Forecasting New Spare Parts Before Product Launch

Starting Point: New Devices, New Spare Parts, but No History

A typical supply-chain use case is the planning of spare parts for new appliances. Existing SAP-supported forecasting procedures often work well for established items. The challenge starts with new devices.

New devices create new spare parts. For these parts, there is initially no consumption history. Classical forecasting methods therefore deliver hardly any reliable demand values. At the same time, parts often need to be ordered early because replenishment times can run for several months.

If you plan too low, you risk out-of-stock situations immediately after product launch. If you plan too high, you tie up capital in the wrong parts.

“The real difficulty was not the forecast itself, but the question of which existing part would behave like a new one.”

The Real Challenge: Understanding Similarity Instead of History

The central question was not simply:

“How do we forecast new spare parts?”

But rather:

“Which existing part is likely to behave similarly to a new part, even though this new part has never existed before?”

To answer this, the solution must do more than apply classical item-number logic. It needs to understand:

• in which existing devices similar parts were installed
• which role a part plays within the bill of materials
• which names, descriptions, and product structures point to functional similarity
• which historical consumption patterns similar parts have shown
• how large the installed base of the devices was in which these parts appeared
• which failure rate can be derived from that

That is exactly what makes this a specialized forecasting case.

Solution: A Specialized Forecasting Layer with BOM Matching and Proxy Logic

One possible solution path is a dedicated forecasting layer alongside SAP that can later be integrated.

At its core, the system analyzes the bills of materials of new appliances and compares them with existing devices. It combines structural information from the BOM with semantic matching based on names, descriptions, product groups, and hierarchies.

The goal is to identify a historical proxy part or a group of similar parts. Their historical consumption serves as the reference for new spare parts.

On this basis, an initial failure rate is calculated:

• historical consumption of similar parts
• relative to the number of devices, or the demand for devices, in which these parts were installed

This rate is then transferred to the expected sales of the new appliances. In this way, an initial forecast is created even before any demand history of its own exists.

What Planners See: No Black Box, but Reasoned Recommendations

The solution does not simply deliver a number. It shows planners:

• which existing part was proposed as the proxy
• why this part was classified as similar
• which historical consumption data was taken into account
• which failure rate was derived
• which initial quantity is recommended
• which uncertainty is associated with the recommendation

The final decision intentionally remains with the planner. AI provides evidence, comparison logic, and suggestions. The human retains responsibility.

Why This Is Not an Excel Case

The domain idea is plausible, but it is not operationally scalable by hand.

Bills of materials, item metadata, appliance histories, spare-part consumption, product descriptions, hierarchies, and new sales plans all need to be connected. With tens of thousands of spare parts, many device variants, and multiple planners, this becomes a search and evaluation problem that Excel cannot solve robustly, transparently, and repeatedly.

That is where the value emerges: a good domain idea turns into a productive, data-driven decision process.

Result: Making Cold Start Plannable

This kind of use case shows how the special case of new spare parts can be modeled in a methodologically sound way:

• Better initial planning: New spare parts are no longer estimated only broadly or manually, but derived from historical similarities and failure rates.
• Lower out-of-stock risk: Especially with long lead times, a reliable planning basis emerges much earlier.
• More transparency for planners: Proxy logic, data basis, and derived rates remain understandable.
• Scalable process: What was previously only conceivable manually in isolated cases can be applied across many new devices and spare parts.
• System-open start: The solution initially runs separately alongside SAP and can later be integrated.

Conclusion

This use case shows by example what supplemental forecasting logic is really about: not replacing a standard system, but modeling exactly those economically critical cases cleanly that cannot be solved reliably within the standard setup.

When new products, long lead times, and missing history come together, a different logic is needed than a standard forecast. That is exactly the kind of logic that can be built with a specialized, transparent layer.