Summarize this blog post with:
| Anaplan CoModeler is an AI agent that helps model builders understand inherited models, build new functionality, generate documentation and catch modelling issues, without losing control. |
TL;DR
In Polestar Analytics' recent Anaplan Community Experience session, we demonstrated these capabilities through two hands-on use cases, currency translation and DISCO-based model optimisation. Across both, CoModeler accelerated the work while the model builder stayed firmly in control.
Traditional Anaplan implementations typically take six to twelve months or longer. That's the kind of manual, expertise-dependent bottleneck agentic AI is now closing across finance. Wolters Kluwer's 2025 CFO survey found that 44% of finance teams are expected to be using agentic AI by 2026, a jump of more than 600% from current adoption levels.
Deloitte's Q4 2025 CFO Signals survey adds weight to that shift: 54% of CFOs say integrating AI agents into their finance departments will be a top transformation priority in 2026 and 87% believe AI will be extremely or very important to finance operations this year. Anaplan CoModeler is part of that shift, built specifically for the model-building work behind every planning cycle.
That gap is the reason CoModeler exists. It was also the backdrop for Polestar Analytics' recent Anaplan Community Experience session, "CoModeler for Clear, Correct, Continuously Improved Models."
The session brought together Steven Gnavotich and Vijay Vyakaranam from Polestar Analytics for a practical walkthrough of Anaplan CoModeler: understanding an unfamiliar model, extending it, and reviewing it against modelling best practices.
Polestar Analytics is an Anaplan implementation partner that builds AI-enabled connected planning solutions for finance and supply chain teams. Here's what the session covered.
Anaplan CoModeler is an AI agent built into the Anaplan platform that helps model builders build, extend, and optimise planning models through natural-language prompts, without replacing the model builder's judgement.
The session opened with where CoModeler fits. It is not meant to replace model builders. Instead it is seen as a tool that helps the work they already do. Anaplan describes the benefit in these words: changing requests written in normal language into organized models, logic and math in minutes instead of the days or weeks it takes to do it manually.
| Build |
Extend |
Optimize |
| Create new modules and model components |
Add functionality to existing models |
Review models for inefficiencies and best-practice gaps |
| Work from natural-language prompts |
Work with existing model structures and dependencies |
Identify potential formula and structural issues |
| Support repetitive development tasks |
Add dimensions, calculations and logic |
Help keep models lean and maintainable |
If you're new to the agent, our glossary on What Is Anaplan CoModeler? breaks down how it works and where it sits within the Anaplan ecosystem, worth a read before the demo walkthrough below.
Documentation: An Overlooked Use Case
The session also highlighted documentation as important, and often overlooked. Anaplan models have a Notes section for a reason, but documentation can fall behind when teams are focused on delivering the next requirement. CoModeler can generate module-level documentation from the existing model structure, giving a new model builder inheriting a model meaningful context from day one.
The live demonstration moved from slides into a scenario familiar to most Centres of Excellence teams.
It's a Tuesday morning. An urgent request arrives from the Finance Director: add currency translation capability to an existing model. There's one immediate problem: the model was inherited only the previous week. It wasn't built by the current team, and there's little documentation to explain how it works.
- No useful module notes.
- No line-item descriptions.
- No obvious starting point.
Instead of going module by module manually, the demonstration used CoModeler to generate documentation from the model itself. A simple natural-language prompt asked CoModeler to add the purpose of each module to the Notes section, from both a technical and functional perspective.
CoModeler first produced an implementation plan for review. Only after the plan was approved did it make the changes. That approval step became a recurring part of the demonstration: the model builder remains in control of what gets applied.
Results of the Documentation Prompt
- CoModeler interpreted existing module names and naming conventions.
- It generated descriptions for the model's modules.
- The demonstration covered 14 modules.
- The user could review the proposed workflow before execution.
- Existing notes could be left untouched, appended to, or replaced.
The practical benefit: documentation that might otherwise be deprioritised could be generated while the model was being understood.
With context around the inherited model in hand, the demo returned to the original Finance request. The business operates across multiple countries and needs to view revenue, cost of sales, and margin in different currencies, not only the base currency.
Normally, this requirement would move through a familiar process:
- Scope the request
- Add it to a sprint backlog
- Work through requirements
- Build the functionality
- Review and test it
The demonstration showed another route. A detailed natural-language prompt described the required currency translation logic. CoModeler then worked through the existing model to determine the objects it needed: currencies, country and city hierarchies, product structures, existing calculation modules, and FX-rate inputs.
It then produced an implementation plan before building the new functionality. The interesting part wasn't simply that it created a module. It was the amount of model context it considered while doing so. CoModeler identified:
- The dimensions required for the new module
- Appropriate line-item formats
- Summary methods
- Existing modules that contained relevant source data
- Mapping relationships across the model
- FX-rate information
- The formulas required to calculate translated revenue and margin
The resulting Currency Translation module was created within the existing model structure. Number formats, summary methods, and formulas can still be adjusted manually or through another prompt, the output isn't treated as final. The objective isn't to remove the model builder from the process. It's to shorten the distance between business requirement and workable model component.
The third part of the demonstration moved from building and extending to optimisation. The question shifted from "what should we build" to "is the model we already have following the right modelling practices."
A simple prompt asked CoModeler to analyse the model's line-item formulas for adherence to the DISCO methodology. The kind of review being demonstrated would traditionally require a senior Anaplan architect to work through modules, formulas, and references, and compare them against established practices.
CoModeler returned two instances where the model didn't adhere to the methodology. More importantly, it explained why they were flagged and suggested how they could be remediated. For example, mapping-related line items sitting in a calculation module were identified as better suited to a System module. The result wasn't just a list of problems, it was an action plan.
That distinction matters when reviewing a large model. Finding an issue is useful. Understanding the issue and knowing what to do next is more useful.
Did You Know?
DISCO isn't just a catchy name, it stands for Data, Inputs, System, Calculations, and Outputs, the five module types Anaplan recommends keeping separate.
One of the recurring themes throughout the demo was control. CoModeler doesn't simply make changes in the background and leave the builder to figure out what happened. The workflow shown in the session involved:
- Prompt — describe what needs to happen.
- Plan — CoModeler proposes an implementation approach.
- Review — the model builder reviews the proposed steps.
- Approve — the user explicitly authorises execution.
- Validate — the resulting changes can be reviewed and refined.
Changes made through CoModeler are also captured in Anaplan's audit history. The user can approve only selected steps rather than accepting an entire plan, which matters when CoModeler is used against an existing, business-critical model.
Want the complete session? Live demonstration, panel Q&A, and the full discussion on model building, extension, optimisation, and governance.
Polestar Analytics is now a preferred global AI delivery partner for Anaplan CoModeler and Anaplan Agent Studio. That means our teams work directly with Anaplan on CoModeler rollouts, helping enterprise customers move from experimentation to trusted, business-critical adoption.
Polestar Analytics has also put together a short video walking through what CoModeler is and how it works.
CoModeler is designed around Anaplan-specific implementation knowledge, including concepts such as DISCO, model sizing, naming conventions, formula patterns, and dimensionality. This allows it to perform tasks such as reviewing a model against DISCO methodology and generating model documentation from natural-language prompts.
Yes. CoModeler can be used with both Anaplan's pre-built application models and fully custom models. The model used during the webinar was a custom model, and the demonstration covered documentation, extending the model, and reviewing its existing structure.
Yes. CoModeler can be used with both Anaplan's pre-built application models and fully custom models. The model used in Polestar Analytics' webinar demonstration was a custom model, where the team used CoModeler to generate documentation, extend the existing model with currency translation, and review its structure against modelling best practices.
The model builder remains in control. CoModeler presents a proposed plan before execution, and users can choose which steps to implement. They can approve the full plan, reject specific steps, or ask CoModeler to revise the plan. Nothing executes without explicit user confirmation.
- Understanding an inherited model
- Creating documentation
- Building repetitive components
- Extending existing functionality
- Reviewing models against best practices
- Identifying potential issues
Bottom line: CoModeler is an accelerant for the model builder, not a replacement.
Further reading: A deeper look at the thinking behind the capability and how it supports smarter planning.