At some point in the last decade, most large organizations bought into connected planning. One platform, one model, one version of the truth. Finance, supply chain, and commercial teams all working from the same numbers, in real time, without the spreadsheet archaeology that preceded every executive meeting.
The vision is compelling. The reality is that you now have the same arguments about whose numbers are right, but they happen inside a very expensive SaaS platform with a much nicer UI.
What "Connected" Actually Means
A planning model connects data from multiple sources. This sounds straightforward until you realize that each source was built by a different team, governed by different rules, and has a different definition of what "this week's actuals" means.
Sales says revenue when the order is placed. Finance says revenue when it ships. Operations says volume when it hits the warehouse. All three are right by their own definitions. None of them are compatible.
Your connected planning model ingests all three, runs its logic, and produces beautifully formatted output that confidently aggregates numbers that were never meant to be aggregated. Then someone in the meeting asks "which line item is this?" and you spend 25 minutes inside a tool that was supposed to save everyone time.
Assumptions Are Load-Bearing and Nobody Labels Them
Every planning model has assumptions built in. Lead times, yield rates, minimum order quantities, replenishment rules, safety stock formulas. These were set by someone, at some point, under conditions that may or may not still exist.
The insidious thing about a mature planning model is that these assumptions become invisible. They were correct when the model was built. They have been wrong for 18 months. Nobody changed them because nobody is quite sure where they live, and the last person who knew is at a different company now.
When the model output stops matching reality, the first instinct is to question the algorithm. The algorithm is fine. The algorithm is faithfully executing assumptions from a world that no longer exists.
The Dashboard Problem
Connected planning tools produce dashboards. Dashboards produce confidence. Confidence produces decisions. And decisions made with confidence in a model that contains stale assumptions produce expensive outcomes.
I have watched organizations plan entire replenishment cycles off scenarios that looked authoritative because they were in a proper planning tool, color-coded, with variance arrows, when the underlying lead time assumption was based on a supplier relationship that had been renegotiated 14 months prior.
The dashboard is not the truth. The dashboard is a rendering of the model. And the model is a rendering of assumptions made by humans who were doing their best at the time.
What Actually Works
None of this means connected planning tools are bad. They are genuinely useful when used honestly. A few things that help:
Treat assumption management as a first-class concern. Every major assumption in your model should have an owner, a review cadence, and a flag when actuals diverge from the assumption by more than a defined threshold. This is not interesting work. It is the work that makes everything else reliable.
Version your model like code. Know what changed between this week's run and last week's run. Not just the numbers but the structure. If a formula changed, that is a release, not a routine update.
Build in explicit uncertainty. A model that shows one number implies certainty it does not have. Models that output a range, with a clear explanation of what drives the variance, produce better conversations and better decisions. Planners argue less about which number is right when the model is honest about not knowing exactly.
Use AI to surface assumption drift. Train a model to flag when historical actuals are systematically diverging from model outputs across specific dimensions. That divergence is usually a stale assumption trying to get your attention.
The Honest Version of the Value Proposition
Connected planning tools do not give you one version of the truth. They give you one place to have the argument about truth, which is meaningfully better than having it across 14 versions of a spreadsheet on a shared drive with conflicting filenames.
That is real value. It just requires you to maintain the model with the same rigor you would apply to production software.
Otherwise you are not doing connected planning. You are doing disconnected planning with better aesthetics.