Cut the Line Item, Lose the Baseline

The saving nobody could find

A plant was shipping 53,456 corner boards a year, four to a pallet across 13,364 pallets, at roughly a dollar apiece. The recommendation was clean: they did not need the corner boards. Stop buying them. On paper that is a fifty-thousand-dollar line walking out the door, the kind of quick win that gets a sourcing project signed off in a single call.

Then the project tried to prove it, and there was nothing to prove it against. You cannot count the corner boards you no longer buy. The only surviving reference was pallet volume, and the plant collected that passively, a "please fill this out" spreadsheet that came back filtered, historical, and impossible to reconcile against a growth number the supplier could not defend. The saving was almost certainly real. It was also, functionally, unverifiable. The change had deleted its own scoreboard.

Why the cut erases the proof

Every sourcing decision quietly assumes a baseline: the thing you were doing before, measured, sitting on a report you can pull next quarter to see the delta. Reduce a resin gauge and the resin invoice still exists at a lower number. Renegotiate a film price and the price per pound is right there on the PO. The baseline survives the change, so the saving is a subtraction anyone can run.

Deleting a component breaks that assumption. When you remove the corner board entirely, you do not lower a number, you remove the row. There is no lighter invoice to compare, because there is no invoice. To reconstruct the saving you have to rebuild a counterfactual: how many pallets would have shipped, times four, times a dollar. Every term in that product is now an estimate, and estimates are what get argued, not booked. The cleaner the deletion, the weaker the proof.

There is a second cost hiding in the same move. A spec attribute is often doing quiet work nobody priced. Colored board is not just more expensive board, it is damage visibility on the line, a fast visual sort that a plain board removes. Corner board is not just a dollar, it is edge protection that shows up, when it is gone, as a few more crushed cartons downstream. Strip the attribute to bank the unit-cost line and its function migrates into scrap, claims, and rework, cost centers that never point back at the sourcing change that caused them. So the one number you can see, the saving, is unverifiable, and the one number that actually moves, downstream damage, is unattributed. That is a bad trade to make blind.

What to do before you delete a line

The fix is not to stop deleting components. Deletions are frequently the best money in the building. The fix is to refuse to approve one until the measurement is designed alongside the change, not after.

Name the field. Before sign-off, state the exact report and column where this saving will appear next quarter, and who owns pulling it. If the honest answer is "we will estimate it from pallet counts," you do not have a baseline, you have a future argument. Stand up the counting mechanism, an active feed from the plant's own system rather than a passive survey, before the change, while the old spec is still running and the count is still real.

Lock the baseline before you touch it. Compare the corner-board scramble to how a premium outdoor-furnishings brand ran its corrugated program. Before anyone opened a bid or moved a spec, the team sent a data request to the incumbent to confirm every dimension and every current cost across the buy. Establish the reference while the incumbent still owns it, then run the change against a number that already exists on paper. When the savings land, there is a confirmed pre-number to subtract from, not a reconstructed one.

Price the silent function. Ask what the attribute was doing besides costing money: damage visibility, protection, sortation, line speed. If it was doing real work, the swap is not free, and the downstream cost belongs in the same model as the unit-cost saving. Nobody put the two in the same model on the corner board, which is exactly why the enthusiasm outran the evidence.

Change one variable. The disciplined clients insist on it for a reason. On a multi-plant film program, the constraint was explicit: single-variable testing, performance preserved, one plant standardized to another's spec and nothing else touched. That is not bureaucratic caution. It is the only way the post-change number means anything. Move three things at once and you can book the saving but never say which move produced it, or which one is quietly bleeding downstream.

What a well-run swap looks like

The change ticket names one measured field, carries the pre-change twelve-month actual pulled from the system of record, and assigns a post-change owner by name. Verification lands inside one billing cycle, not a year-end reconstruction. Exactly one variable moves per test. The silent function of the removed attribute is priced and sits in the same model as the unit saving, so a swap that looks free on the invoice has to prove it is free on the floor. Under that standard a deletion is still often the best win available. It just arrives as a number instead of a story.

The scoreboard leaves with the line

The corner boards really were unnecessary, and the fifty thousand dollars was probably real. But "probably real" is what you are left holding when the cut takes the scoreboard with it, and probably-real does not survive the first hard question from a CFO. Design the measurement before you delete the line, or you will win the argument and never be able to prove it.