LVMH carried Oracle support on licenses nobody had run for years. A utilization assessment, a shelving plan, and a termination sequence built around Oracle repricing rules cut 10.5 million euro across three years.
LVMH, the global luxury group with roughly two hundred thousand employees across more than seventy five countries, paid annual Oracle support on a license estate built up through years of acquisitions and projects. A material share of it ran nothing.
A structured utilization assessment and a careful termination sequence saved 10.5 million euro across three years. This is the method chapter of the story: how a buyer finds what is genuinely unused, and proves it well enough to act.
LVMH cut 10.5 million euro from Oracle support across three years by terminating support on licenses with no deployment evidence, shelving licenses with plausible reuse, and sequencing the terminations so Oracle technical support policies could not reprice the remainder upward.
The estate spanned Oracle Database Enterprise Edition, Real Application Clusters, and management packs across dozens of maisons and regions. Years of projects had left entitlements nobody reconciled.
Four cases in this series each own one layer of the same discipline. This page owns the evidence method; the termination mechanics, the program phasing, and the negotiation controls live with Costco, American Airlines, and Chevron respectively.
Read them in that order if you are building a program. The evidence question always comes first, because every later step inherits its quality.
Oracle support bills annually at roughly 22 percent of the net license fee, with a yearly uplift on top. A license bought for a project that ended in 2015 still invoices every year until someone acts.
The invoice does not distinguish between a production cluster and a shelf. That distinction is the buyer's job, and it is exactly the work most estates never do.
Acquisitive groups buy software locally, then consolidate infrastructure centrally. Workloads retire in the consolidation, but the contracts behind them keep renewing, because no single owner sees both the purchase history and the running estate.
In a group built through acquisitions, entitlements sit in different agreements, bought in different decades, under different names. The support renewal is the only place the whole history surfaces, and it surfaces as one number.
The assessment matched every supported entitlement against deployment evidence: Oracle Database instances, options usage, and pack telemetry. A material tranche of supported licenses had not run in years.
Buckets are not permanent. The classification was designed for annual revisit, with shelvable entitlements decaying toward terminable after repeated idle cycles and active entitlements retested at each renewal.
Classification is only as strong as the proof behind it. Each bucket carried a defined standard, and an entitlement that failed the standard stayed in the safer bucket until the proof improved.
Bucket, evidence standard, and decision rule
| Bucket | Evidence required | Decision rule |
|---|---|---|
| Active | Current discovery hit plus workload owner confirmation | Retain support, review pricing |
| Shelvable | No deployment in the window, a named owner claiming a credible reuse scenario | End support, keep the license, log the reinstatement price |
| Terminable | No deployment in the window, no owner claiming reuse, no license set dependency | Terminate license and support together |
Oracle support pricing is anchored to license sets and matched service levels, and the matching rule binds the license set, not the CSI. Terminating part of a set can let Oracle reprice the retained support, erasing the saving. Every termination batch was tested against that rule before submission.
You prove it by reconciling an entitlement register against deployment discovery, holding the result to a trailing twelve month standard, and making a named owner sign each classification. Proof is a file, not a feeling.
The burden of proof is asymmetric. Oracle needs no evidence to invoice; the buyer needs a defensible file to stop the invoice. Accepting that asymmetry early, and budgeting for it, is what separates programs that close from reviews that drift.
The register is the denominator of the whole exercise. It is assembled from ordering documents, migration paperwork, acquisition history, and the support renewal detail, then reconciled line by line against what Oracle's renewal quote actually bills.
Where the register and the quote disagreed, the quote was challenged rather than assumed correct. Renewal quotes inherit decades of clerical history, and that history is not always in the buyer's favor.
No single source is sufficient. Discovery without attestation misreads dormant systems, and attestation without discovery misses installations nobody remembers. The classifications that survive Oracle's scrutiny rest on at least two independent sources.
Twelve trailing months is the working standard, because it captures one full business cycle including year end processing. Shorter windows misclassify seasonal workloads; longer windows delay savings for little added certainty.
The window is a test, not a verdict. Anything idle for twelve months still passed through the false positive checks below before its bucket was final.
Every false positive terminated in error becomes a future reinstatement or repurchase at Oracle's pricing, not yours. The window and the owner signature exist to catch them before the letter goes out.
Less than vendors claim. The register lives comfortably in a spreadsheet with disciplined version control, and discovery can come from the tooling most infrastructure teams already run.
What matters is independence. Running Oracle's own measurement scripts hands the vendor a data set on your estate before you have interpreted it yourself, so the assessment used internal collection first and shared nothing until the position was understood.
They complicate the denominator, not the method. In VMware clusters and other soft partitioned platforms, Oracle's counting position can attach license demand to hosts where nothing Oracle ever ran, so an entitlement that looks surplus may be quietly absorbing that exposure.
The assessment therefore mapped cluster boundaries before declaring anything spare. A license is only surplus once the deployment position it might be defending has been priced.
In roughly one quarter of focused work, provided the renewal calendar allows it. The phases below reflect how the LVMH exercise was structured and how we sequence the same work elsewhere.
Assessment phases and their outputs
| Phase | Work | Output |
|---|---|---|
| Weeks 1 to 2 | Assemble the entitlement register, reconcile against the renewal quote | The denominator, with quote discrepancies flagged |
| Weeks 3 to 6 | Discovery and usage collection across the estate | Deployment evidence per entitlement |
| Weeks 7 to 8 | Owner interviews and written attestations | Signed classifications, reuse claims tested |
| Weeks 9 to 10 | False positive review, license set mapping | Final buckets, batch design |
| Weeks 11 to 12 | Decision log, notice drafting, executive approval | Termination and shelving package ready to submit |
Support changes take effect at the next renewal, so an assessment that finishes after the invoice locks in another full year of waste. Working back from the renewal date, with the notice period subtracted, gives the true start date.
At a 3.5 million euro annual reduction, each month of slippage past the renewal had a visible price. That arithmetic, presented early, is what kept the assessment resourced.
Four roles, kept small: a licensing lead who owns the register, an infrastructure lead who owns discovery, a procurement lead who owns the Oracle paper, and an executive sponsor who owns the decision. Everyone else contributes evidence on request.
Large committees stall this work. The estates that finish before the renewal are the ones where a small group holds both the data and the authority to act on it.
Three levers produced the saving: clean termination of dead entitlements, shelving of plausible reuse licenses, and renewal timing that landed the changes before the next support invoice, not after it.
The engagement in numbers
| Metric | Before | After |
|---|---|---|
| Supported entitlements | Full historical estate | Active deployments only |
| Annual support spend | Baseline | Reduced by roughly 3.5 million euro per year |
| Three year saving | n/a | 10.5 million euro |
| Repricing exposure | Untested | Zero, by sequencing |
| Audit posture | Unreconciled estate | Documented entitlement baseline |
For retained products, Oracle lifetime support stages determined what Premier Support actually bought. Where products had aged into Sustaining Support, which carries the same fee with fewer rights, the value question was asked explicitly rather than renewed by habit.
Shelving ends the support fee while preserving the perpetual license right, and its worst case is known in advance. Oracle's reinstatement formula computes from the last annual fee paid, with back support for the lapsed period on top.
That number was calculated for every shelved entitlement before the decision, so a future redeployment could be priced against a fresh purchase. The mechanics are covered in the guide to dropping Oracle support and reinstatement.
No running workload lost coverage, no service level was downgraded, and no support model switch was forced through. The 10.5 million euro came from scope: paying only for what the estate actually operates.
That framing matters internally. A program sold as "cutting support" invites operational resistance; a program sold as "ending fees on things we do not run" gets signatures.
The closing sequence was utilization evidence, license set mapping, batch design, then termination letters timed against the renewal date with the retained estate protected in writing.
The set mechanics deserve their own treatment, and they get it in the companion Costco Wholesale case. This page stays on the question that comes first: whether the evidence justifies touching the line at all. For the phasing of a comparable program, see the American Airlines case.
Every classification carried a named owner, a date, and an evidence reference in a decision log. When Oracle later probed individual terminations, the answer was a document, not a meeting.
The same log doubled as the entitlement baseline for audit defense. Estates rarely get to buy that artifact this cheaply: here it was a byproduct of the savings work.
Predictable, which is a compliment to the preparation. Renewal representatives queried the largest reductions, asked for justification, and escalated internally when the numbers held.
Because every line traced to evidence, the queries burned days rather than quarters. The pattern we see repeatedly: the vendor tests the buyer's proof once, then reprices its own effort accordingly.
Starting the register earlier. Every week spent debating whether the estate had waste was a week the evidence work was not running, and the waste percentage was never really in doubt.
The other candid note: owner attestations arrive slowly unless an executive sponsor chases them. Build that chasing into the plan rather than discovering the need in week seven.
The standard advice says run a discovery tool, export the idle list, and cancel what shows no activity. We disagree. Tool output alone misclassifies disaster recovery nodes, seasonal systems, and deliberate buffers, and it cannot see entitlements at all, so it misses the licenses that were never installed anywhere. In the reviews Fredrik Filipsson ran in 2024 and 2025, the programs that banked savings without repurchase pain treated discovery as one of four evidence sources and made a named owner sign every classification. The scan starts the argument. The register, the window, and the signature are what finish it.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
The invoice was paying for our history, not our infrastructure. Separating the two was worth ten million euro.
More Oracle cost analysis lives in the Oracle knowledge hub and the Oracle third party support guide. If a renewal is close, start with optimizing your license footprint before renewal and the guide to conducting internal Oracle license audits.
The saving came from terminating support on entitlements with no deployment evidence, shelving those with plausible reuse, and sequencing the work around Oracle repricing rules so retained support could not be repriced upward. Roughly 3.5 million euro came off the annual line.
At least two independent sources over a trailing twelve month window: infrastructure discovery, database level usage data, operational records, and a written owner attestation. A discovery scan alone is not proof, because it cannot distinguish dormant from dead.
Yes, but the order of operations matters. Matching service levels binds the license set rather than the CSI, so partial cancellations done naively can trigger repricing of what you keep. Map the sets first, then batch the terminations.
Shelving ends support on a license while preserving the license right itself. It fits entitlements with plausible future reuse, and its worst case is calculable in advance because reinstatement computes from the last annual fee paid plus back support.
In our 2024 to 2025 reviews, 15 to 30 percent of supported licenses showed no deployment evidence in the trailing twelve months. At a roughly 22 percent annual support rate, that waste compounds every year it goes unaddressed.
Not when it is evidence based. A documented utilization baseline built for the termination work doubles as audit defense. In our engagement file, estates that completed the exercise entered subsequent audits with stronger positions, not weaker ones.
Not as the first step. Internal collection keeps the data under your control while the position is still forming. Oracle's scripts produce output formatted for Oracle's process, and sharing it prematurely surrenders the interpretation.
Roughly one quarter for a large estate, run as five phases from entitlement register to signed decision log. The real constraint is the renewal calendar: finishing after the invoice locks in another year of the old number.
Every year we paid for software that retired before some of our staff joined. The assessment made the invisible line items visible.
Confidential consultation. No follow up sales call unless you ask for one.
Oracle support benchmarks, license utilization patterns, ULA signals, and the buyer side moves across the Oracle estate.
