The most expensive Oracle mistakes happen after the ULA ends, not during it. Certifying too low, dropping licenses without modelling repricing, losing the evidence file, and the panic re ULA each quietly destroy value. Avoiding four errors protects the position you worked to certify.
By the Meridian advisory team · Ex Oracle licensing analysts · Updated June 2026
The end of a ULA feels like a finish line. It is closer to a handover. The value you certified is now an asset to protect, and the period after exit holds the mistakes that erode it. Here are the four that cost the most, and how to avoid each one.
Four mistakes do most of the damage after a ULA: certifying too conservatively, dropping licenses without modelling repricing, failing to retain the evidence file, and the panic re ULA. Each is avoidable with discipline rather than spend. Certify everything legitimate, model the net before reducing, keep the file, and license genuine growth deliberately instead of reaching for a new unlimited term.
This is the most expensive error in the cycle, and the most common, because it feels prudent. Support fees continue at the ULA level no matter what you certify, so a higher count costs nothing extra in recurring fees. Every legitimate deployment you leave out of the certification is therefore free perpetual entitlement handed back. The undercount saves nothing, because support does not fall with it, and it turns licenses you could have owned into licenses you must later purchase or defend in an audit. Certified counts often land well above the first estimate once cloud, disaster recovery, and non production deployments within the term are handled properly, a multiple that is indicative and contract dependent but consistently real. The fix is to certify every defensible deployment, supported by evidence, rather than trimming the number out of a misplaced sense of caution.
The instinct to cut unused licenses to save support is reasonable, but the saving is rarely what it appears. Matching service levels and the pricing rule that applies when licenses are reduced can lift the rate on the licenses you keep, offsetting much or all of the saving from the ones you drop. The headline figure, support removed, is not the real figure, net saving after repricing. A reduction that looks attractive on the number removed can leave you barely better off, or worse, once the remainder is repriced. The fix is to model the net on the specific support set, using your ordering documents, before committing to any reduction, and to act only where the net is genuinely worth it.
Certification produces a number, but the number is only as strong as the record behind it. The evidence file, the server lists, the tool output, and the documented methodology, is what defends the certified counts if they are questioned, and audit interest tends to rise in the first two years after exit. A file that is captured at certification and then lost in a reorganisation or a system migration cannot answer an audit later, and it cannot be convincingly recreated after the fact. The fix is cheap and entirely within your control: retain the evidence file as a deliberate, owned, findable record from the day you certify. The cost of keeping it is trivial against the cost of facing a review without it.
The most damaging mistake arrives under pressure. When deployment grows beyond the certified count and an audit surfaces the gap, the tempting escape is a new unlimited agreement that makes the finding disappear. That settlement resets the certification cycle, raises the support base for years, and trades a defined, one time problem for a far larger commitment. Genuine growth is a normal cost to plan for, not an emergency to solve by re entering the cycle you just left. The fix is to track deployment against the certified position over time, see growth as it happens, and license it deliberately on its own terms, so a gap never accumulates into a crisis that a panic re ULA seems to answer.
| Mistake | What it costs | The fix |
|---|---|---|
| Certifying too low | Free entitlement given away | Certify every defensible deployment |
| Dropping licenses blindly | Repricing erases the saving | Model the net on the support set first |
| Losing the evidence file | Undefended counts in an audit | Retain the file as an owned record |
| Panic re ULA | Reset cycle, higher support for years | License genuine growth deliberately |
Figures referenced throughout are indicative and depend on your specific contract language.
Each mistake has a deeper treatment worth reading before the decision arises. Start with why support stays flat at certification, which underpins the case against undercounting, then repricing risk when dropping licenses for the modelling that protects against the second error. Our post certification audit and cost guide is the pillar that frames the whole period after exit. When you want an independent read on whether your position is fully protected, the next step is a confidential assessment.
Certifying too low. Because support stays flat regardless of the count, every legitimate deployment left uncertified is free perpetual entitlement given away. The undercount does not save anything, since support is unchanged, and it converts owned licenses into licenses you must later buy or defend. A conservative certification is the costliest common error in the whole cycle.
They model the headline saving and ignore repricing. Matching service level and reduction pricing rules can raise the rate on the licenses kept, so the net saving is often far smaller than expected, sometimes nil. Dropping licenses can be the right move, but only after the net effect is modelled on the specific support set, not assumed from the number removed.
Track deployment against the certified count and license genuine growth deliberately as it happens. The panic re ULA occurs when an audit gap appears with no plan, and a new unlimited term is grabbed to settle it. Keeping the evidence file, monitoring growth, and treating any new ULA as a forward strategy rather than a settlement removes the panic that drives the mistake.