has. You can watch it travel. Sales is the ceiling, so you sharpen the pipeline and close more. For a month it feels solved. Then fulfillment cannot ship what sales now brings in, and the deals that felt like wins sit aging in a queue. You fix fulfillment, and support inherits all the volume fulfillment finally let through. You staff support, and cash becomes the limit, because everything downstream now moves faster than the money coming back. Every fix worked. The pressure relocated to the next station, and it will keep relocating for as long as the business grows.

The reason it keeps catching teams off guard is that attention does not travel with it. A constraint that gets solved also gets celebrated, and celebration is a spotlight pointed at the place the problem used to be. The new limit sets up exactly where the light is not: in the handoff nobody profiled, the approval step that quietly became a bottleneck, the one person who turned into the human queue for every request. It does not announce itself. It just starts adding days, in a corner of the business still trusted because it was never the problem before.

So the next mistake writes itself. The team keeps optimizing the constraint it already beat. More tooling for a sales motion that is no longer the limit. A faster report on the process that stopped being slow. This feels like progress and produces almost none, because effort spent anywhere other than the current constraint does not lift the system. An hour saved at a station that was never the bottleneck is an hour saved nowhere. The dashboards built to watch the last constraint keep glowing green while the real delay grows somewhere off-screen.

Finding the new one cannot be done from memory. By the time a constraint is obvious it has usually already moved again, so the honest answer always takes current data. The signal is boring and reliable: look for where work piles up and waits. The constraint is wherever the queue is longest, where lead time concentrates, where work waits between steps. Intuition points at the last fire. Measurement points at the next one.

Which is why analytics itself never finishes. The target it exists to track keeps moving through the business, so the analytics has to move with it. A dashboard designed around last year’s constraint is now watching the wrong station with great precision. The same discipline you would apply to the business applies to the analytics that watches it: assume the important question has changed, check where the limit actually sits now, and keep the models current enough to answer honestly. Analytics that cannot evolve at the speed of its own business slowly becomes a very confident record of a problem you already solved.

That is the whole case for treating analytics as living infrastructure you keep current as the business shifts under it. The constraint will move again next quarter. The only question worth asking after any win is where it went while everyone was still celebrating the last one.