The paid Advanced Analytics option died in December 2019, but its detection footprint did not. This is what actually triggers a machine learning finding in 23ai and how to close it before an auditor does.
The paid Advanced Analytics option died in December 2019, but its detection footprint did not. This is what actually triggers a machine learning finding in 23ai and how to close it before an auditor does.
If a vendor rep, a reseller, or an internal DBA tells you that using Oracle Machine Learning in 23ai requires you to buy the Advanced Analytics option, they are working from documentation that is more than six years out of date. Effective December 5, 2019, Oracle folded the Advanced Analytics option (its formal name was Oracle Advanced Analytics, or OAA) and the Spatial and Graph option into the base Database license. Oracle's own licensing change FAQ states that customers receive Machine Learning, Spatial, and Graph "without any increase to the price of Oracle Database licenses or DBCS subscriptions." That rule is still current. An Oracle-adjacent technical reference updated on January 21, 2026 confirms plainly that "Oracle Machine Learning (formerly Advanced Analytics) no longer requires an extra cost license."
So the license cost of in-database ML in 23ai is zero, on top of whatever edition you already own. That is the good news. The problem, and the reason this page exists, is that the feature-usage tracking Oracle built while OAA was a paid option never went away. The internal name Oracle records is still "Data Mining." That single string is what an auditor's script pulls, and it is what generates confusion, false findings, and unnecessary panic. Your risk in 23ai is not a license bill for ML itself. Your risk is that ML usage rides on Enterprise Edition, and EE is where the real money sits. Read on for where the leverage and the exposure actually are.
The license cost of in-database ML in 23ai is zero. The exposure is the Enterprise Edition it rides on, and the audit script that still calls it 'Data Mining.'
The naming history is not trivia. It is the reason your audit-defense position depends on knowing exactly what to search for. When OAA was a paid Enterprise Edition option, Oracle's own product blogging (Patrick Davies, September 2018) stated bluntly that "Advanced Analytics includes only one feature: Data Mining." Oracle Data Mining was the SQL and PL/SQL algorithm engine. The option also historically bundled Oracle R Enterprise, the enterprise wrapper around the open-source R language. Mike Dietrich of Oracle summarized it in December 2019 as "The Option Formerly Known as Oracle Advanced Analytics (OAA). Today we call it Machine Learning."
Here is the practical consequence. The legacy Oracle Data Mining documentation still frames the feature as "a component of the Oracle Advanced Analytics option of Oracle Database Enterprise Edition." Anyone reading a pre-2019 manual, and Oracle has thousands of them still live, will conclude they owe money. They do not. In 25 years of negotiating with this vendor, we have seen this exact misreading used, sometimes unintentionally, to seed doubt during a soft audit. When a Software Investment Advisor or LMS auditor points at a Data Mining usage flag, the correct response is to cite the December 5, 2019 licensing change FAQ, not to concede.
Included at no extra license cost does not mean available everywhere in the same way. Oracle's licensing change FAQ is explicit: "All features are included to the extent that the underlying database supports them. For example, Oracle Database SE2 does not support parallel operations, so Machine Learning, Graph or Spatial features will execute serially on SE2." On Standard Edition 2 the ML algorithms run, but single-threaded, which for large training sets is functionally punishing. On some non-EE configurations the operation fails outright with ORA-40107: "Operation Requires Data Mining Option To Be Installed."
That ORA-40107 error is a landmine in negotiations because it reads like a licensing message. It is not. It is an install-state and edition-support message. Do not let it be used to argue that you were running an unlicensed option. The real cost question is upstream: are you on Enterprise Edition at $47,500 per processor list (SKU A90611, with 22 percent annual support at $10,450 per license per year), or on SE2? ML is free on both, but if a team justified moving to EE partly to "get" ML, that decision was made on a false premise and should be revisited. Review the boundary in our companion piece on where the 23ai Free versus Enterprise Edition feature line sits before you accept any EE upsell tied to analytics.
This is the section to bookmark. Oracle detects ML feature use through the data dictionary, and the tracked name is still "Data Mining." The core audit query is simple: SELECT * FROM DBA_FEATURE_USAGE_STATISTICS WHERE NAME IN ('Data Mining');. When an LMS engagement runs, the comprehensive weapon is the script options_packs_usage_statistics.sql, documented in My Oracle Support Doc ID 1317265.1. That script rolls up every option and pack across your estate, and Data Mining appears in its output whether or not you consider yourself an ML user.
The columns that decide your position are in the DBA_FEATURE_USAGE_STATISTICS view. DETECTED_USAGES counts how many times the system has flagged the feature. CURRENTLY_USED shows TRUE or FALSE for the last check. FIRST_USAGE_DATE and LAST_USAGE_DATE bound the window. There is also a binary-level control: the chopt tool can add or remove the Advanced Analytics component from an Oracle home, though it requires all instances from that home to be shut down. For the broader mechanics, our guide on how to check Oracle license information the three reliable ways walks through the LMS scripts, Enterprise Manager, and the CSI portal, and where each one misleads.
| Detection artifact | What it shows | Buyer interpretation in 23ai |
|---|---|---|
| NAME = 'Data Mining' in DBA_FEATURE_USAGE_STATISTICS | In-database ML feature usage flag | No license cost since Dec 5, 2019. Not a finding on its own. |
| DETECTED_USAGES / CURRENTLY_USED | Frequency and last-check status | Use to prove non-use or bounded historical use, not liability. |
| ORA-40107 | Operation requires Data Mining option installed | Edition/install-state message, not a licensing breach. |
| V$OPTION 'Data Mining' | Install and enable state at the binary | Confirms component presence; controllable via chopt. |
| options_packs_usage_statistics.sql | Estate-wide option and pack rollup | The auditor's primary tool. Run it yourself first. |
A Data Mining flag in DBA_FEATURE_USAGE_STATISTICS is not a finding. Since December 5, 2019 it is a free feature. Do not concede on the string alone.
The 23ai release meaningfully expanded what you get at no incremental license cost. Oracle documents an array of in-database algorithms covering classification, clustering, ranking, time series, and anomaly detection. The May 2, 2024 database blog listed "significant enhancements to data mining and machine learning functionality, including improvements to XGBoost and the Generalized Linear Model algorithm to enhance predictive accuracy." In practice you exercise these through DBMS_DATA_MINING (the OML4SQL package) and, for embeddings, DBMS_VECTOR.
The most consequential 23ai addition is ONNX model loading. Oracle Machine Learning now lets you load transformer models into the database as first-class objects, in ONNX format, for use with AI Vector Search. The operative privilege in the real world is GRANT CREATE MINING MODEL, which is granted directly to a user and also travels inside the new DB_DEVELOPER_ROLE. When you scope what a schema can do, that privilege is the entitlement you are actually managing, not a separate license line item.
In-database ML and AI Vector Search are adjacent and often bundled in the same conversation, but they are distinct features. Both are now included in Enterprise Edition at no additional license cost, and AI Vector Search is available at no additional charge in Oracle Database cloud services such as Autonomous Database. That inclusion is a genuine value lever for existing EE customers and a structural threat to the standalone vector database market (Pinecone, Weaviate, Milvus, Chroma, Qdrant), as our pillar analysis notes.
The danger is the reverse of a bill: it is scope creep. Because ONNX model loading uses DBMS_VECTOR and touches both feature families, teams sometimes assume that heavy vector or ML workloads justify additional Oracle spend. They do not, at the feature-license level. Confirm the boundary directly in is Oracle 23ai AI Vector Search included in Enterprise Edition or an extra, and read the whole entitlement map in what actually triggers a bill for 23ai vector and AI features. Neither ML nor vector search is where Oracle collects. Cores and edition are.
Oracle Machine Learning for Python (OML4Py) is the Python API for the full ML workflow against an Oracle AI Database or Autonomous AI Database. Buyers regularly confuse an installation requirement with a licensing requirement, and Oracle's own documentation feeds that confusion. On-premises, you must manually install Python, the required libraries, the OML4Py server components in the database, and the OML4Py client. That is operational work, not a separate SKU. In Autonomous Database, OML4Py is pre-installed and a Python interpreter ships with OML Notebooks.
The same pattern applied historically to Oracle R Enterprise, Spatial Studio, and Property Graph In-Memory Analytics (PGX): the 2019 FAQ noted these "will still require separate installation." File this correctly. If someone raises a separate-install component as evidence of an unlicensed option, the answer is that installation and licensing are different axes. None of these Python or R paths carries an incremental license charge on top of your database edition.
The negotiation reality is straightforward. Because ML is free and rides on Enterprise Edition, the only cost lever that matters is core counting. EE at $47,500 per processor list, multiplied by the applicable core factor in Oracle's Core Factor Table, is where any ML-adjacent conversation should be redirected. If an ML initiative is being used to justify new EE processors, challenge whether the workload genuinely needs EE at all, or whether it needs the compute rather than the edition. Our Oracle Database EE options pricing and audit trap breakdown has the per-processor and core-factor math to run that check.
In summary: in-database machine learning in Oracle 23ai is a free capability that Oracle still tracks under an old, paid-sounding name. Treat the detection footprint as a documentation exercise, not a liability. Push every dollar of the conversation upstream to edition and core counting, which is the only place Oracle actually earns revenue from your analytics ambitions. That is where your negotiation should live.
No. Effective December 5, 2019, Oracle folded Advanced Analytics (OAA) into the base database license. In-database ML, including the algorithms and OML packages in 23ai, carries no incremental license cost on top of your edition. Any source telling you otherwise is using pre-2019 documentation.
Oracle never renamed the internal feature-usage tracking. The DBA_FEATURE_USAGE_STATISTICS view and the options_packs_usage_statistics.sql script still record ML activity under the name 'Data Mining.' A flag there is not a license finding. Since December 5, 2019 it reflects a free, included feature.
It is included on both, but SE2 does not support parallel operations, so ML runs single-threaded and some operations can fail with ORA-40107. That error is an edition and install-state message, not a licensing breach. For serious ML workloads, EE performance matters, but EE is where the real cost sits at $47,500 per processor list.
No, they are distinct features that Oracle sometimes discusses together, especially around ONNX model loading. Both are included in Enterprise Edition at no extra license cost, and vector search is free in Autonomous Database. Do not let heavy vector or ML workloads be used to justify additional Oracle feature spend.
No. On-premises OML4Py and R components require manual installation, which is operational work, not a separate license. In Autonomous Database, OML4Py is pre-installed. Installation requirements and licensing requirements are different axes; neither Python nor R path adds a charge on top of your database edition.
Run options_packs_usage_statistics.sql yourself first, using MOS Doc ID 1317265.1, so you know your Data Mining and every other flag before an auditor arrives. Keep the December 2019 OAA/OSG licensing change FAQ in your evidence pack, and redirect all cost discussion to Enterprise Edition processors and core counting, which is the only real exposure.
Oracle Database 23ai bundles options you may never deploy. The buyer side guide to edition right sizing, option pruning, and AI Vector Search licensing.
Gated with a work email on the download page. No sales follow up you did not ask for.
Get the White Paper →
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. Renewal signals, audit moves, and the levers that work. No vendor spin.
