Oracle
Costco Wholesale. 4.2 million dollars off Oracle support.
Support renewed on licenses nobody ran. The utilization assessment found the gap, the license set math protected it, and the program banked it.
Support renewed on licenses nobody ran. The utilization assessment found the gap, the license set math protected it, and the program banked it.
Costco Wholesale cut 4.2 million dollars from Oracle support by terminating unused licenses in the right order, with license set rules and repricing traps managed before any cancellation notice went in. This page owns the mechanics: what a license set is, why it and not the CSI controls the outcome, and how the surgery is done.
Costco reviewed the support line because it had grown by inertia: support renewed annually on licenses bought across years of projects, while the deployed estate consolidated underneath. Support tracked purchase history, not the running footprint.
Oracle support renews each year at about 22 percent of the net license fee under the technical support policies, so every unused entitlement still on support is pure carrying cost.
How you build that evidence to a standard Oracle will respect is its own discipline, and it is the spine of the companion LVMH case. This page assumes the evidence exists and concentrates on what happens next: the surgery.
The assessment found a material block of support spend with no deployment behind it. Consolidated data centers and retired projects had left licenses idle while their support lines kept renewing.
Findings and actions by support category
| Category | Finding | Action |
|---|---|---|
| Idle database options | Licensed and supported, never installed | Terminate support after license set check |
| Retired project licenses | Deployment decommissioned years prior | Terminate in grouped license sets |
| Oversized processor counts | Hardware consolidation cut core needs | Resize at the next support renewal |
| Active core estate | Deployed and current | Retain support unchanged |
Findings were confirmed with system owners before classification, so the termination list carried operational signatures rather than tool output alone. That step cost two weeks and prevented every category of internal dispute later.
Because the matching service levels rule in the Oracle technical support policies document requires every license within a license set to carry the same support level, and the repricing rules let Oracle recalculate retained support after a partial reduction. Naive cancellations walk straight into both.
A license set is Oracle's grouping of a program's licenses together with the licenses that depend on them, and it is the unit on which support decisions actually operate. The CSI is an administrative container for billing; the set is the licensing boundary that Oracle's policies enforce.
Buyers who negotiate by CSI discover this distinction at the worst possible moment: after the termination notice, when Oracle recalculates a fee the buyer believed was settled. Learning the set map before acting is what this case exists to teach.
Take a hypothetical set of 100 processor licenses paying a contracted support fee. Terminate 40 of them naively and the surviving 60 do not keep paying 60 percent of the old fee; Oracle reprices them as a fresh support calculation at current rates.
Depending on the contract history, the recalculated fee on 60 licenses can approach the old fee on 100. That is how a termination that looks like a 40 percent saving on paper can net close to zero, and it is why every batch was modeled before notice.
From four sources read together: ordering documents, migration and conversion paperwork, the support renewal quote, and the support portal's own line detail. Each source is incomplete alone; the map is the reconciliation of all four.
Assign it to one named owner with licensing depth. Committees produce set maps with gaps, and a gap in this map is precisely where a repricing surprise enters.
Reducing the quantity within a set, as with the oversized processor counts, is a partial reduction, and partial reduction is exactly what triggers the repricing recalculation. Those lines were therefore handled as negotiated resizes at renewal, with the recalculated support fee agreed in writing before commitment.
Terminating a whole set avoids that recalculation entirely, which is why whole set exits led the program and resizes followed at their own pace.
License migrations, metric conversions, and acquisition consolidations over the years change which entitlements belong together. The set map therefore had to be built from the ordering documents and migration paperwork, not assumed from the current support renewal layout.
This is unglamorous archival work, and it is the single highest leverage step in the program. A wrong set map does not fail loudly; it fails as a surprise invoice a year later.
The dangerous candidates are the ones with invisible dependencies: entitlements whose paper or technical linkage to the retained estate only surfaces after the notice is served. Four patterns account for most of the accidents we see.
Deceptive candidates and the safe move for each
| Candidate | The hidden linkage | The safe move |
|---|---|---|
| Database option beside a live database | Support levels must match across linked programs on the estate | Confirm the option was never installed, then terminate the license outright rather than lapsing its support |
| Named User Plus counts that look surplus | Per processor user minimums still apply to the retained servers | Recount the minimums after the proposed cut before serving notice |
| Processor licenses freed by consolidation | Core factor arithmetic on the new hardware may still need them | Redo the core count on the target platform first |
| Entitlements named in certification or migration papers | Prior agreements may reference them as the basis of current rights | Trace the paper chain before touching the line |
Nothing terminates on the strength of a usage report alone. Every candidate passed a paper review and a counting review as well, because the report answers whether the license runs, not whether the estate still needs the right.
The program grouped terminations by license set, modeled the repricing effect of each group before notice, and timed notices against renewal dates. Nothing was cancelled until the net saving of the group survived the repricing math.
Precision, mostly. Each notice identified the complete license sets leaving support, referenced the governing agreements, took effect at the renewal date, and requested written confirmation that pricing on the surviving lines was unchanged.
That last request is the quiet masterstroke available to any buyer. Confirmation in writing before the effective date converts a policy argument you might have next year into a commercial fact you hold now.
Support contracts renew automatically unless notice lands inside the contractual window, and a missed window means twelve more months at the full rate. The program worked backward from each renewal date and treated the notice deadline, not the renewal itself, as the milestone.
Three things: which agreement governs each set, what the notice and auto renewal clauses require, and that the letters made no statements Oracle could later read as compliance admissions. Ten minutes of contract reading per set, against millions in exposure, is the cheapest insurance in the program.
Typically one to two renewal cycles, because each batch can only take effect at its own renewal date. The modeling is weeks of work; the calendar is what stretches the program, and rushing it by ignoring notice windows is how savings get postponed a full year.
Reinstatement is expensive by design, and its formula starts from the last annual fee paid, with charges for the lapsed period and an uplift per the lifetime support policy framework. The program computed that number for every candidate set before deciding.
Where reinstatement exposure felt uncomfortable, the entitlement stayed on support another cycle. The full mechanics are covered in our walkthrough of dropping Oracle support and reinstatement.
The perpetual license right survives; what stops is access to updates, patches, and service requests. That differs from Sustaining Support, which continues charging the same fee while delivering fewer rights, and from termination of the license itself, which ends both fee and right.
Terminated support on a still held license is therefore reversible at a price, which is exactly why the reinstatement math was run first. Nothing in the program relied on a door that could not be reopened.
The program banked 4.2 million dollars in support savings with no compliance exposure and no repricing surprise on retained lines. The support base now matches the deployed estate, and the annual review keeps it that way.
The number understates the value. A verified set map, a signed utilization baseline, and a clean invoice history change the tone of every later Oracle conversation, from renewals to audits.
An annual utilization pass before each renewal, run against the same set map the program built. The map is an asset now; maintaining it costs days per year and keeps every future termination decision one modeling exercise away.
New purchases update the map on arrival. That single habit prevents the next decade from rebuilding the pile this program just cleared.
Not when it is executed cleanly. Oracle treats well documented terminations as routine commercial hygiene; what damages the relationship is a compliance gap, and the utilization assessment confirmed there was none.
Costco's number is the smallest of the four cases, and its lesson is the most portable: mechanics are what make any of the larger numbers real. The phased program that scales this work is the American Airlines case; the contract controls that protect it long term are the Chevron case.
The standard advice says never drop Oracle support because reinstatement penalties and repricing make it a one way door that always costs more later. We disagree. In the 30 to 40 Oracle licensing and support reviews Fredrik Filipsson worked in 2024 and 2025, estates that ran license set analysis before terminating kept 70 to 100 percent of modeled savings, and reinstatement was needed in almost none of them because the terminated entitlements were genuinely dead. The one way door argument protects Oracle's renewal base, not the buyer. The buyer side move is to model repricing per license set, terminate in clean groups, and accept reinstatement exposure only where redeployment is implausible.
Three numbers from the advisory file put boundaries on what set surgery is worth.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
Six steps make up the surgery checklist, and the order is the protection. Each one exists because skipping it has a documented cost somewhere in an advisory file.
White Paper · Oracle
When third party support is the right call for Oracle Database, Apps, and Middleware. Read it free.
Costco Wholesale saved 4.2 million dollars by terminating support on unused Oracle licenses, grouped by license set so retained support lines did not reprice upward. The saving recurs, because a cancelled line stops billing in every subsequent year as well.
It requires every license within a license set to carry the same support level. The rule operates on the set, never on the CSI, so dropping support on part of a set has consequences for the whole of it.
Yes, routinely. The CSI is a billing and administration container, while the set is the licensing boundary Oracle's policies enforce, and years of migrations can also spread one set across several CSIs. Terminations planned on the CSI map alone inherit that confusion.
Repricing is Oracle recalculating support on the licenses you keep after a partial reduction, at the pricing then in effect. Grouped correctly, terminations avoid it entirely; grouped badly, it can consume the whole saving.
From the last annual fee paid, with charges covering the lapsed period plus an uplift under Oracle's policies. Because the formula is known, reinstatement exposure can be computed per set before any termination decision is made.
Yes, when three conditions hold: verified zero deployment, a modeled license set position, and notice served inside the contractual window. Skipping any of the three is where terminations go wrong.
Shelve when a credible redeployment scenario exists, terminate when none does. Shelving keeps the perpetual right while ending the fee, with reinstatement as the priced way back; termination is final and suits entitlements no owner will claim.
That pricing on the surviving support lines is unchanged, which sets survive under which agreements, and the effective date. Written confirmation before the renewal converts every future repricing argument into a settled commercial fact.
The support optimization sequence, the license set rules, and the repricing traps.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders running the next renewal cycle.
Oracle support renews by inertia. The estates that force the support base back to the deployed estate every year never pay for nothing.
500+ enterprise clients. 11 vendor practices. Industry recognized. One conversation can change what you pay for the next three years.
One buyer side briefing a week. Pricing moves, audit signals, and the levers that work. No vendor spin.
