In many organizations, if you ask 'who owns the demand forecast?' you hear several different answers: sales says it belongs to planning, planning says sales does not provide information, and finance has its own number from the budget. The result is several 'versions of the truth', each built for a different purpose, and it is unclear which one inventory and production decisions ride on.
Roles must be separated. The demand planner owns the process and the baseline statistical forecast: cleaning data, running the model and measuring accuracy. Sales owns market information: key customers, contracts, projects. Marketing owns information on promotions, new-product launches and pricing. Finance owns valuation and alignment with the budget.
The key principle is that sales and marketing provide inputs, but one person owns the final number. If the final owner is unclear, each side pushes its own number and the S&OP meeting argues about the number instead of deciding. Usually the process owner is best placed in the demand planner or their manager, with explicit authority to accept or reject inputs based on evidence.
For sales inputs to be valuable they must be structured. Instead of 'I think sales will rise', ask the salesperson to say which customer, what volume, with what probability and from when. And track the accuracy of each source's input (forecast value added): if the sales team's adjustment improves accuracy on average, it earns more weight; if it worsens it, it needs review.
Incentive alignment matters too. If sales is measured only on achieving the sales target, it has an incentive to under-forecast (to make the target easier) or over-forecast (to secure stock). Measuring and making transparent the forecast accuracy of each input changes behavior without punishment.
A common mistake is leaving the forecast to software alone and removing humans from the process. The statistical model learns from the past, but it does not know that a big customer is coming tomorrow or that a competitor is entering the market. The opposite mistake is unaccountable manual intervention that turns the forecast into personal opinion. The right combination is a baseline model plus documented, evaluable manual adjustment.
A simple framework for starting: name the process owner, write down each role's inputs and delivery deadlines, define a structured input template, and hold a pre-S&OP meeting to reach a consensus forecast before the main meeting.
A practical starting point: this week, ask each of the four roles 'which forecast number do you use?' If you hear several different answers, your problem is not the algorithm but ownership, and naming a single owner will be the first improvement step.