Every CFO I speak with is asking the same question: "We have the data, we have the planning system, so why does it still take three weeks to close the books - and another week to explain what happened?"

Stages of Enterprise AI Maturity (W XO-Claude 2) (1)

The diagram below is the architecture we built at Octane - 5 layers, one flow.

Users ask a question in plain English. It flows down through IBM watsonx Orchestrate (governed, audited, secured) into four specialist Finance Agents, each powered by Claude AI talking directly to IBM Planning Analytics TM1 via MCP Server - the Model Context Protocol. IBM Planning Analytics (TM1) is one of the most capable financial planning engines on the market. The data model is there. The governance is there. What's missing is the intelligent interface that makes all of that accessible to every finance professional, not just the TM1 specialists.

The result comes back in under two minutes. No export, no spreadsheet, no wait. The answer is rarely the technology. That is precisely what the Agentic Finance Operating Model delivers.

A Five-Layer Architecture: From Question to Decision

The model is not a chatbot bolted onto a spreadsheet. It is a governed, multi-layer architecture in which every component has a specific role - and nothing moves to the next layer without passing through the controls of the one above it.

01	USERS CFO, FP&A, Finance Analysts, Controllers, Business Leaders Anyone with a financial question - asked in plain English, no training required.
02	ORCHESTRATION IBM watsonx Orchestrate Governs every interaction. Applies security, identity, audit trails and compliance controls before routing to the right agent.
03	FINANCE AGENTS Fast Close, Variance, Forecast & CFO Insight Agents Specialist AI agents that perform the work: reconciliation, root cause analysis, scenario planning, board narratives.
04	INTELLIGENCE ENGINE Claude AI, Open AI, IBM - connected to TM1 via MCP Server The Model Context Protocol gives Claude real-time, read/write access to live TM1 cubes, dimensions and hierarchies. No export, no lag.
05	SYSTEM OF RECORD IBM Planning Analytics (TM1) The trusted financial source of truth. Budgets, forecasts, actuals, workforce plans and driver trees - all live, all governed.

How It Works: Following a Question Through the Stack

Take a real example. A CFO asks: "Why did our EBITDA miss by $4.2 million this quarter?"

Step 1 - The question enters watsonx Orchestrate. Security and identity are verified. The request is logged for audit. Governance policies are applied. Only then is the question routed to the appropriate agent.

Step 2 - The Variance Agent is activated. This specialist agent is built for exactly this task: comparing actuals against budget across dimensions, surfacing outliers, and attributing variance to specific drivers.

Step 3 - Claude AI connects to TM1 via the MCP Server. The Model Context Protocol is not a data export. It is a live connection. Claude understands the cube structure, the dimension hierarchies, and the business rules inside TM1. It runs the queries directly against the source of truth - no manual extract, no intermediate file, no risk of stale data.

Step 4 - The answer comes back up the stack. Claude synthesises the financial data into a narrative: "Personnel costs in the South region were $2.1M above budget, driven by three unplanned contractor engagements in June. Supply chain costs added $1.8M, reflecting the logistics surcharge announced in Q2." The Variance Agent formats this for the user. watsonx Orchestrate logs the interaction.

Total elapsed time: under two minutes. The same analysis would previously have required an MDX query, a manual export, an Excel model and a commentary document - consuming the better part of a working day.

The Four Finance Agents: What Each One Does

Fast Close Agent

Finance teams spend an average of three to five days on manual reconciliation tasks at period-end. The Fast Close Agent automates the full reconciliation workflow across entities: journal validation, exception flagging, and close readiness checks. Controllers move from doing the work to reviewing the output.

Reconciliation - automated cross-entity matching against TM1 actuals
Journal validation - AI-reviewed before posting, with exceptions escalated
Exception management - priority-ranked, with suggested remediation

Variance Agent

Commentary on variance is one of the most time-consuming tasks in FP&A. The Variance Agent compares actual results against budget and prior periods across all dimensions in TM1, identifies the key drivers, and generates structured narrative ready for an executive audience.

Actual vs Budget - full dimensional analysis across cost centres and entities
Root cause analysis - driver-level attribution, not just summary totals
Driver attribution - links financial movement to specific operational events

Forecast Agent

Static annual budgets are losing relevance. The Forecast Agent supports continuous, driver-based rolling forecasts - modelling multiple scenarios against live TM1 data, identifying trend inflection points, and providing recommendations for FP&A teams to evaluate.

Scenario planning - multiple "what if" models built and compared in minutes
Driver analysis - links operational assumptions to financial outcomes
Recommendations - prioritised next actions based on forecast trajectory

CFO Insight Agent

The CFO does not have time to run queries. The CFO Insight Agent translates TM1 data into board-ready language: KPI summaries, risk narratives, and structured briefings that can go directly into a board pack or executive communication.

Board narratives - generated from live TM1 data, formatted for executive audiences
KPI summaries - trend analysis across the metrics that matter most
Risk insights - early warning signals surfaced from the financial model

The Technical Foundation: Why MCP Changes Everything

Most AI integrations with financial systems work by exporting data to a flat file, passing it to an AI model, and returning a result. This approach introduces latency, data freshness risk, and a break in the governance chain.

The Model Context Protocol (MCP) eliminates all three. Claude does not read an export of your TM1 model. Claude reads your TM1 model - live.

Through the MCP Server, Claude can:

Query any dimension, hierarchy or cube in real time
Execute MDX queries on behalf of the Finance Agent
Read the rules and logic embedded in the TM1 model
Write data back through governed workflows - no spreadsheet required

This is what makes the IBM-Anthropic partnership significant in practice. IBM watsonx Orchestrate uses Claude AI as its base language model and has been purpose-built for enterprise integration. The governance, audit trail and identity management in watsonx Orchestrate combined with Claude's financial reasoning capability creates an architecture that is not just powerful - it is enterprise-ready.

Business Outcomes

These are not projected benefits. They reflect what organisations implementing this architecture are demonstrating in practice.

3-5 days

Faster Month-End Close

Fast Close Agent automates reconciliation, journal validation and exception management across entities.

20-30%

Better Forecast Accuracy

Driver-based rolling forecasts updated continuously with live TM1 data rather than point-in-time snapshots.

70-85%

Reduction in Reporting Effort

Variance Agent generates structured commentary directly from TM1 cubes, eliminating copy-paste workflows.

Lower

Risk with Stronger Controls

Governed AI with full audit trail and compliance logging through IBM watsonx Orchestrate.

Real Time

CFO Insights on Demand

Instant answers from live TM1 data. No query, no export, no wait.

Why Octane: A Best-of-Breed Approach

At Octane, we take a best-of-breed approach to AI in Finance - partnering with the world's leading AI innovators: IBM, Anthropic and OpenAI.

For our customers, these collaborations deliver industry-specific AI Finance applications and AI-augmented workflows purpose-built for the finance function. Internally, they drive greater operational speed and productivity across our own teams every day.

For customers across EMEA, APJ - and specifically here in Australia and New Zealand - we have gone a step further by running Transformation Roadmaps that showcase the governance and trust frameworks enterprise AI adoption demands. That local focus gives us the end-to-end expertise to turn global advancements into secure, responsible local value for you.

KEY INSIGHT

"The teams that win the next decade will not be the ones with the most data. They will be the ones who can move from question to decision the fastest - with confidence that the answer is governed, audited and right."

100 Finance Prompts for Claude: A Practical Starting Point

To help Finance teams move from curiosity to capability, we have compiled a guide of 100 ready-to-use Claude prompts covering the full scope of the finance function. Each prompt specifies which Claude model to use (Opus, Sonnet or Haiku), which interface to use (Claude Chat, Claude Code or Claude in Excel), and is designed to deliver immediate, practical value.

Here is a taste of what is inside:

-> Monte Carlo Simulations	-> Cohort Revenue Analysis
-> Variance Analysis & Commentary	-> Multi-Scenario Financial Modeling
-> What-If Analysis	-> Headcount Planning
-> Executive Briefings	-> Data Consolidation & Cleanup
-> Revenue Forecasting	-> *And 91 more...*

FOR YOU IF

CFO - Strategic scenario analysis, board narrative generation, risk insight summaries

FP&A Director - Rolling forecasts, driver-based models, multi-scenario financial modeling

Financial Analyst - Variance commentary, data consolidation, Monte Carlo simulations, cohort analysis

No guesswork. No wasted time.

Where to Start: A Practical Sequence

For Finance teams already running IBM Planning Analytics, the architecture is available today. The question is sequencing.

1. Establish the knowledge and governance foundation first. AI agents are only as reliable as the data and policies they operate within. Governed cubes, clear dimension structures and documented business rules are the prerequisite.

2. Start with a high-frequency, high-effort process. Month-end close is the right starting point for most organisations. The Fast Close Agent delivers visible ROI quickly and builds confidence in the architecture.

3. Extend to continuous forecasting. Once the close is accelerated, the natural next step is making forecasting continuous. The Forecast Agent connects to the same TM1 model and works on the same governed data.

4. Unlock the CFO layer last. Board-level narrative and strategic insight are the highest-value outputs. They are also the ones that depend on confidence in everything below them.

Want to See It Live?

In 60 minutes we will map your close process, identify your top three automation opportunities, and run a live Agentic Finance demo - built on IBM watsonx Orchestrate and powered by Enterprise AI.