PlantOps BI

ABC Classification

ABC sorts parts by how much value flows through them, so planners spend their attention where the money is. On the demo data, 57 of 200 parts carry 80.2% of 2025 usage. The question worth asking next is how much stock each class sits on.

2025 usage value (order quantity times unit cost) by class, against stock on 2025-12-31. Days of supply is stock over a day's usage.

ClassPartsUsageShare of usageStockShare of stockDays of supply
A57$48.12M80.2%$7.18M83.2%54
B42$8.92M14.9%$942K10.9%39
C101$2.93M4.9%$508K5.9%63

What It Shows

The 101 C parts move 4.9% of the value and hold 5.9% of the stock, 63 days of supply against 54 for A and 39 for B. In real plants the usual cause is cheap parts bought in large lots and reviewed rarely. They are not where most of the cash is, but they take shelf space and counting effort, and a tighter reorder point on C parts is often a quick win.

The cut-offs are conventions, not laws. 80% and 95% are common; some plants use 70 and 90. Pick them once, write them on the report, and don't move them to make a review look better.

The DAX

The class is a measure, not a column, so it follows the date filter: a part can be A this year and B last year. The ranking clears the product filter so a part keeps its class when someone filters to one family. Two parts with the same usage get the same class, because each is compared with the parts strictly larger than it.

-- A calculated table with one row per class, for the rows of a visual.
-- "No usage" catches parts with stock but no orders in the period.
ABC = DATATABLE ( "Class", STRING, { { "A" }, { "B" }, { "C" }, { "No usage" } } )
Usage Value =
SUMX ( fact_orders, fact_orders[qty] * fact_orders[unit_cost] )

-- One part at a time: a part's class from the share of usage held by parts
-- larger than it. Parts are ranked against all parts in the period, whatever
-- product slicer is on. HASONEVALUE (not ISINSCOPE) so it also works when
-- another measure iterates parts. On a family or total row it is blank.
ABC Class =
IF (
    HASONEVALUE ( dim_product[part_id] ),
    VAR ThisPart = [Usage Value]
    VAR Parts =
        CALCULATETABLE (
            ADDCOLUMNS ( VALUES ( dim_product[part_id] ), "@v", [Usage Value] ),
            REMOVEFILTERS ( dim_product )
        )
    VAR Total = SUMX ( Parts, [@v] )
    VAR Before = SUMX ( FILTER ( Parts, [@v] > ThisPart ), [@v] )
    VAR Share = DIVIDE ( Before, Total )
    RETURN
        IF (
            ISBLANK ( ThisPart ),
            "No usage",
            SWITCH ( TRUE (), Share < 0.8, "A", Share < 0.95, "B", "C" )
        )
)

-- Any measure split by class: iterate the classes in context, so ABC[Class]
-- works on rows, in a slicer (one class or several) and on the total row.
Usage by ABC =
SUMX (
    VALUES ( ABC[Class] ),
    VAR Cls = ABC[Class]
    RETURN
        SUMX (
            FILTER ( VALUES ( dim_product[part_id] ), [ABC Class] = Cls ),
            [Usage Value]
        )
)

Swap [Usage Value] in the last measure for any other measure, such as the closing stock from Inventory snapshots, to split it by class; the No usage row then holds stock nobody ordered, often the most interesting row on the report. Every part in the demo data has orders in each year, so that row is empty here. The class is recomputed for every part in every cell, which is fine for a few thousand parts; for tens of thousands, store the class for a fixed period as a calculated column instead. The site computes the classes with the same strictly-larger rule, and its tests check that the classes cover every part and all usage, and that A starts below 80%.

Inventory turns · Days on hand · About the demo dataset