AI in enterprise planning is often reduced to a single question: can it produce a better forecast?

Predictive forecasting is certainly part of the opportunity, but it is only one part. A Planning Analytics environment is supported by developers who build and maintain applications, administrators who operate and govern the platform, and end users who analyze information and make planning decisions. Each group performs different work, faces different constraints, and can benefit from AI in different ways.

At IBM TechXchange 2026, Cubewise will explore this broader opportunity through the lens of AI-assisted forecasting with IBM Planning Analytics and watsonx Orchestrate. Rather than presenting AI as a replacement for established planning models or human judgment, the session will examine how it can extend the existing platform and improve the experience for the people who build, manage, and use it.

One Platform, Three Different Experiences

When organizations discuss improving a planning system, the conversation usually focuses on the end user. That is understandable. The ultimate purpose of the platform is to help the business plan, forecast, analyze results, and make decisions.

But the quality of that experience depends on work happening across the entire environment.

Developers write and maintain rules, processes, integrations, and application logic. Administrators manage security, access, deployments, monitoring, and other operational responsibilities. End users navigate models, investigate changes, update assumptions, and interpret results.

AI can assist all three groups in different ways.

For developers, the opportunity is greater speed and consistency in building and maintaining solutions. For administrators, it is the automation of repeatable operational work with appropriate controls. For end users, it is easier access to information, analysis, and predictive capabilities.

Looking at these groups separately helps move the discussion away from vague promises about “AI transformation” and toward practical questions about where assistance can make a measurable difference.

Helping Developers Work More Efficiently

Planning Analytics developers spend significant time working with code. They create TurboIntegrator processes, rules, MDX queries, REST integrations, Python utilities, and supporting documentation. They also inherit existing applications that may contain years of accumulated business logic.

AI-enabled development tools such as Cursor, IBM Bob, Claude, and other coding assistants have the potential to accelerate parts of this work. A developer might use them to:

• Review existing code and explain unfamiliar logic
• Identify repeated or inconsistent patterns
• Suggest updates to rules, processes, or scripts
• Generate an initial version of routine code
• Assist with documentation and test creation
• Support refactoring while preserving the intended business logic

The benefit is not that the AI becomes the Planning Analytics developer. It is that the developer can spend less time on mechanical work and more time making architectural decisions, validating business requirements, and reviewing the quality of the finished solution.

That distinction matters. Planning Analytics code is not isolated from the business. A seemingly small change to a rule, process, hierarchy, or data flow can affect calculations and decisions elsewhere in the model. AI-generated code still requires review, testing, and an understanding of the application’s wider context.

Used responsibly, these tools can act as development assistants. They can help experienced developers move faster and help less experienced developers understand applications more quickly, while keeping people accountable for what ultimately enters the environment.

IBM has demonstrated a similar development pattern in which IBM Bob assists with building and validating MCP tools before those tools are imported into watsonx Orchestrate. The developer remains involved at key approval points rather than turning over the entire process to the AI.

Streamlining Planning Analytics Administration

Many administrative activities follow repeatable patterns but still require someone to collect information, make a series of updates, verify the outcome, and resolve exceptions. These tasks are often straightforward individually, but together they consume time and create operational overhead.

Consider user provisioning. An administrator may need to determine the appropriate role, add the user to predefined groups, confirm that access was applied correctly, and record the result. The decision rules may already exist, but someone still has to move the request through each step.

An agent could potentially coordinate that process by:

1. Receiving a request and collecting the required details
2. Identifying the predefined role or access pattern
3. Calling approved tools to perform the required actions
4. Checking whether each action completed successfully
5. Escalating an exception or requesting human approval when necessary
6. Recording the outcome for review

This is more than executing a fixed script. An agentic workflow can use multiple tools, evaluate the result of each step, and determine what should happen next. If a check fails, the workflow can loop back, gather more information, retry an appropriate step, or hand the issue to an administrator.

Other potential administrative use cases could include routine environment checks, validation of configuration, scheduled reconciliations, process monitoring, or the collection of diagnostic information.

The strongest candidates share a few characteristics. They are repeatable, governed by known rules, supported by reliable tools, and are capable of being verified after execution. They also have clear boundaries for when a person needs to intervene.

The goal is not autonomous administration without oversight. It is to let administrators define how routine work should be handled, give an agent access only to the tools required for that work, and preserve appropriate review points for sensitive or exceptional actions.

Making Planning More Accessible to End Users

For end users, the most visible opportunity is a natural-language interface to planning data.

Planning Analytics models can contain extensive multidimensional structures, business rules, assumptions, and organizational detail. Experienced users may know exactly which book, cube, view, dimension, or report they need. Occasional users may understand the business question perfectly but struggle to translate it into the navigation and query steps required to find the answer.

A natural-language interface can reduce that gap.

Instead of manually navigating through several views, a user could ask a question about a variance, trend, forecast, or scenario. The assistant could help retrieve relevant information, explain results, or direct attention to an area that deserves further analysis. IBM describes the Planning Analytics Agent as supporting natural-language questions grounded in TM1 logic, along with model-based insights, explanations, and governed actions.

Predictive forecasting adds another form of assistance. Rather than treating a statistically generated forecast as the final answer, it can provide a baseline for review. The planner can compare that baseline with the current plan, investigate differences, apply business knowledge, and adjust assumptions where appropriate.

This preserves the role of the planner. AI handles more of the initial analysis and mechanical effort, while the planner provides context, judgment, and accountability.

The future end-user experience may therefore combine several capabilities:

• Asking questions about planning data in natural language
• Generating a baseline predictive forecast
• Explaining variances, drivers, or unusual results
• Initiating an approved workflow
• Reviewing recommendations before information is updated or published

IBM’s current direction for Planning Analytics AI similarly emphasizes model-aware analysis, predictive baselines, governed workflows, and controlled actions within the planning process.

Connecting AI to Planning Analytics

Useful AI assistance depends on access to the right data and actions.

For Planning Analytics, that connection is provided through the REST API. But technical connectivity alone is not enough. An implementation must also determine whose identity is being used, what that identity is allowed to access, and how actions will be governed and recorded.

A non-interactive credential can be appropriate for controlled back-end processes that run independently of a particular user. Examples might include scheduled integration, monitoring, or another predefined service operation.

Tasks initiated by an end user raise a different question. If an assistant is answering a question or performing an action on someone’s behalf, the preferred pattern is often to authenticate in the context of that user. OAuth can support this approach by connecting access to the user’s identity rather than giving every request the permissions of a shared technical account.

This is particularly important for natural-language analysis. A simpler interface should not become a way around Planning Analytics security. The assistant should operate within the user’s existing access, so the answers it provides are limited to information that person is already permitted to see.

The upcoming TechXchange session will examine these authentication patterns and the circumstances in which each can be appropriate. The broader lesson is that identity and security must be designed into an AI-enabled planning architecture from the beginning, not added after an agent has already been given access to the platform. The current presentation design identifies OAuth as the preferred direction where access should be tied to a real user identity, while retaining service-oriented patterns for suitable back-end automation.

Turning Planning Analytics Capabilities into Agent Tools

Once secure connectivity is established, the next question is what the agent should actually be allowed to do.

An agent does not need unrestricted access to the entire Planning Analytics REST API. Instead, specific capabilities can be made available as defined tools. Depending on the use case, a tool might retrieve data, execute a view, run a process, inspect metadata, perform a validation, generate a forecast, or submit a controlled update.

There are several ways to create this tool layer.

Direct REST API calls

Direct REST calls provide precise access to Planning Analytics capabilities. They can be useful when the required operation is well understood and the implementation benefits from direct control over requests and responses.

TM1py-based tools

TM1py provides a Python interface to the Planning Analytics REST API and can simplify the development of reusable functions. Existing TM1py-based code can be wrapped as tools that an agent is permitted to call.

This approach can be especially useful when an organization already has Python utilities for integration, administration, validation, or data access. Rather than rebuilding the underlying capability for AI, the organization can expose an appropriately controlled version of it to the agent.

MCP tools

The Model Context Protocol provides a standardized way for agents and AI applications to discover and use tools. In practical terms, MCP can reduce the amount of custom integration required each time a capability needs to be made available to another agent or development environment.

IBM’s evolving Planning Analytics MCP capabilities can provide a more standardized access layer, while organizations can also create their own MCP tools for narrower or customer-specific operations. The upcoming Cubewise session is expected to explore how direct REST access, TM1py-based tools, and MCP can each contribute to the overall architecture.

These methods are not necessarily alternatives competing for a single winner. A solution may use different approaches for different purposes. The important decision is not simply how quickly a tool can be created. It is whether the tool has a clear purpose, controlled scope, predictable behavior, and an output that the agent or a person can verify.

From Individual Tools to Agentic Workflows

A single tool performs an action. An agent can combine tools to pursue a broader goal.

For example, an AI-assisted forecasting flow could potentially:

1. Check whether required source data is available
2. Clean or validate the incoming information
3. Initiate a baseline predictive forecast
4. Return the result for analysis
5. Identify areas requiring attention
6. Ask a planner to review or adjust assumptions
7. Continue the workflow only after the appropriate human decision

More advanced designs may involve multiple agents. One agent might focus on data preparation, another on forecasting, and another on communicating results or coordinating approvals. A supervising agent could manage the flow between them and determine when a step should be repeated, stopped, or escalated.

These patterns make it possible to move beyond isolated AI interactions toward repeatable business processes. But they also increase the importance of boundaries, monitoring, and human review. The more actions an agent can coordinate, the more clearly the organization must define what it can decide independently and what still requires approval.

watsonx Orchestrate provides the environment in which tools, agents, multi-step processes, and human checkpoints can be coordinated. Planning Analytics remains the governed planning foundation, while Orchestrate helps connect individual capabilities into a broader workflow. The presentation currently frames this as a target architecture and roadmap rather than a completed production implementation.

Bringing the Agent into the Planning Experience

AI assistance becomes more useful when it is available where the user is already working.

Agents created through watsonx Orchestrate can be incorporated into the embedded Planning Analytics AI experience, allowing users to access custom capabilities without leaving Planning Analytics Workspace. IBM has described support for customer-managed watsonx Orchestrate environments within the Planning Analytics Assistant, including prebuilt and custom tools and agents.

This creates the potential for an agent to respond with a better understanding of the planning environment and the task the user is trying to complete. Rather than beginning every exchange without context, an embedded assistant can be designed to work closer to the model, workflow, and point of analysis.

That can make interactions faster and more relevant. A user investigating a forecast should not have to explain the entire structure of the planning application before asking a useful question. The closer the assistant is to the governed planning context, the more effectively it can help the user interpret information and decide what to do next.

Start with Assistance, Then Expand

The most credible path to AI-enabled planning is not to automate an entire forecasting cycle immediately.

A better approach is to begin with bounded use cases that solve recognizable problems:

• Help a developer review and update code
• Automate a repeatable administrative workflow
• Give an end user a natural-language path to analysis
• Generate a predictive baseline for planner review
• Connect two or three existing steps through a governed agent

These use cases can create value independently while also establishing the technical foundation for more advanced workflows. They force organizations to address identity, security, tool design, testing, monitoring, and human approval before attempting broader automation.

Over time, additional tools and agents can be introduced for developers, administrators, and end users. The Planning Analytics MCP ecosystem will continue to evolve, and watsonx Orchestrate can provide a common environment for combining those capabilities into reusable workflows.

But the objective should remain consistent: AI should make the planning environment easier to build, easier to manage, and easier to use.

Continuing the Conversation at TechXchange 2026

Last year, Cubewise presented practical approaches to scaling Planning Analytics through modern integration, parallel processing, and greater operational transparency. This year, the conversation moves to the next stage: how AI-assisted experiences can extend that foundation across forecasting and the wider Planning Analytics lifecycle.

At IBM TechXchange 2026, the session will explore the potential benefits for developers, administrators, and end users, along with the technical patterns that can connect Planning Analytics to watsonx Orchestrate. The discussion will cover secure REST connectivity, identity and authentication, direct and reusable agent tools, MCP, agentic workflows, and the delivery of custom agents through the Planning Analytics experience.

The future of AI in enterprise planning is not simply a forecast generated by a model. It is a planning platform in which people spend less time navigating mechanics and more time applying expertise, evaluating options, and making decisions.

That is the opportunity we look forward to exploring at IBM TechXchange 2026.

About Cubewise

Cubewise helps organizations get more value from IBM Planning Analytics through consulting, application development, managed services, education, and purpose-built software. Our teams combine deep Planning Analytics expertise with modern technologies to help customers improve performance, integration, governance, and the overall planning experience.