Forecasting the mix: demand planning for ETO and CTO manufacturers

Part 2 of 3: Planning what you haven’t designed yet

In Part 1, we made the case for planning the model and the mix instead of the part number. That raises the obvious next question. How do you forecast demand for products that haven’t been configured yet, and what should demand planning hand to supply planning? The answer combines model forecasts, attach rates that reflect each customer, and statistical forecasts of the options themselves, blended into one prediction.

Why demand planning is harder in ETO and CTO

Sales history is spread thin across thousands of configurations, so few of them have enough history to forecast. Yet both the models and the options need to be considered to decide which materials to buy and which resources get loaded. Demand also arrives as projects and quotes, so a single large order can swing the mix for months.

The common starting point, and its three limits

Most ETO and CTO planners start the same way. They forecast the model, where history is deep enough to be meaningful, then apply attach rates: the share of model orders that carry each option, taken from past orders. A forecast of 100 forklifts with a 40% attach rate for the heavy chassis becomes 40 heavy chassis.

It works, up to a point. A single average rate treats every customer the same. Each option is treated as if it were chosen independently of the others. And rates built on history react slowly when buying patterns shift.

Fixing the three limits

Customer differences. Customers don’t buy the same mix, so attach rates are kept by region, by key customer and for large CRM projects, with a rest-of-market rate for everyone else. When a state roadworks department that always orders heavy chassis puts a fleet opportunity in the pipeline, the forecast mix shifts toward what that customer buys, not toward the average.

Option dependencies. Options travel together and rule each other out. A larger engine may require the heavy chassis, and some combinations can’t be built at all. Multiplying independent rates can call for combinations that never ship and misstate component demand. The fix is to keep attach rates in the context of related choices and to check the combined forecast against the configuration rules, so the predicted mix is one that can actually be built.

Changing demand. Some options have demand patterns of their own: a new engine type gaining share every quarter, a seasonal option, or an option sold across many models. A statistical forecast of the option itself, using its demand history across all models, picks up these trends sooner than attach rates can.

One prediction, using the best method

Neither method wins everywhere, so the best approach uses both. For each option, the attach-rate forecast and the direct option forecast are measured against actual results, and the more accurate method, or a weighted blend, is used. The result is reconciled so the options stay consistent with the model forecast and the configuration rules, and it comes with a likely range, not just a single number. Planners see which method was used and why, and can adjust it.

Decisions in inventory target setting

That likely range drives the inventory decisions. Buffers belong at shared components and sub-assemblies, especially long-lead items, not at finished goods. Their size comes from the forecast range and supplier variability, set against a target service level, so the trade-off between service and working capital is visible. A medical device maker, for example, can order common long-lead imaging components ahead, based on forecasted usage across every configuration that uses them.

What gets passed to supply planning

Demand planning hands supply planning a forecast of each model by period, the expected option and component mix including customer and project specifics, known CRM projects with their configurations, and inventory targets for buffered components. As configured orders arrive, they consume this forecast, and the configurations actually sold refresh the attach rates for the next cycle. Because Eyelit SIOP from Eyelit Technologies runs demand, inventory and supply on one model, nothing is re-keyed along the way.

Part 3 of the series looks at supply planning: how that demand becomes capacity and material plans without a part number for every configuration, and how S&OP hands off to S&OE as orders arrive.