AI that understands your profitability, cost and planning models
Oracle EPM already ships with its own AI. Claude, by Anthropic, is the layer between your Oracle EPM and your team: it prepares the data, models the business and communicates the impact, in an agentic, automated and controlled way. Here is what each one does and how we implement it.
Request your Oracle EPM + Claude assessmentTake the 3-minute Radar ↓Where does AI live in your Oracle EPM?
Oracle's native AI
Features that already come with Oracle EPM: predictions, alerts and agents.
Best when you want quick results with your current license.
Risk: nobody configures it or uses it.
Claude connected to your model
Claude is the layer between your Oracle EPM and your team. With Claude Code and MCP connectors on the Oracle EPM REST API, it prepares the data, models the business and communicates the impact, with controls at every step.
Best when you want to automate model administration.
Risk: permissions; that is why every action is logged.
Claude in your own cloud
Claude on Microsoft Foundry, Amazon Bedrock or Google Cloud Vertex AI, inside your account and under your controls.
Best for banking, insurance and healthcare.
Risk: security configuration and consumption cost.
Oracle EPM's native AI
| Module | AI feature | What it does |
|---|---|---|
| Enterprise Profitability and Cost Management | IPM Insights | Monitors your data and alerts you to anomalies, bias and correlations. |
| Enterprise Profitability and Cost Management | PCM Agent | An agent inside the profitability and cost module. |
| Planning | Predictive Planning · Auto Predict · Advanced Predictions | Forecasts built from your history. |
| Planning | EPM Assistant | An agent to work the plan in natural language. |
Source: Oracle Cloud EPM documentation, November 2026.
An agentic layer between Oracle EPM and your team
Prepares the data
Loads, validates and ties out to the source before every run.
Models the business
Translates drivers, rules and scenarios into your profitability and planning model.
Communicates the impact
Turns Oracle results and alerts into the narrative leadership needs.
Control: everything runs with an audit trail. Changes to the production model require a person's approval.
The Sáasil™ framework for AI-enabled management accounting
Decision-to-Value™ connects people, strategy, data, processes, applications and governance to deliver measurable performance and business value.
Strategy and enterprise value
Decision and performance assessment · roadmap · business case
Talent
Training · organization design · workforce
D2V™ processes
End-to-end flows · activity analysis · information needs · ROIC optimization
Agentic EPM orchestration
Intelligent agents · workflow orchestration · embedded analytics · natural-language interaction · proactive alerts and actions
Domain models
Revenue · costing · investment
AI model governance
Model choice · controls · traceability of every result
EPM applications
Core · profitability · planning · narrative reporting · embedded AI
Data and analytics
Integration · data platform · knowledge base
Infrastructure and security
Cloud · security · scalability · APIs · extensibility
The three columns follow the capability logic McKinsey sets out in Rewired (Lamarre et al., Wiley, 2023; 2nd ed. 2026). Decision-to-Value™ adds the trace from each capability to the decision it enables and the P&L line it moves.
Good uses and bad uses
Use it to
- Profitability: check that every allocation rule ties out to the source total before running the POV.
- Profitability: prioritize IPM Insights alerts and separate data errors from real business changes.
- Cost: draft the first version of the driver dictionary from your cost center catalog.
- Planning: review the assumptions of the predictive forecast before the committee meets.
- Administration: document an inherited model nobody ever documented.
Don't use it like this
- Accepting a number without tying it out to the source. Instead: automatic tie-out and a human review of the total.
- Letting an agent change production rules without an audit trail. Instead: Claude proposes, a person approves.
- Pasting customer data into a personal account. Instead: Claude Team or Enterprise, or Claude in your own cloud.
- Using the predictive forecast as the official budget. Instead: document assumptions and drivers.
Three prompts to learn by doing
These prompts don't hand you a finished answer. They put Claude in the role of tutor: it explains, asks you questions, reviews what you propose and leaves you with the judgment to keep going on your own. Use them inside a Claude Project with your model documentation uploaded, and don't paste real customer data into a personal account.
Profitability and cost · EPCMLearn to choose allocation drivers
What you will learn: the criteria that make a driver defensible.
I want to learn how to design allocation drivers in Oracle EPCM, not have you hand them to me. Here is my cost center catalog [PASTE]. Let's work like this: 1. Explain in a few lines what makes a good driver (cause and effect, measurable, the data actually exists). 2. Pick 3 cost centers from my list and ask me to propose the driver for each one. 3. Review my proposal: tell me what works, what you would change and why. Don't give me the answer before I try. 4. When we finish the 3, summarize the criteria I applied so I can do the rest on my own. If anything doesn't come from what I gave you, flag it as ASSUMPTION.
Then check: that every driver you choose exists in your source systems.
Planning · PlanningLearn to move your budget to drivers
What you will learn: to find what really moves each line of the plan.
I want to learn how to turn my [department] budget into a driver-based model in Oracle Planning. Here is its current structure [PASTE]. Act as my tutor: 1. Before proposing anything, ask me what moves each line. 2. With my answers, help me write the line → driver formula. 3. If my logic is weak, don't fix it yourself: ask me a question so I fix it. 4. At the end, give me 3 control questions to test the model against last year's actuals.
Then check: run the model with last year's data and compare it to actuals.
Model administrationLearn to read an inherited model
What you will learn: how your model is built, piece by piece.
I want to deeply understand an Oracle EPM model I inherited. Here are the dimensions, main members, rules and jobs [PASTE]. Walk me through it like a class: 1. First explain the overall logic in 5 lines. 2. Let's go through one dimension and one rule at a time: you ask me what I think it does, I answer, and you correct me. 3. Every piece I understand, we write into the documentation together. 4. At the end, give me a 5-question quiz to confirm I understand the model.
Then check: review the documentation with whoever runs the model.
Get the other 9 prompts
The other 9 teach you to review allocation rules, explain a margin variance, design cost to serve, prepare the POV, read IPM Insights alerts, challenge the predictive forecast, build scenarios, write the plan narrative and keep the change log.
You get the PDF right away. No spam: we only write to you about Oracle EPM + Claude.
Non-negotiable rules
Every number ties out
Everything the AI produces or touches is tied out to its source.
Every action leaves a trail
Who, what, when and with what approval.
Periodic review
Of the model and of the AI, with our Model Governance Review.
How ready is your Oracle EPM operation to work with AI?
12 questions, about 3 minutes. You instantly see your radar across 6 fronts and where to start. It is the starting point of the assessment.
The radar is an indicative self-assessment; it does not store your answers or send them to anyone.
From the radar to your first use case in production
Oracle EPM + Claude Radar
On your own, in 3 minutes. You see which fronts are strong and where to start.
Oracle EPM + Claude assessment
A working session with Pedro San Martín and a review of your application (dimensions, rules, jobs and Oracle AI enabled). No write access to your environment.
First use case in production
A single process, with Claude operating under an audit trail and human approval from day one.
- Your radar, validated against the real model.
- The Oracle AI you already pay for and aren't using.
- 3 Claude use cases prioritized by value and risk.
- A 90-day plan for the first use case.
Duration: 2 weeks · For: teams running Oracle EPCM or Planning in production.
What teams ask before they start
Do you need write access to our Oracle EPM environment?
Not for the assessment: we review your application without write access. In production, Claude works through the Oracle EPM REST API with the permissions you grant, and every change to the model is logged and approved by a person.
Where does our data go?
We follow your data policy. We work with anonymized data or on Claude for business (Team or Enterprise), under Anthropic's commercial terms. For regulated data, Claude runs in your own cloud (Microsoft Foundry, Amazon Bedrock or Google Cloud Vertex AI), inside your account and your controls.
Does Claude replace Oracle's native AI?
No. IPM Insights, Predictive Planning and Oracle's agents keep doing their job. Claude reads what they produce, explains it and acts on it within your controls.
Do we need technical skills to start?
No. You can start this week with a Claude Project and the prompts on this page. Claude Code and MCP connectors come in when we automate the operation of the model.
How is this different from a chatbot?
A chatbot answers questions. Here Claude works on your model: it loads and ties out data, reviews rules and drafts the narrative, with an audit trail and human approval on every production change.
Which Oracle EPM modules does this cover?
Enterprise Profitability and Cost Management and Planning, including FreeForm. That is where we operate Claude today.

Decision-to-Value
An Implementation Guide for Finance and EPM Professionals. The method behind this page, chapter by chapter: understand the business, communicate with impact, model the business, prepare relevant data and lead the change.
Content by Asher & Company. Claude and Anthropic are trademarks of Anthropic PBC; Oracle is a registered trademark of Oracle Corporation.

