Spare parts inventory management: lumpy demand, supersession, and why min-max fits

Spare parts break the reorder rules that work fine for finished goods, and the breakage is structural, not a skill issue. A $40 replacement part for a $400 product sells maybe four times a year, in bursts — two in one week after a failure cluster, then nothing for four months. Any method that computes an average daily velocity on that history produces a reorder point so low it guarantees a gap right when a customer with a dead unit needs the part most. Spare parts inventory management is therefore a different game with different rules: demand is lumpy, parts get superseded by successors, and the cost of being wrong is measured in lost product sales, not just the part sale. Short answer: for a small parts catalog, a min-max system with a floor sized to worst-case failure demand beats velocity-based reordering almost every time.

This article starts with a fill-in worksheet: two inputs from your own records, and what each one does to the velocity math that's been failing you. Then it explains the three properties that make parts different and what each one does to your method. The method itself — the min-max formula, its math, and its tradeoffs — is owned by the min-max inventory guide; here we cover why it's the right tool for parts and how to set it up for a parts catalog without repeating that derivation.

The worksheet: fill in two numbers first

Pick your ten most-returned or most-requested parts and, for each, write down two numbers you can actually defend:

1. Installed base size: ______ — how many units you've sold that use this part (from fitment records, order history, or serial registrations).
2. Worst demand month in the last year: ______ — the highest number of units of this part requested in any single month, not the average.

Now hold each number against the velocity math your current reordering runs on:

• The worst-month number breaks velocity math because it isn't in the average. A part that sold 3 units in its worst month and 0 in the other eleven has an average of 0.25/month — a reorder point built on that average was dark for the entire worst month. The failure cluster is the plan, not an outlier.
• The installed-base number breaks velocity math because velocity can't see it at all. A part serving 340 units in the field and a part serving 30 units may have identical trailing sales, but their forward demand pools differ by an order of magnitude — and velocity, which only counts what sold, has no column for it.

Where the worst-month number exceeds what your current floor or reorder point would have covered, that's the SKU to fix first — the sections below explain why the standard method fails here and what to use instead.

Property one: lumpy demand

Finished-goods demand is smooth enough that averages mean something: a phone case selling 2.3/day has a real "2.3" in it. Parts demand doesn't. A replacement gasket for a camping stove doesn't sell because shoppers want gaskets — it sells because units in the field fail, and failures don't average politely.

What a failure-driven sale looks like in your order data:

• Clusters after a defect batch. A bad lot of seals goes out in month three; months five and six see a burst of replacement requests, then silence.
• Clusters after usage milestones. Wear parts (bike chains, blender blades, filter cartridges) fail on a rough usage clock, so demand arrives in waves timed to when you sold the units, not when it rains.
• Spikes after product recalls or viral repair content. One forum thread about a common fix can triple a part's annual demand in a fortnight.

The consequence: a SKU-level velocity average is close to meaningless for parts, and every downstream number built on it — reorder points, days-of-cover, forecast-driven buying lists — inherits the meaninglessness. The classic velocity-based method assumes yesterday predicts tomorrow; failure demand says the next sale could come from a unit you shipped last week or one you shipped two years ago. This is the same zero-history problem that breaks forecasting for new products, just arrived from the other end of the product lifecycle — and it's why the reorder-point method built on smooth demand (covered properly in the reorder point formula guide) quietly fails on parts.

Property two: the installed base is your demand pool

Here's the reframe that makes parts demand thinkable: a part's demand comes from the population of units you've sold that use it — the installed base — multiplied by a failure or replacement rate you can estimate from warranty claims and service history. You don't have "a slow SKU"; you have 340 units in the field with a part that fails maybe once per 40 unit-years. (Treat that failure rate as an assumption to estimate from your own claims data, not a benchmark.)

Two practical uses of the installed-base lens:

1. A part attached to a product you discontinued is on a countdown. The installed base stops growing, ages, and eventually thins out. Parts demand for a three-year-old product line is declining demand wearing a flat-sales disguise — order accordingly, and plan an end-of-life date for the part itself.
2. A part attached to a fast-growing product line is on a ramp. Every unit you ship this quarter joins the failure pool next quarter. Parts demand lags product demand by roughly the time-to-first-failure — a reason to hold a little more of the part when the product line is accelerating.

The verticals where this bites hardest — service parts at bike shops, replacement components for powered goods — live inside their own buying rhythms; the bike shop inventory guide covers the fitment and seasonal side of that world.

Property three: supersession

Parts don't just sell — they get replaced by their own successors. A redesigned gasket (part B) supersedes the original (part A); the supplier may honor A for a transition window or accept it as a return. Supersession is the parts-world version of a product line refresh, and mismanaging it creates the worst parts-inventory failure: stock of the superseded part you can't sell, and no stock of the successor part everyone now needs.

A lightweight supersession record needs only four fields per part, maintainable in your existing catalog or spreadsheet:

The transition-window discipline: when A hands off to B, sell A to zero first (it's usually unsellable once the window closes), then move its min-max floor to B — but re-size the floor, because B serves both A's residual installed base and new production. A floor copied from A to B unmodified is how stores end up with a successor part perpetually backordered during its own launch quarter.

Why min-max fits parts — and how to apply it without re-deriving it

The min-max method — reorder to a maximum level whenever stock hits a minimum — fits parts because it doesn't need smooth demand to work. The floor is sized to a service-level judgment ("never be without one floor unit"), not to an average; the ceiling caps capital tied up in a SKU that might sell four times a year. The full method, formula, and worked arithmetic live in the min-max inventory guide — what's parts-specific is the inputs:

• Set the floor from worst-case failure demand over your supplier's lead time, not average demand. If the part's cluster behavior is "up to 3 in a bad month" and lead time is 5 weeks, the floor needs to cover a cluster landing at the start of a replenishment cycle. A floor built on the 12-month average (which might be 0.6/month) fails exactly when the cluster shows up.
• Set the ceiling from capital honesty. A part costing you $9 with a floor of 4 and a ceiling of 8 ties up at most $72 — trivial. A $140 control board with the same levels ties up $1,120 in stock selling rarely; that's a ceiling you shrink deliberately, accepting a service-level tradeoff you've named.
• Review floors quarterly, not monthly. Parts demand shifts when the installed base shifts (new product launches, discontinuations, supersessions), which happens on quarters, not weeks. Monthly fiddling with parts floors is motion, not progress.

One honest limitation: min-max caps your service level at what your floor covers. For genuinely critical parts — the failure that takes a customer's whole product down and takes your support queue with it — some stores accept a deeper floor on the few SKUs where failure cost dwarfs part cost. That judgment is yours; the method just makes the tradeoff visible.

The high-value-part corner

A brief note on the tail end of the parts catalog: a handful of expensive parts (control boards, drive units, motor assemblies) behave less like parts and more like serialized assets — warranty claims, unit-level history, and theft risk all attach to the individual piece. When a part's unit cost crosses into "we track each one individually" territory, that's the point where serial number tracking earns its overhead, on five SKUs rather than the whole catalog.

Common questions

Why is spare parts demand so lumpy? Because parts sell on failure events, not shopping intent — a replacement sells when a unit in the field breaks, and failures cluster (defect batches, wear milestones, usage spikes) rather than averaging smoothly. A part can sell three units in one week and none for four months, which makes average-velocity math unreliable for parts reordering.

What is parts supersession? Supersession is when a successor part officially replaces an older one — a redesigned gasket supersedes the original. Track it in both directions ("supersedes" on the new part, "superseded by" on the old), sell through old stock during the transition window, and re-size the successor's min-max floor to serve both old units still in the field and new production.

Why use min-max for spare parts instead of reorder points? Min-max floors can be sized to worst-case failure demand over the lead time, so the method survives lumpy demand without pretending an average exists. Velocity-based reorder points assume yesterday's sales predict tomorrow's — an assumption parts history routinely violates, producing floors that fail during exactly the failure clusters they exist for.

How do I know how many spare parts to hold? Work from the installed base: units sold that use the part, times an estimated failure/replacement rate from your own warranty and service history, sized to cover worst-case demand during your supplier's lead time. Treat failure rates as estimates from your own data, review floors quarterly, and cap ceilings by the capital you're honestly willing to park in a slow SKU.

Should I keep stocking parts for discontinued products? Only while the installed base justifies it — discontinued products stop adding to the failure pool, and the existing pool ages and thins. Set an end-of-life review date for each orphaned part, keep the fitment record so you know which products feed it, and wind the floor down as the base ages rather than cutting it off while customers still need repairs.

Next step

You've already filled in the worksheet above — act on it. Where the worst-month number exceeds what your current floor would have covered, raise that floor first, using the min-max setup rather than an average that was never real. Then set a quarterly review date for the parts whose installed base is shifting (launches, discontinuations, supersessions), since that's when floors go stale.