Optimization at renewal time is a confession that there was no program
Estates cleaned up as a one time project drift back toward the old waste level within four to six quarters, because the estate keeps moving after the cleanup ends. The estates that stay cheap work a quarterly rhythm: measure in month one, decide in month two, execute in month three, with the levers pulled in cost order. This is the operating manual for that rhythm, and the point of it is that the baseline is never more than 90 days old when a commercial event starts.
Prepared by Redress Compliance · August 10, 2026 · Oracle advisory. Based on 25 to 35 Oracle cost and license reviews, 2024 to 2025.
Executive summary
Optimized once is not optimized, because five mechanisms put the waste back and none of them announces itself. Tooling enables options, so Diagnostics Pack, Tuning Pack, and Advanced Compression switch on through routine administration and become licensable retroactively.
Virtualization boundaries grow as clusters gain hosts. Hardware refresh changes core counts and core factors, so the old position no longer maps. People churn leaves leavers inside Named User Plus counts.
And renewals roll the support base forward whether or not the licenses underneath are still deployed. In our file, estates optimized as a project drifted back within six quarters.
Reharvesting comes first, because an idle license you already own is cheaper than any discount. The program keeps a reallocation register: every decommission event releases entitlement back to a pool, and every new project checks the pool before a purchase request is raised.
Middleware is a rich vein, since migration programmes free processor licenses that would otherwise be repurchased. This lever runs before the metric work because it changes the numbers every downstream lever operates on, which is why the sequence is not interchangeable.
The metric has a crossover, and below it the wrong metric is pure overspend. Database Enterprise Edition carries a minimum of 25 Named User Plus per processor, so a population below that floor prices better on Named User Plus while a large or uncountable population prices better per processor.
A reporting database with 60 known users on two processors carrying processor licenses is on the wrong metric.
Count real users quarterly, remembering that Named User Plus covers humans and devices accessing the database directly or through middleware, and that multiplexing does not reduce the count.
Support is where shelfware bleeds, and three policy mechanics decide whether a cut survives. Shelfware costs roughly 22 percent of net license fee per year in support.
Terminating support on genuinely unused licenses is allowed, but matching service levels mean every license in a set must carry the same level, repricing recalculates the remainder and can claw back most of the saving, and reinstatement is expensive enough that only confident cuts should be made.
Run the support review every quarter three, at least 120 days before the anniversary, because a decision made inside the notice window is not a decision, it is an acceptance.
The quarterly rhythm, and the annual overlay
| Month | Work | Artifact produced | What it feeds |
|---|---|---|---|
| One: measure | Refresh discovery, reconcile entitlements to deployment | Current baseline, gap register | Everything downstream |
| Two: decide | Rank gaps by cost and risk, pick levers, get sign off | Decision log with owners and dates | The execution month |
| Three: execute | Reharvest, file support changes, remediate options | Savings ledger entry | The CFO report and the next cycle |
Each quarter carries one theme on top of the standard cycle, so every corner of the estate gets a deep pass once a year. Quarter one takes options and packs, scanning every database for feature usage and closing or licensing each finding.
Quarter two takes infrastructure and metrics, reviewing cluster boundaries, core counts, and the processor against Named User Plus position per product. Quarter three takes the support base, reviewed at least 120 days before the anniversary while notice periods can still be met.
Quarter four takes contracts and demand, mapping next year's projects against entitlements so growth lands on licenses you already own. The deep cost material sits in the total cost optimisation guide and the support stack in the support cost strategies.
The four levers, in cost order
- Reharvest idle licenses before buying anything. Take entitlement off decommissioned or shrunken workloads into a pool, and make every new project check the pool before raising a purchase request. The middleware case sits in the migration business case.
- Right size the metric per product. Test the 25 Named User Plus per processor crossover, count real users quarterly, and document the decision and the evidence, dated, in the baseline file.
- Retire shelfware and the support attached to it, after modelling matching service levels and the repricing effect, because a cut that triggers a reprice on the remaining set can return almost nothing.
- Prune the support base on a calendar, not on impulse, every quarter three and at least 120 days out, since notice periods turn a late review into an automatic renewal.
- Report five numbers every quarter: baseline age in days, support cost per deployed processor, shelfware ratio, open option exposures, and reharvest rate. A program that reports five numbers outlives every reorganisation.
The Oracle cost optimisation playbook
The lever by lever detail underneath the rhythm: options and packs, virtualization boundaries, metric crossovers, and the support stack.
Get the white paper →The five level maturity ladder
| Level | Name | What happens | The tell |
|---|---|---|---|
| 0 | Reactive | Nothing until an audit letter or renewal quote forces it | Baseline older than a year, or none |
| 1 | Project | A one time cleanup after a shock, then decay | Savings fade within six quarters |
| 2 | Scheduled | Quarterly cycle with a named owner and a decision log | Baseline never older than 90 days |
| 3 | Instrumented | Continuous discovery, monthly option scans, automated alerts | Gaps surface in weeks, not at renewal |
| 4 | Commercial | Program outputs open every negotiation and audit response | No renewal signed without a fresh position |
Standing the program up from a cold start means running the first quarter as a founding cycle: heavier on measurement, lighter on execution, and honest about what is unknown. The goal of cycle one is not savings, it is a trustworthy picture and a working calendar.
Month one assembles the paper, pulling every ordering document, amendment.
And support renewal from the last ten years and building the entitlement register from the documents rather than from memory, which is where most estates discover entitlements nobody knew they owned and metrics nobody remembered agreeing to.
Month two measures the metal, refreshing discovery across servers, virtual machines, and cloud instances, running the database feature usage scripts, and reconciling the two views until every delta is explained, because an unexplained delta in cycle one becomes an audit finding in year two.
Month three holds the first decision meeting, ranking every gap by annual cost and by audit risk, assigning an owner and a date to each, and publishing the first savings ledger even if it holds three lines, since the ledger is the program's credibility instrument with the CFO.
The roster is small and named rather than a committee: a program owner from software asset management at roughly a day a week, a database delegate who answers for what is deployed, an infrastructure delegate who owns the virtualization map, a procurement owner who holds the notice date calendar.
And an executive sponsor who unblocks month two.
The reference points are Oracle's own documents, including the technology price list.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
What we saw across Oracle optimization engagements, 2024 to 2025
The common advice is to run optimization when the renewal lands, because that is when the money is visible.
We disagree, because a renewal window optimization starts after Oracle has already built its own picture of your estate, leaves no time to execute the levers with long lead times, and hands the clock to the vendor:
How long a one time cleanup held before the estate returned toward its old waste level, because nothing kept measuring after the project closed.
The threshold a scheduled program holds, which is what guarantees a commercial event never opens with the vendor holding fresher data than you.
Across the 25 to 35 Oracle cost and license reviews we ran, the leaks were the same three every time: database options and packs switched on by tooling with nobody logging the licence consequence, support still billing on licences that had not been deployed for years.
And Named User Plus counts frozen at a headcount the company no longer had.
The estates that saved most did the unglamorous work in the quiet quarters, reharvesting in month three, remediating options in quarter one, filing support notice 120 days early, so that by the time the quote arrived the negotiation was about their numbers rather than Oracle's.
The quarterly cycle is deliberately boring, and the drama it prevents is the audit finding priced at list. The wider practice sits in the Oracle library.
Your first five moves
- Build the entitlement register from the paper, not from memory, pulling every ordering document, amendment, and support renewal from the last ten years in month one of the founding cycle.
- Reconcile discovery against entitlement until every delta is explained, because an unexplained delta in cycle one becomes an audit finding in year two.
- Pull the levers in cost order: reharvest the idle licences first, then test the 25 Named User Plus per processor crossover per product, then shelfware, then the support base.
- Model matching service levels and the reprice before filing any support change, and run that review every quarter three, at least 120 days before the anniversary.
- Report the five numbers every quarter and keep the baseline under 90 days old, so no renewal or audit ever opens with Oracle holding a fresher picture than you do. The Oracle practice runs the rhythm with you.
Frequently asked questions
Why does Oracle optimization fail as a one time project?
Because the estate keeps moving after the cleanup ends. New projects add servers, platform teams enable licensable features, people join and leave, and the support renewal auto renews the old base.
In our file estates optimized as a project drifted back toward their old waste level within four to six quarters, and the leaks were the same three every time.
What does the quarterly rhythm actually look like?
One quarter, three jobs. Month one measures: refresh discovery and reconcile entitlements to deployment, producing a current baseline and a gap register. Month two decides: rank gaps by cost and risk, pick levers, get sign off, producing a decision log with owners and dates.
Month three executes and publishes a savings ledger entry that feeds the CFO report and the next cycle.
What is the Named User Plus crossover?
Database Enterprise Edition carries a minimum of 25 Named User Plus per processor, so a population below that floor prices better on Named User Plus while a large or uncountable population prices better per processor.
Test the crossover per product, count real users quarterly, and remember that multiplexing through middleware does not reduce the Named User Plus count.
Can you terminate support on unused Oracle licences?
Yes, but three mechanics decide whether the saving survives. Matching service levels require every licence in a set to carry the same support level, so partial drops pull the whole set in. Repricing recalculates support on the remainder and can claw back most of the benefit.
And reinstatement after a lapse is expensive, so only cut what you are confident stays cut.
Why does the support review run 120 days before the anniversary?
Because notice periods make a late review worthless. A support decision made inside the notice window is not a decision, it is an acceptance of the renewal.
Running the review every quarter three, at least 120 days out, is what leaves time to model the reprice, obtain sign off, and file the change while the option still exists.
How do you grade an optimization program's maturity?
On a five level ladder. Level 0 is reactive with no current baseline. Level 1 is a one time project whose savings fade within six quarters. Level 2 is a scheduled quarterly cycle with a named owner and a baseline never older than 90 days. Level 3 adds continuous discovery and monthly option scans.
Level 4 is commercial, where no renewal is signed without a fresh position.
Which five numbers should the program report?
Baseline age in days, which is the program's pulse and should never exceed 90. Support cost per deployed processor, the truest unit cost of the estate. Shelfware ratio, supported but undeployed entitlement as a share of the support bill. Open option exposures awaiting remediation.
And reharvest rate, the share of new demand met from the pool instead of a purchase.
Should optimization wait until the renewal?
No. A renewal window optimization starts after Oracle has already built its own picture of your estate, leaves no time for the levers with long lead times, and hands the clock to the vendor.
The estates that saved most did the work in the quiet quarters, so that when the quote arrived the negotiation ran on their numbers rather than Oracle's.
