ORDOrders & serviceStop 8 of 9
Service parts planning
Stocking spare parts when demand is sparse and lumpy but every stockout can stop a customer's machine.
Read atChange any time
A company selling tractors, lifts, aircraft engines or medical scanners also has to supply spare parts for years after the sale. Service parts look like an inventory problem from a nightmare:
- Huge variety: tens of thousands of part numbers, many for old models.
- Sparse, lumpy demand: most parts sell rarely and unpredictably, the intermittent and lumpy patterns.
- High stakes: when a customer’s machine is down waiting for a part, the cost of that downtime can far exceed the part’s price.
- Long tails: parts must be available long after the product stops being made.
So planners work differently from finished-goods planning:
- Forecast from the installed base: how many machines are in use, how old they are and how hard they work (hours or kilometres), not just past part sales.
- Use intermittent-demand methods such as Croston, SBA or TSB, and plan stock from the distribution of demand rather than an average.
- Set service levels by criticality: a part that stops the machine gets high availability, possibly stocked near customers; a cosmetic part can wait.
- Plan across echelons: a central warehouse holds slow parts for the whole network, while fast-moving or critical parts sit in regional or field stock with service engineers.
- Plan the end of life: when a supplier stops making a part, a last-time buy has to cover all future demand.
Done well, the aftermarket is often one of a manufacturer’s most profitable businesses, because service revenue continues for the life of the installed base.
Words to use
- Installed base
- The number, age and usage of products in the field that will need parts.
- Criticality
- How badly a missing part affects the customer, such as a machine down or just cosmetic.
- Last-time buy
- A final purchase of a part before its production ends, sized for all remaining demand.
Forecasting from the installed base
2,000 lifts in service. Failure rates rise with age:
| Age group | Lifts | Failures per lift per year | Expected demand |
|---|---|---|---|
| Under 3 years | 800 | 0.02 | 16 |
| 3 to 6 years | 700 | 0.05 | 35 |
| Over 6 years | 500 | 0.10 | 50 |
| Total | 2,000 | 101 a year |
As the fleet ages, demand will rise even if no lifts are added, which a forecast based only on past part sales would miss.
Central or field stock?
Demand for this board is low and random, so a Poisson distribution describes it better than a normal one. With a one-month replenishment lead time and a 95% availability target:
- One central warehouse: average demand over the lead time is about 8.4 boards; covering 95% of months takes about 13 boards.
- Five regional stocks, each facing a fifth of the demand (about 1.7 boards a month): each needs 4 boards for 95%, so 20 in total.
Pooling saves 7 boards, but a lift waiting for a board from the central warehouse stays down longer. For critical parts the answer is often a mix: a small field stock for the fastest-needed items and a central pool for the rest.
Service levels by criticality
| Criticality | Example | Target |
|---|---|---|
| Machine down, safety risk | Lift brake part | Very high availability, stock near customers |
| Machine down, no safety risk | Controller board | High availability, regional or fast central supply |
| Degraded performance | Fan, sensor | Moderate, central stock |
| Cosmetic | Panel trim | Low, order on demand |
Sizing a last-time buy
When a supplier announces the end of a part, one final order must cover all future needs. If demand is 101 a year now and expected to fall 15% a year over the 8 years of remaining support, expected future demand is about 101 × 0.85 + 101 × 0.85² + … over 8 years ≈ 416 boards. Adding a buffer for uncertainty, often 10–20%, gives roughly 460–500. Too few means stranded customers; too many means scrap at the end.
Repairable parts and multi-echelon models
Many expensive parts (engine modules, circuit boards) are repairable: a failed unit goes to a repair shop and back into stock. The question becomes how many spares to hold at each base and depot, given repair times and transport. Sherbrooke’s METRIC model (1968), developed for US Air Force spare parts, set out the multi-echelon approach still used today: depots and bases are planned together, and the expected number of machines waiting for parts is the key measure.
Optimize availability, not part fill rates
Customers care whether their machine runs, not whether each part number hit 95%. Availability-based planning asks: for a given budget, which mix of spares maximizes machine uptime? A marginal analysis answers it: repeatedly add the spare that gives the largest reduction in expected downtime per rupee, until the budget runs out. Cheap parts that often stop machines get stocked deep; expensive parts that rarely fail are pooled centrally.
Uptime contracts
Many manufacturers sell service as a guarantee: a contracted uptime (say, 98%) or response time, sometimes paid per hour of use. These performance-based contracts shift the risk of downtime to the manufacturer, which then has strong reasons to plan spares, technicians and preventive maintenance well. Cohen, Agrawal and Agrawal (2006) describe how leading companies treat the aftermarket as a business, with service levels designed by customer segment.
Predictive signals
Machines with sensors report hours run, temperatures, vibration and error codes. That data supports better forecasts (which parts are likely to fail soon, where) and better maintenance (replacing a part before it fails, at a planned visit). It also feeds the installed-base forecast with real usage instead of assumptions.
New supply options
For old products whose parts are no longer made, alternatives to the last-time buy include repair and refurbishment of returned parts, redesigned replacement parts, and, for some plastic and metal components, additive manufacturing (3D printing) on demand, which can avoid holding decades of stock.
Interview kit
Practise it out loud
Answer each question yourself before opening the model answer.
Service parts have huge variety, intermittent or lumpy demand and very high stockout costs. I’d forecast from the installed base and usage, use intermittent-demand methods like SBA or TSB, set service levels by criticality, and position stock across echelons, with slow parts central and critical fast parts close to customers. End-of-life parts need last-time-buy planning.
How is service parts planning different from finished-goods planning?
Demand is driven by failures in the installed base, not by consumer buying, so it’s sparse and lumpy. There are many more SKUs, the cost of a stockout is often the customer’s downtime rather than a lost sale, and parts must be supported for many years, including after production ends.
Where would you stock a critical but slow-moving part?
Usually centrally, to pool demand across the network, combined with fast shipping, unless the downtime cost and required response time are so high that regional or field stock is justified. Multi-echelon planning weighs those trade-offs part by part.
Connected stops
Where this idea feeds in, or gets its inputs from.
Sources and further reading
- Cohen, M. A., Agrawal, N. & Agrawal, V. (2006). Winning in the aftermarket. Harvard Business Review, 84(5), 129–138.
- Syntetos, A. A., Boylan, J. E. & Croston, J. D. (2005). On the categorization of demand patterns. Journal of the Operational Research Society, 56(5), 495–503.
- Sherbrooke, C. C. (1968). METRIC: A multi-echelon technique for recoverable item control. Operations Research, 16(1), 122–141.