SKU rationalization: how to decide which products to cut, keep or reprice
SKU rationalization is a regular review of every item you sell, to decide which to keep, cut, reprice or supply only to order. A SKU, or stock keeping unit, is one item as your business stocks it: one size, one colour, one pack. The review ranks SKUs on contribution and demand together, then checks what each one does for your customers before anything is cut. AI brings together data that usually sits in separate reports and keeps the ranking current. This guide covers the method, a worked example from an auto-parts distributor and how to start.
What SKU rationalization is
A SKU is a code your own business assigns. GS1 US, the standards body behind retail barcodes, explains the difference in What is a GTIN?: “While a GTIN identifies a product across the entire global marketplace, a SKU (Stock Keeping Unit) is strictly internal.” Every business decides for itself how many SKUs to carry, and SKU rationalization is the regular check on that decision.
The review ends with one decision for each SKU:
- Keep it in stock as it is.
- Reprice it, so that its margin covers what it costs to stock and serve. Price optimization covers setting the new price.
- Supply it to order, with a longer lead time and no stock. The Business Development Bank of Canada (BDC) covers this case in its guide to optimizing inventory management. For low-volume items with high carrying costs, “you can offer them on backorder or suggest a substitute product to customers.”
- Merge it with a duplicate, such as the same part bought from two suppliers under two numbers.
- Phase it out: sell down the stock, tell the customers who buy it, and offer a replacement.
Retailers call the same work assortment planning. A review chapter by Gürhan Kök, Marshall Fisher and Ramnath Vaidyanathan defines its goal as an assortment “that maximizes sales or gross margin subject to various constraints,” such as a purchase budget and shelf space. In a plant, the constraints are machine time and changeovers. In a distribution business, they are warehouse space and working capital.
Why the long tail costs more than it shows
Sales concentrate in a few items. BDC notes that “eighty percent of a company’s sales are typically generated by 20% of its products,” and MIT’s supply chain key concepts describe the same pattern as a power law. The items in the long tail carry costs that a standard margin report spreads thinly or leaves out:
- Holding cost. BDC lists inventory holding cost as “storage space, utilities, insurance, shrinkage, financing, warehouse personnel, inventory counts, obsolescence.” A slow item pays all of them for longer.
- Changeovers. Every short production run needs a setup. Plante Moran, an accounting and advisory firm, warns in The art of SKU rationalization that when costing leaves out changeover time, high-volume parts end up subsidizing the low-volume ones.
- Write-downs. Under the international standard IAS 2, “Inventories are measured at the lower of cost and net realisable value.” Stock that can no longer sell at a profit is written down, and the loss is an expense in that period.
- Complexity. More SKUs mean more pick locations, purchase orders and labels, and more chances to ship the wrong one.
Some tail items are still the reason a customer buys everything else from you. The review has to find those before it cuts anything.
How SKU rationalization is done by hand
The usual method is an ABC analysis in a spreadsheet. Items are sorted by revenue, the top sellers become A, the middle B and the tail C, and someone reviews the C list. BDC describes C items as “low-volume items with high carrying costs that you don’t need to carry.”
Revenue alone is the wrong sort key. A high-revenue item can make little or no margin, and a low-revenue item can be one of the most profitable in the catalogue. The spreadsheet also misses the facts that decide whether a cut is safe:
- Which customers buy the item, and what else they buy with it.
- Whether a substitute exists that customers will accept.
- Whether the item is new, seasonal, or tied to a contract, warranty or service obligation.
Those facts sit in different systems, so they get checked from memory, if at all. Because the review takes weeks of spreadsheet work, it often gets postponed.
What AI changes
Software can put margin, demand and customer data into one view per SKU and keep it current. People still make every decision. Here is the work it takes over.
A ranking on contribution
For each SKU, software computes the contribution after the costs that item causes: landed or production cost, freight, rebates, the holding cost of its average stock and the setup cost of each run. Margin analysis covers how. It then ranks items on contribution per unit of the scarce resource. OpenStax’s managerial accounting text gives the rule in its section on decisions when resources are constrained. Products “should be ranked based on their unit contribution margin per production restraint.” For a distributor, the constraint is usually inventory dollars or warehouse space. For a plant, it is usually machine hours.
A demand outlook for each item
A forecast shows which items are declining, which are seasonal and which are new and still growing. An item that sold little last year because it was out of stock for four months looks dead in a revenue report. A model that reads stockout periods does not count those months as zero demand. AI demand forecasting covers the forecast.
Customer ties
For each tail SKU, software lists the customers who bought it, their total contribution across everything they buy, and how often the item shared an order with other lines. A SKU that one large customer orders every month alongside 30 other lines needs a different decision from one that a walk-in buyer ordered once.
Duplicates and near-duplicates
Language models read item descriptions, specifications and supplier part numbers, and propose pairs that look like the same product under two numbers. A person confirms each merge. Merging also repairs the history, because one item’s sales are no longer split across two records.
Substitution
When an item is cut, some of its buyers switch to a similar item and some buy elsewhere. The Kök, Fisher and Vaidyanathan review lists substitution estimation as one of its subjects. Your own history holds evidence too: the weeks when an item was out of stock show whether its buyers took the alternative or went elsewhere.
A draft decision list
The output is a list with a proposed decision and the reasons behind it, for the product manager, sales and finance to review together. The software discontinues nothing on its own.
The list also totals what the decisions would free up: inventory dollars from the sell-down, warehouse locations, purchase orders a year and production hours saved on changeovers. Finance weighs that total against the contribution the cut items earned.
A worked example
An auto-parts distributor reviews one product family. The numbers are made up for this example.
| SKU | Units a year | Contribution after cost to serve | Trend and customer ties | Proposed decision |
|---|---|---|---|---|
| Brake pad set, common sedan | 9,400 | $61,000 | Stable. Many customers | Keep |
| Brake pad set, older model | 310 | $1,900 | Falling 15% a year. Bought by 12 repair shops that buy many other lines | Keep and raise the price 8% |
| Ceramic pad set, same fit as another SKU | 140 | −$600 | Flat. Bought interchangeably with the other part | Merge into the other SKU |
| Rotor for a discontinued vehicle | 25 | −$1,400 | Falling. Two small accounts | Phase out and offer it to order |
| Caliper kit, new model | 90 | $700 | Growing since its launch five months ago. Four customers so far | Keep and review in six months |
| Hardware kit | 4,800 | $3,100 | Stable. On most brake pad orders | Keep |
Two of the six change category when customers and trends are in view. The older-model pad set ranks near the bottom on revenue, but the shops that buy it also buy many other lines, so the draft raises its price instead of cutting it. The hardware kit makes little on its own and appears on most pad orders, so it stays.
The same review works in other industries. A food plant ranks flavours and pack sizes by contribution per hour of line time, because changeovers and cleaning take that time from production. A building-supply chain ranks colours and sizes store by store, because each store has a fixed amount of shelf space. A manufacturer can apply it to what it buys as well as what it sells. A fabricator that stocks many thicknesses and grades of sheet can standardize on fewer, after the bills of materials show which products use each one.
Find the items that cost more than they earn
Tell Derik which ERP holds your item and sales history and how many active SKUs you carry. He will tell you what your data can support.
Start a conversationThe data SKU rationalization needs
| Data | Where it usually lives | Used for |
|---|---|---|
| Sales lines and credit notes, 24 months or more | ERP | Volume, price and contribution per SKU |
| Item cost, setup times and scrap | ERP item cards, routings, production records | Cost per SKU, including changeovers |
| Inventory, receipts and stockout periods | ERP inventory and purchasing | Holding cost and true demand |
| Item descriptions, attributes and supplier part numbers | ERP item master, supplier catalogues | Finding duplicates and substitutes |
| Customer master and contracts | ERP, CRM, contract files | Customer ties and obligations |
| Bills of materials, where used | ERP manufacturing module | Parts that go into other products |
| Supplier minimums and lead times | Purchasing records, supplier agreements | Make-to-order and phase-out options |
Duplicate item numbers and changing units of measure distort every number in the review. AI for inventory management covers how to find and fix them, and AI for ERP covers getting the data out of the ERP you run.
How to start small
- Pick one product family with a long tail and a clear owner.
- Pull 24 months of data for it: sales, cost, stock and stockout periods.
- Compute the contribution of each SKU after the costs it causes, and tie the total back to the ledger.
- Add the demand trend and the customer ties for each SKU.
- Review the draft list with sales and finance in one meeting, and record the reason for each decision.
- Phase out over a planned period, following the steps in phasing out an item.
- Check the result after six months: what the customers of the cut items did, and what happened to the family’s total contribution.
Phasing out an item without losing the customer
A cut item still has stock, open orders, quotes and customers who expect it. A planned phase-out avoids surprises:
- Block new purchases of the item in the ERP, and leave it open for sale until the stock is gone.
- Map each cut item to its substitute, and show the substitute on the item card, the price list and the website.
- Tell the customers who bought it in the last year, with the end date and the replacement.
- Sell down the remaining stock, or return it to the supplier where the agreement allows.
- Update customer catalogues, electronic order files and open quotes that still list the old number.
- Keep the old number searchable and linked to its replacement.
A language model can draft the customer notices and find every quote, contract and catalogue that mentions the old number. A person approves each notice before it goes out.
Risks and limits
- Cutting an item can cost a customer. Check what else the buyers of each item buy before you cut it.
- Contracts and service obligations come first. Some items must stay available for warranty, safety or contract reasons, whatever their margin.
- Seasonal items look dead off-season. Use at least two years of history before you judge one.
- Allocated costs are estimates. Two reasonable allocation rules can rank the same SKU differently, so check that a decision holds under both.
- Write-downs are an accounting decision. The software lists the stock, and your accountant decides any write-down under IAS 2 or the standard you report under.
- Merges change the history. Keep a record of merged item numbers so that past orders and documents still resolve.
How ThriveAI helps
ThriveAI is an AI engineering company in Ottawa that builds private AI systems on a company’s own data, for businesses that make, move or sell physical goods. For a product range, that means ranking every SKU on contribution, demand and customer ties from the records you already keep, and drafting the cut, keep and reprice list for a named person to decide. Every answer shows where it came from.
ThriveAI’s systems read your ERP and accounting system as they are, and every connection only reads data. They are designed to keep each client’s data on its own server in Canada. You choose a model on that server or a hosted model under a written zero data retention agreement, and a hosted model may process requests outside Canada. Derik Lawlis, the founder, leads every project and stays close to the build. About ThriveAI covers the company.