Every forecasting method you’ve read about assumes one thing: history. New product spare parts planning has none. On launch day you have a bill of materials, a sales plan that is probably optimistic, and a service network that wants to know which parts to stock by Friday.
So planners guess. Some guess high and end up writing off screens for a model that sold half its forecast. Others guess low and watch repair turnaround stretch while parts sit on a ship.
There’s a better way to guess. This guide covers the cold-start problem and the full spare parts lifecycle: launch, ramp, maturity, end of life and the last-time buy. Once a product has a few months of real consumption data, our post on machine learning for spare parts demand forecasting picks up where this one ends.
Why new product spare parts planning is a cold-start problem
A mature product gives you three things a new one doesn’t: a failure rate per part, a known installed base and a consumption pattern by region. Without them, standard time-series forecasting has nothing to work with.
What you do have is more useful than it looks. You have the service BOM, which tells you exactly which parts can be replaced in the field. You probably have similar products from earlier generations. And within weeks of launch you’ll have activations and the first warranty claims. The job is to combine those three sources, then shift weight from assumptions to real data as fast as the data allows.
This is the gap the NPI side of our Spares Planning and Forecasting product addresses. It includes an NPI Management module for new launches and supply planning driven by the service BOM (SBOM) from ODM vendors, alongside ML demand forecasting for products that already have history.
Spare parts planning across the product lifecycle
The right method changes as the product ages. Here’s how we think about each phase:
- Pre-launch: service BOM analysis plus analog product failure rates. Main risk: stocking parts that never fail, missing the ones that do.
- Launch (first 8-12 weeks): analog rates adjusted daily with early claims and activations. Main risk: an early-failure spike from a design or batch issue.
- Ramp: a blend of the analog model and actual consumption. Main risk: over-trusting early data from a small installed base.
- Maturity: ML forecasting on claims, sales and activation history. Main risk: complacency when a new failure mode appears.
- End of life: installed-base decay model and last-time buy. Main risk: buying too much for a shrinking base, or too little for legal obligations.
- Post-support: run-down of remaining stock, harvesting from returns. Main risk: holding stock long after demand has gone.
Look at the two ends of the list. Launch and end of life are where the least data is available and the most money is committed, which is why we’d put your best planners there.
Initial spares provisioning before launch
“Initial provisioning” is the term aerospace and defence planners use. In consumer electronics, people are more likely to say launch stocking or NPI spares. Same problem: what goes on the shelf before anyone has a failure record.
Start from the service BOM
The service BOM is the list of parts your repair network can actually replace. It’s usually much shorter than the manufacturing BOM. A phone might have hundreds of components but only a couple of dozen field-replaceable parts: display assembly, battery, back cover, camera modules, charging port board, main board.
Get the SBOM agreed with the ODM early, with part numbers, lead times and minimum order quantities. A classic launch gap is a part that’s field-replaceable in the service manual but was never set up as an orderable spare.
Borrow failure rates from analog products
Pick one to three earlier models that are close in design, price point and usage. Map each new part to its nearest equivalent and take that part’s failure rate as your starting point. Then adjust for what’s changed. A new display supplier, a larger battery or a first-generation hinge deserves a higher assumption than a carry-over part.
Be honest about which assumptions are weak. Flag them. Those are the parts you’ll watch hardest after launch.
Size the launch stock
A simple starting formula for each part: expected units in customers’ hands during the period, times the analog failure rate for that period, plus safety stock for the lead time. To take a purely illustrative example, if you expect 50,000 active units in the first quarter and the analog display failure rate is 0.8% per quarter, that’s 400 displays of expected demand before safety stock. The formula is crude. That’s fine, because you’ll replace it with real data within weeks.
Where you put the stock matters as much as how much you buy. Regional hubs with short replenishment to service centres usually beat spreading thin stock everywhere.
Reading the first failure signals after launch
The early weeks are when a new product is most likely to surprise you. The NIST Engineering Statistics Handbook describes the start of the bathtub curve as an early failure period with “a high but rapidly decreasing failure rate” that typically lasts several weeks to a few months. That’s exactly the window when your analog assumptions are weakest.
Your best real signal is warranty claims. Each validated claim tells you which part failed, on which model and batch, in which country. If claims are decided by rules rather than by hand, that data is available the same day, not after month-end. We explain the mechanics in automating warranty claim adjudication.
The second signal is activations. Shipments tell you how many units left the factory. Activations tell you how many are actually in use and where, which is the real denominator for any failure rate. A part with ten claims against 5,000 activations is a very different story from ten claims against 50,000.
Compare actual claim rates to your analog assumptions weekly for each SBOM part. When a part runs well above its assumption, reorder early and escalate to engineering. When it runs well below, stop replenishing and let stock draw down.
Ramp and maturity: handing over to ML forecasting
At some point, real consumption becomes more reliable than the analog model. There’s no universal date for that. It depends on volume. A high-volume phone might get there in a couple of months. A niche accessory might never get there, and stays on a simpler model for life.
A practical approach is to blend. Weight the analog forecast heavily at launch and shift weight to actual consumption as the installed base and claim count grow. Once you have enough history per part and region, hand over to ML forecasting on claims, sales and activation data.
Some products sit in between: low volume, lumpy demand, few analogs. Off-the-shelf models tend to struggle there. That’s where a custom model can help, built for sparse data with explicit uncertainty ranges. Our AI as a Service team builds that kind of model on Azure, AWS or on-premise infrastructure, typically in four to eight weeks.
End of life and the last-time buy
Spare parts end-of-life planning starts well before the product stops selling. The ODM will announce the end of production for the components in your service BOM, and you’ll get one last chance to order.
The obligations are also getting longer. In the EU, new ecodesign rules for smartphones and tablets that started applying on 20 June 2025 require manufacturers to supply key spare parts within 5-10 working days for at least 7 years after a model is no longer sold in the EU. That turns the last-time buy from a commercial choice into a compliance question for any brand selling there.
Last-time buy calculation
A workable last-time buy calculation for each part:
- Estimate the installed base still in use, year by year, over the support horizon.
- Multiply by the part’s failure rate for each year, using real consumption history.
- Add safety stock for the service level you’ve committed to.
- Subtract stock on hand and expected supply from repaired, refurbished or harvested parts.
- Round to the supplier’s minimum order quantity and check the storage and write-off cost.
Step 4 is the one most teams skip. Returned devices are a parts source, and ignoring them inflates the buy. If you want a structured process around this, IEC 62402 is the international standard for obsolescence management, and it’s a reasonable framework to borrow from.
Planning a launch across 70+ countries
Multi-country launches add a layer. Each region has different sales timing, warranty terms, service partners and import lead times. A part that’s fine in one country can be short in another simply because the ship arrived three weeks later. We covered the regional side of this in scaling global after-sales operations.
At HMD Global, our spares planning platform runs across 70+ countries. The HMD Global case study covers SBOM and supply planning with factory and vendor lead times, ML demand forecasting from claims, sales and activation data, and a service vendor portal with consumption-driven replenishment. For new products, the combination that matters is SBOM planning with real lead times plus a live claims feed. Together they let planners correct launch assumptions quickly instead of waiting for the first stock-out report.
Frequently asked questions
How do you forecast spare parts for a new product?
Start with the service BOM to list field-replaceable parts. Take failure rates from one to three similar earlier products and adjust for design changes. Multiply by the expected installed base for each period and add safety stock for lead time. After launch, replace assumptions with actual claim and activation data as quickly as volume allows.
What should an initial spare parts list for a product launch include?
It should include every field-replaceable part in the service BOM, with part number, supplier, lead time, minimum order quantity, expected failure rate and the analog product each rate came from. Add launch stock quantities by region and flag the parts with the weakest assumptions, such as new suppliers or first-generation components, for close monitoring.
How do you calculate a last-time buy for spare parts?
Estimate the installed base still in use for each year of the support period, multiply by each part’s failure rate and add safety stock. Then subtract stock on hand and parts you expect to recover from repairs, refurbishment or harvesting. Round to the supplier’s minimum order quantity and weigh the cost of excess stock against the risk of running out.
What is the difference between a service BOM and a manufacturing BOM?
A manufacturing BOM lists every component needed to build the product. A service BOM lists only the parts a repair network can replace in the field, usually as assemblies such as a display module or battery. The service BOM is shorter and is the starting point for spare parts planning, pricing and repair procedures.
How long do you need to stock spare parts after a product is discontinued?
It depends on the market and product. In the EU, ecodesign rules that started applying in June 2025 require smartphone and tablet makers to supply key spare parts for at least seven years after a model stops being sold there. Elsewhere, obligations vary, so check local consumer law and your own warranty commitments before setting the support horizon.
Plan spares for your next launch
If you have a launch coming up and no history to plan from, talk to us about planning spares for your next launch. We’ll look at your service BOM, your analog products and your claims data, and show you where the launch assumptions are weakest. Once the product has real history, ML-based spare parts forecasting takes over from there.
