You’re forecasting at item level, and yet you regularly end up with too much of one thing and too little of another. A big customer drops off and you only notice once the orders stop coming in. Another one starts a project and hits you with an order that wipes out your stock.
If that sounds familiar, the problem probably isn’t the quality of your item-level forecast. You are most likely only doing half the work: you forecast at item level, and that’s where it stops. That isn’t wrong. It’s just incomplete.
Two directions, two kinds of knowledge
Worth being precise about what top-down means here, because the word gets used two ways. This isn’t about a revenue target for the whole business that you break down by region and channel. The highest level you forecast at is the item: one expectation per item, across all your customers. The model looks at that item’s history, picks up trend and seasonality, and extends the line. It’s good at that. Better than a person, and it scales across your entire range.
Bottom-up starts one layer lower: at the demand of one customer for one item. You add those customer-item lines up, and that brings you to the same item level. What you gain here isn’t computing power but knowledge: what your account manager knows about a customer appears nowhere in the history.
Two routes, then, with the same destination. They aren’t competing methods but two sources that see different things — and most planning tools only offer you one of them.
What disappears when you add customers up
The moment you add customers up into a single number per item, everything happening at customer level disappears into the aggregate.
Take a new project. A customer is adding a new product line and will order double for the next six months. None of that shows up in the item’s historical data, so the item forecast misses it entirely. Your sales rep knew weeks ago.
Or take churn. A customer winds down, or moves to a competitor. At item level that’s offset by growth elsewhere, so the number still adds up while the situation underneath shifts. You only see it in the figures once the gap gets too big to mask.
And then there’s changing behaviour. A customer starts ordering in larger quantities less often, or switches to a different pack size. The order pattern changes, the number per item stays the same, and so the model doesn’t adjust.
Every one of these is knowledge that exists somewhere in your organisation, with the people who talk to the customer, but never reaches the planning. A better model at item level doesn’t solve that. No model extracts from history what isn’t in there yet.
Mature planning teams do both
The teams that have this right don’t choose. They run two forecasts side by side, with a clear owner for each.
Supply owns the item forecast. One statistical forecast per item, across the whole range. That’s the baseline, and it stays exactly where it is.
Sales owns the customer-item forecast. Pre-filled from history, so reps don’t start from scratch. They look at the customers they know and correct where they know something the model doesn’t: that project, that customer winding down, that tender coming up.
That division of labour is what makes it workable. Sales doesn’t fill in the entire range, only what they know better than the model. Supply doesn’t have to invent customer knowledge they don’t have.
The difference between the two is your signal
This is where the actual work happens. You aggregate the customer forecast up to item level and put it next to supply’s item forecast. Two independent answers to the same question, at the same level.
Where they agree, you don’t need to do anything. That’s the vast majority of your range, and you can let it run with a clear conscience.
Where they diverge, someone knows something. Maybe sales sees a project coming that the model can’t see. Maybe an account manager’s forecast is too optimistic. Either way that’s a conversation worth having, and now you know exactly which items that conversation should be about.
That’s managing by exception: not reviewing every item by hand, but only the cases where your two sources disagree or the risk is high. Without that second forecast you don’t have that signal. A deviation only becomes visible once it has already turned into a shortage or a surplus.
That precision carries through to your stock. A sharper demand forecast is the foundation under good inventory optimisation: less buffer on items you’re confident about, more targeted intervention where it pinches. At Lienesch, better planning translated into one and a half million euros less tied up in inventory.
You don’t have to replace anything
The biggest concern with something like this usually isn’t the technology, it’s the project. Rolling out a new planning system feels like months of implementation and a lot of resistance.
That’s not the case here, because you’re not replacing anything. Your item forecast stays right where it is, in the tool you already use. What gets added is the customer layer underneath it, and the comparison between the two. If you work with a tool that’s strong at item level, Innico plugs in on top of it as an add-on. If you don’t have a supply forecast yet, Innico can do that for you too.
The real change is involving sales. And that’s easier than you’d think, because your reps get something back. Their knowledge about customers finally counts in the planning, instead of evaporating into a single number per item.
From a forecast that’s right to a forecast you can steer on
A forecast that’s right at item level isn’t the same as a forecast you can steer on. The difference isn’t a better method, it’s a second one: the customer layer underneath, and the gap between the two as your signal.
Curious where your two forecasts would disagree?
Curious where your two forecasts would disagree?
See in a demo what a customer forecast alongside an (existing) item forecast looks like.
Banner image by Vitaly Gariev on Unsplash