The AI Already Running in Your Oracle Fusion Tenant
An honest feature-by-feature inventory across all six Fusion modules — what is actually live, what genuinely works, what still needs conditions to deliver, and what requires governance before you enable it.
Find your reading path
Business & Functional
Finance Directors, HR Leads, Operations Managers
You want to know what AI Oracle has actually switched on in your modules — and whether it is working. Skip the technical detail and focus on what each feature does, how mature it is, and what your team needs to know before relying on its outputs.
Start here →
IT & Configuration
System Admins, Technical Leads, Fusion Config Specialists
You need the prerequisites, feature flags, OCI dependencies, and governance controls for each feature before you enable anything. The framework section defines the four assessment dimensions; the module sections give you the detail per feature.
Start here →
Consultant & Architect
Solutions Architects, Implementation Consultants
This is your reference inventory for every architecture review and client conversation. The scorecard gives you the full 26-feature matrix; the prerequisite section tells you the deployment readiness conditions; the module sections contain the implementation detail.
Start here →
Not sure? Start from the top — the article covers every module and works for all audiences.
00 From Map to Territory
Article 1 gave you the vocabulary — the taxonomy of AI types, the three branches, the architecture stack. This article is the inventory. What has Oracle actually shipped? What is running in your Fusion tenant right now? What genuinely delivers, and what still needs specific conditions to be useful?
Oracle’s release notes describe features in marketing language. This article uses the taxonomy from Article 1 to classify each feature precisely, assess it honestly, and tell you what your organisation needs before enabling it. The goal is not to be negative about Oracle’s AI capabilities — they are genuinely substantial. The goal is to give practitioners an accurate picture rather than a promotional one.
“Oracle’s AI capabilities in 25D and 26A are genuinely substantial. This article’s job is not to be sceptical — it is to be accurate. There is a difference.”
One important framing note: maturity ratings here reflect implementation readiness, not Oracle’s release status. A feature can be Generally Available in Oracle’s terms and still require 18 months of clean historical data before it works reliably in your tenant. GA is an engineering milestone. Maturity — as used here — is a practitioner judgement. The two are often different.
01 How to Read This Inventory
Every feature in this article is assessed against the same four dimensions. Understanding them once makes the rest of the article a reference you can return to.
Branch
Which AI type from Article 1’s taxonomy. A = Perception, B = Analytical/Predictive, C = Generative/Agentic. AB or BC = combined.
Maturity
Independent assessment — not Oracle’s GA/Preview label.
Mature — validated, predictable, real-world evidence
Live — functional, limited track record
Emerging — right architecture, expect rough edges
Data Threshold
What your tenant needs before this feature works reliably. The single most common reason Fusion AI features underperform is data immaturity — not configuration errors.
Governance Flag
What oversight the feature requires.
Advisory — human decides action
Review required — human approves output
Override required — agent takes actions
02 Financials AI
Financials
6 AI features assessed · Oracle Fusion 25D / 26A
Governance note — Financials
1 feature in this module requires human review of outputs before use. Build the review workflow before enabling the feature.
AP Invoice Automation
Document Intelligence + Supervised ML
Mature
Perception + Analytical
Reads supplier invoices in any format, extracts structured fields, matches to POs, routes exceptions. High-confidence matches post automatically.
Data Threshold
6+ months invoice history for reliable match scoring. Supplier data quality is critical — poor supplier master data is the most common cause of low match rates.
Governance
Advisory — Auto-posting applies to high-confidence matches only. Exception queue requires human review. Standard AP audit controls apply.
Smart Close / GL Anomaly Detection
Supervised ML
Mature
Analytical AI
Scores every journal entry against historical patterns at period close. Flags statistical anomalies before the close is finalised — catching errors, duplicates, and unusual postings that would otherwise pass undetected.
Data Threshold
13+ months period-end data for seasonal pattern learning. Clean chart of accounts required. Features degrade significantly on tenants with frequent COA changes.
Governance
Advisory — Anomaly flags are advisory. Human review required before close action. Strong fit with existing close checklists.
Cash Flow Forecasting
Predictive Analytics
Live
Analytical AI
Forecasts short and medium-term cash position using AR/AP patterns, historical payment behaviour, and open transaction data. Outputs daily and weekly cash position projections.
Data Threshold
18+ months AR/AP history. Clean customer payment data. Payables terms data populated. Accuracy drops significantly on tenants with inconsistent payment term coding.
Governance
Advisory — Forecast feeds treasury planning. Human sign-off on assumptions required before use in board or investor reporting.
Collections AI
Prescriptive AI
Live
Analytical AI
Prioritises AR collection activities by scoring customer accounts on payment risk, relationship value, and days outstanding. Recommends contact strategy and sequence.
Data Threshold
Customer payment history, segment and credit data populated. 12+ months AR history. Effectiveness is significantly lower without customer segmentation data.
Governance
Advisory — Prescriptive recommendations should be reviewed before automated outreach or credit actions. Particularly important where customer relationships are strategically sensitive.
Period-End Commentary
Generative AI
Live
Generative / Agentic
Generates draft narrative commentary explaining variances between actuals and budget/forecast. Produces structured finance language from structured data inputs — variance tables become readable narrative.
Data Threshold
Structured actuals vs budget data in EPM or Financials. Variance data clean and labelled. Quality of generated commentary directly reflects quality of underlying variance data.
Governance
Review required
REVIEW REQUIRED before any distribution. Finance team must review generated commentary before it reaches management, board, or external parties. Not a final output — it is a draft.
Oracle 26A release notes ↗
Duplicate Invoice Detection
Supervised ML
Mature
Analytical AI
Identifies potential duplicate invoices using fuzzy matching on supplier, amount, date, and reference fields. Flags before payment run — catching duplicates that exact-match controls miss.
Data Threshold
Low — operates on current invoice data without significant historical training requirement. One of the easiest Fusion AI features to enable with minimal data prerequisites.
Governance
Advisory — Flags only — no auto-rejection by default. Human review of flagged items. High-confidence duplicate pairs can be configured for auto-hold.
03 Procurement & SCM AI
Procurement & SCM
4 AI features assessed · Oracle Fusion 25D / 26A
Governance note — Procurement & SCM
1 feature in this module requires human review of outputs before use. Build the review workflow before enabling the feature.
Supplier Risk Scoring
Supervised ML
Live
Analytical AI
Scores suppliers on financial stability, delivery performance, and compliance risk. Can integrate with external data providers for enriched risk signals.
Data Threshold
Supplier transaction history in Oracle. External data feeds (D&B, etc.) significantly improve accuracy but are optional. Scores are thinner for suppliers with limited transaction history.
Governance
Advisory — Risk scores are advisory. Procurement decisions should not rely solely on AI score. Particularly important for single-source or strategic suppliers where relationship context matters.
Contract Intelligence
NLP / Document AI
Live
Perception AI
Extracts key clauses, obligations, dates, and risk indicators from contracts. Classifies contract type and flags non-standard terms against a configured baseline.
Data Threshold
Contracts uploaded and indexed in Oracle Contract Management. Extraction quality varies with contract format complexity — heavily formatted or scanned PDFs reduce accuracy.
Governance
Review required
REVIEW REQUIRED. Clause extraction requires legal review. AI identifies and extracts — it does not interpret. Legal and compliance sign-off mandatory before acting on extracted terms.
Oracle 26A release notes ↗
Demand Sensing
Predictive Analytics
Live
Analytical AI
Forecasts short-term demand using real-time signals — POS data, order patterns, external signals — to adjust supply plans dynamically within the planning cycle.
Data Threshold
24+ months demand history. Clean item master. Integration with POS or external demand signals significantly improves accuracy. Thin SKU history weakens reliability.
Governance
Advisory — Forecast feeds supply plan. Human review of assumptions before plan approval. Particularly important for high-value or long-lead-time items.
Spend Classification
Unsupervised ML
Live
Analytical AI
Automatically classifies spend transactions against a taxonomy (UNSPSC or custom). Identifies maverick spend and reclassification opportunities across the purchase-to-pay cycle.
Data Threshold
Clean supplier and PO data. Custom taxonomy configuration improves accuracy significantly. Initial classification run always requires data review — do not deploy without a validation cycle.
Governance
Advisory — Auto-classification is applied; periodic accuracy review recommended. Misclassification risk is higher in complex or overlapping category structures.
04 HCM & Payroll AI
HCM & Payroll
5 AI features assessed · Oracle Fusion 25D / 26A
Governance note — HCM & Payroll
2 features in this module require human review of outputs before use. Build the review workflow before enabling the feature.
Attrition Prediction
Supervised ML
Mature
Analytical AI
Scores employees on flight risk using historical attrition patterns, engagement signals, compensation relativities, tenure, and role factors. Outputs per-employee risk scores and cohort-level analysis.
Data Threshold
2+ years employee history. 500+ employee population for reliable model training. Engagement survey data significantly improves accuracy. Models trained on smaller populations carry higher false-positive rates.
Governance
Advisory — ⚠ Scores are advisory only. Must not be used as the sole basis for employment decisions. Use policy should be reviewed by HR, legal, and ethics stakeholders before deployment. Disparate impact risk requires monitoring.
Candidate Scoring
Supervised ML + NLP (CV parsing)
Live
Perception + Analytical
Parses CVs and applications using NLP, then scores candidates against job requirements using ML trained on historical hiring outcomes. Surfaces ranked shortlists.
Data Threshold
Historical hiring data with outcome labels (hired/not hired, subsequent performance). Quality of training data directly determines bias risk — poor historical hiring practices are amplified, not corrected, by the model.
Governance
Review required
⚠ REVIEW REQUIRED. AI scoring in hiring carries inherent bias risk. Human decision required for all shortlisting and rejection. Bias audit of training data strongly recommended before deployment. Legal review of use policy required in regulated jurisdictions.
Oracle 26A release notes ↗
Skills Intelligence
Unsupervised ML + NLP (skill inference)
Live
Perception + Analytical
Infers skills from job descriptions, CVs, and work history. Builds a dynamic skills graph across the workforce. Identifies skill gaps and internal mobility opportunities.
Data Threshold
Job descriptions and employee profiles populated in HCM. Skills framework configured. The system infers skills not explicitly stated — the quality of inference depends heavily on profile completeness.
Governance
Advisory — Inferred skills (not directly stated by the employee) require human validation before use in talent decisions, particularly promotions or development planning.
Payroll Anomaly Detection
Supervised ML
Live
Analytical AI
Flags statistically unusual payroll elements — unexpected increases, new allowances, pattern breaks — before payroll run confirmation. Reduces post-payment correction cycles.
Data Threshold
12+ months payroll run history. Clean payroll element configuration. High false-positive rates in the first two to three months after major payroll structure changes.
Governance
Advisory — Anomaly flags are advisory. Standard payroll review and approval process applies. Not a substitute for four-eyes payroll review.
HR Document AI
Document Intelligence
Emerging
Perception AI
Extracts and classifies information from HR documents — offer letters, contracts, certifications, compliance documents — uploaded to HCM document management.
Data Threshold
Documents uploaded to HCM. Initial extraction accuracy should be validated on a representative sample before broader deployment. Accuracy varies significantly by document type and format.
Governance
Review required
REVIEW REQUIRED. Extraction accuracy must be validated before use in compliance workflows. Not suitable for unreviewed automated processing of legal or regulatory documents — particularly in jurisdictions with strict employment law requirements.
Oracle 26A release notes ↗
05 EPM AI
EPM
4 AI features assessed · Oracle Fusion 25D / 26A
Governance note — EPM
1 feature in this module requires an override protocol before enabling. These are agentic features that take autonomous actions — the governance framework must exist before the feature is switched on, not after.
Predictive Planning
Predictive Analytics
Mature
Analytical AI
Generates statistical forecasts from historical actuals within EPM planning models. Driver-based — learns relationships between drivers and financial outcomes. Runs within Planning or PBCS.
Data Threshold
3+ years actuals. Clean driver data and driver-outcome relationships defined. Planning model well-structured with stable dimensionality. Models degrade on tenants with frequent structural changes.
Governance
Advisory — Model assumptions must be documented and signed off by FP&A before use. Forecast outputs are an input to the human-led planning process, not a substitute for it.
Variance Explanation
Generative AI
Live
Generative / Agentic
Generates natural language narrative explaining variances between actuals and plan/forecast. Integrates with SmartView and the EPM web interface. Produces draft commentary finance teams can edit and distribute.
Data Threshold
Structured actuals vs budget/forecast data in EPM. Variance commentary configuration required. Output quality is directly proportional to the quality and granularity of underlying variance data.
Governance
Review required
REVIEW REQUIRED. Draft commentary must be reviewed by finance team before any distribution. Not suitable for unreviewed board or executive reporting. The model can produce confident-sounding commentary on thin or ambiguous data.
Oracle 26A release notes ↗
Agentic Planning Assistant
Agentic AI (multi-step autonomous)
Emerging
Generative / Agentic
Accepts natural language instructions and autonomously creates, populates, and versions planning scenarios across EPM model dimensions. The most architecturally significant AI feature in 25D/26A — it takes actions, not just recommendations.
Data Threshold
Well-structured EPM planning model. Clean driver hierarchy. Natural language instruction quality directly affects output quality. Requires thorough testing in a non-production environment before any live use.
Governance
Override required
⚠ OVERRIDE PROTOCOL REQUIRED. This agent takes autonomous actions across your planning model. A human checkpoint before any model submission, publication, or use in financial reporting is mandatory. Audit trail of all agent actions must be configured before enabling. Escalation path for unexpected agent behaviour must be defined.
Oracle 26A release notes ↗
Scenario Modelling AI
Prescriptive AI
Live
Analytical AI
Recommends scenario assumptions and parameter ranges based on historical volatility and driver sensitivity analysis. Surfaces the range within which drivers have historically moved, and flags assumptions that sit outside observed bounds.
Data Threshold
Driver-based EPM model. Historical actuals for sensitivity calculation. Thinner history means wider and less useful recommendation ranges.
Governance
Advisory — Scenario recommendations are advisory. FP&A team retains full ownership of scenario selection and assumptions.
06 CRM & CX AI
CRM & CX
4 AI features assessed · Oracle Fusion 25D / 26A
Governance note — CRM & CX
1 feature in this module requires an override protocol before enabling. These are agentic features that take autonomous actions — the governance framework must exist before the feature is switched on, not after.
Win Probability
Supervised ML
Mature
Analytical AI
Scores open opportunities on likelihood to close based on historical deal patterns, engagement signals, stage velocity, and competitive factors. One of the most mature AI features in the Fusion suite.
Data Threshold
12+ months closed/lost deal history. CRM data quality — stage updates, contact engagement logging — directly affects accuracy. Sales teams that do not update CRM diligently get unreliable scores.
Governance
Advisory — Scores are advisory. Sales team review is expected and appropriate. Should not replace sales judgment for complex deal strategy.
Next Best Action
Prescriptive AI
Live
Analytical AI
Recommends next sales or service action based on customer interaction history, product affinity, lifecycle stage, and propensity models. Surfaces in the sales rep interface and CX agent desktop.
Data Threshold
Customer interaction history in Oracle CX. Product catalogue data. Customer segment data. Without these, recommendations default to generic playbooks rather than personalised suggestions.
Governance
Advisory — Recommendations are suggestions. Human judgment applies to all customer-facing actions.
Service Resolution Agent
Agentic AI (multi-step autonomous)
Emerging
Generative / Agentic
Handles service cases from receipt to resolution recommendation autonomously — triages, queries knowledge base, drafts responses, escalates when confidence is low. The most mature agentic feature in the CX suite.
Data Threshold
Knowledge base populated and actively maintained. Case history in Oracle Service. Escalation path and confidence thresholds must be configured and tested before enabling in production.
Governance
Override required
⚠ OVERRIDE PROTOCOL REQUIRED. This agent takes resolution actions that affect customer relationships and may carry contractual or regulatory implications. Human escalation path must be defined, tested, and documented before go-live. Confidence threshold configuration is critical — get this wrong and the agent will handle cases it should not.
Oracle 26A release notes ↗
Sentiment Analysis
NLP
Live
Perception AI
Analyses customer interactions — emails, chat transcripts, call summaries — to score sentiment and identify at-risk relationships or escalation signals.
Data Threshold
Customer interaction data in Oracle CX. Volume of interactions affects model reliability. Low-volume accounts produce less reliable sentiment scores.
Governance
Advisory — Sentiment scores are advisory. Customer relationship decisions require human judgment. Sentiment scoring is a signal, not a verdict.
07 Projects AI
Projects
3 AI features assessed · Oracle Fusion 25D / 26A
Cost Forecasting
Predictive Analytics
Live
Analytical AI
Forecasts project cost-to-complete using historical actuals, burn rate patterns, and resource consumption trends. Outputs EAC projections that update dynamically as actuals are posted.
Data Threshold
18+ months project actuals across comparable project types. Clean project structure (WBS) required. Accuracy degrades significantly when comparing structurally different project types.
Governance
Advisory — Forecast feeds EAC. Project manager review required before EAC is submitted or used in financial reporting or client billing.
Resource Optimisation
Prescriptive AI
Live
Analytical AI
Recommends resource assignments based on skills match, availability, cost, and project demand. Identifies over and under-utilisation across the portfolio. Surfaces internal mobility opportunities.
Data Threshold
Resource profiles (skills, availability, cost rates) populated. Project demand data current. Works best with 100+ resources — smaller pools produce fewer meaningful optimisation opportunities.
Governance
Advisory — Recommendations are advisory. Resource managers retain decision authority. Particularly important where resource assignments have contractual or client-relationship implications.
Margin Prediction
Predictive Analytics
Emerging
Analytical AI
Predicts project margin at completion using early-stage signals — contract type, resource mix, scope change frequency, and comparable project outcomes. Most useful in the first 20% of project lifecycle.
Data Threshold
Integrated financials and project data required. Comparable project history essential. Early-stage feature — validate accuracy on your own project population before relying on outputs in client or investor reporting.
Governance
Advisory — Advisory only. CFO or controller review required before use in financial commitments, investor reporting, or bid pricing.
08 The Honest Scorecard
All 26 features across all six modules in one place. Filter by module, maturity, branch, or governance requirement. This is the reference view — use it before any AI readiness conversation, implementation scoping, or vendor discussion.
The maturity ratings here are independent assessments reflecting implementation readiness, not Oracle’s GA/Preview labels. A feature rated Live may be Generally Available in Oracle’s terms but still require specific data conditions. A feature rated Emerging may be technically preview or very early production.
Oracle Official Resources: For authoritative release notes, roadmap, and feature documentation across all modules, see the
Oracle Fusion Cloud documentation index ↗.
Simpl’IT Consulting is an Oracle Certified Partner. Independent analysis only — always verify feature availability in your specific tenant configuration.
Looking for “what should we prioritise”?
This scorecard answers what Oracle shipped. The AI Use Case Navigator answers what you should do with it — matching your organisation’s industry, modules, and challenges to recommended use cases. Two different questions, two different tools.
09 Before You Enable Anything
Three things every organisation needs in place before enabling AI features in Fusion — not as a bureaucratic checklist, but because their absence is the most common cause of AI feature failures we see in the field.
Data Readiness
The single biggest predictor of AI feature success in Fusion. Most features in this article have explicit data thresholds — they underperform or produce misleading outputs when those thresholds are not met.
Audit data completeness and history depth per module before enabling
Clean master data (suppliers, customers, employees, items) before AI reads it
Consistent coding and categorisation — AI learns your patterns, including bad ones
Historical data continuity — gaps and structural changes degrade models
Configuration Prerequisites
Several AI features require specific configuration steps, OCI service connections, or initial model training runs that are not automatically done at go-live.
Feature flags enabled in the correct sequence — some AI features depend on others being active
OCI AI Services connections configured where required (particularly for document AI and generative features)
Initial model training runs completed — predictive models need to be trained before they produce outputs
Integration with external data sources (D&B, POS, etc.) where relevant
Governance Placeholders
For generative and agentic features especially, the review workflow and override protocol must exist before the feature is enabled. Not after the first incorrect output surfaces in a board pack.
Define who reviews AI outputs before they are used or distributed
Document the override path for agentic features — what triggers human intervention
Establish the audit trail requirements before the first agent action is taken
For HCM AI: legal and HR review of use policy before deployment in hiring or workforce decisions
Continue your reading path
10 What Comes Next
This article is the inventory. The module articles — Articles 3 through 8 — go deep. Each one covers a single Fusion module in full: the complete feature set, configuration walkthrough, data prerequisites in detail, governance framework, and real implementation patterns from the field.
Looking ahead — Articles 19 and 20 cover the full AI governance framework for Oracle Fusion: configuration controls, audit trail requirements, override protocols, role-based access, and adoption strategy. If you are planning to enable any of the agentic features covered in this article, reading those before you configure is strongly recommended.
Oracle Fusion AI Series — 20 articles across Foundation, Module, Vertical, and Governance layers. Published by Simpl’IT Consulting, an Oracle Certified Partner. Independent analysis anchored to Oracle Fusion Cloud 25D / 26A. View the full series →