That gap is the real risk with AI agents in finance. Not that they fail, but that they succeed in isolation: another layer of tools sitting next to the process, rather than inside it.

AI agents create value in finance only when they are embedded in the workflows where decisions are actually made.

The adoption is real. The value is not yet.

McKinsey surveyed 102 CFOs and found that 44% were using generative AI in more than five finance use cases in 2025, up from 7% the year before. That is a remarkable jump.

But the same research points to the problem: nearly two-thirds of organizations have not begun scaling AI beyond pilots. The reasons are familiar to anyone who has run a finance transformation. Pilots break down under real conditions. They do not adapt as new data arrives. And they remain poorly integrated into core processes.

In other words, the constraint is not the model. It is the workflow.

What “embedded” means in utility finance

The opportunities in utility finance are not abstract. They have names: budget preparation, rolling forecasts, variance analysis, capital planning, rate scenarios, affordability analysis, board reporting, and regulatory requests.

Consider a simple case. The load comes in below the forecast for two consecutive months.

An isolated AI assistant can tell you that. It can even write a clear paragraph about it.

An embedded AI agent does something different. It detects deviations in the forecasting process. It traces which assumptions drove the original number. It identifies which capital projects and rate scenarios depend on that load. It recommends which scenario the team should review. And it routes the issue to the people who own the decision, with the context already attached.

The first is information. The second is a step in a decision.

That is the difference between AI as a tool and AI as part of a decision system, and it is why the previous articles in this series matter here. Decision-ready data, traceable assumptions, and a planning process built for continuous updating are not separate topics. They are the conditions that make an AI agent useful rather than merely interesting.

Why is this harder, and more important, for utilities

Financial decisions in a utility do not reside in a single department. A change in load forecast affects capital needs. A capital delay affects reliability or affordability. A cost variance reshapes rate scenarios. A regulatory request reorders reporting priorities.

An AI agent designed solely for the finance team will keep reproducing the silos the CFO is trying to break down. The agent has to understand the workflow that connects finance, operations, engineering, regulatory, and leadership, because that is where the decisions happen.

EY and Deloitte reach the same conclusion from different angles: organizations that get results treat AI as part of an orchestrated system of data, workflows, and controls, with agents reasoning and recommending within defined guardrails. Deloitte describes agents becoming part of the organization’s decision ecosystem. That phrase is exactly right, and it is exactly what an isolated chatbot is not.

For utilities, the guardrails are not optional. Finance cannot use AI effectively if the data is not trusted. Scenarios are not useful if assumptions cannot be traced. Automation creates risk if controls are unclear. And regulators will ask how a number was produced.

How a CFO should evaluate an AI agent

Less by how it performs in a demo. More about what it changes in the planning cycle.

Does it reduce manual reconciliation? Does it explain variances more quickly, including the assumptions behind them? Does it connect forecasts to scenarios, and scenarios to decisions? Does it leave an auditable trail? Does it get a decision made sooner?

If the answer to most of these is no, the organization has bought a very capable assistant, not a better way of planning.

The future of AI in finance is not an isolated chatbot. It is a governed, workflow-aware decision capability.

Planning platforms are already moving in this direction. IBM‘s work on AI in FP&A describes AI embedded throughout the planning process rather than attached to it, and the AI agents in IBM Planning Analytics are explicitly designed to operate inside the planning model: they read the same data, respect the same assumptions, and act within the same controls as the plan itself. The direction of the market is clear: the agent belongs inside the planning system, not beside it.

What this looks like in practice

In our own work with utilities, we made this choice deliberately. DaPlan for Utilities runs on IBM Planning Analytics, and its agents are embedded in the utility planning workflow: a variance agent that explains deviations in terms of the assumptions that changed, a forecasting agent that proposes a revised load-driven scenario, and a review flow that routes the impact on capital and rates to the people who own those decisions.

None of that is a chatbot next to the plan. It is the plan, working faster.

Closing the loop

This series started with the utility CFO becoming a decision orchestrator. AI agents are the most powerful instrument that role has been handed so far.

But an instrument is only as good as the system it plays in.

Utilities do not need AI agents that generate more analysis. They need AI agents that close the loop between data, insight, decision, and action, inside the workflows where the utility actually plans.

That is where AI stops being an experiment and becomes part of the way utility finance works.

Sources

• McKinsey & Company — How finance teams are putting AI to work today https://www.mckinsey.com/capabilities/strategy-and-corporate-finance/our-insights/how-finance-teams-are-putting-ai-to-work-today
• EY — How AI is transforming financial planning and analysis https://www.ey.com/en_us/services/consulting/finance-consulting-services/how-ai-is-transforming-financial-planning-and-analysis
• Deloitte — AI’s impact on the future of finance https://www.deloitte.com/us/en/what-we-do/capabilities/finance/articles/ai-future-of-finance.html
• IBM — AI in Financial Planning and Analysis https://www.ibm.com/think/topics/ai-in-financial-planning-and-analysis
• IBM — Planning Analytics AI https://www.ibm.com/products/planning-analytics/ai

This article was prepared using ChatGPT’s research capabilities, with additional support from Grammarly to refine the English grammar and clarity.