Skip to content

Planning tools and AI agents: Anaplan, Pigment and Adaptive Planning

5 min readFor FP&A leaders and finance systems owners connecting planning models to AI agents

Anaplan, Pigment and Workday Adaptive Planning can answer planning questions in Claude, ChatGPT and other MCP clients through their APIs. The hard part is choosing the version or scenario, the switchover between actuals and forecast, and the sign convention for variance, because a model cannot guess those. Anaplan and Adaptive Planning tie API calls to a user, while Pigment uses API keys an administrator creates, which changes who sees what.

Planning questions sound simple

How are we tracking against budget? What changed since last month’s forecast? Which cost centers have not submitted? FP&A answers these every cycle, usually by opening the model, picking a view and exporting it. They are exactly the questions people now want to ask an assistant such as Claude or ChatGPT.

Planning data is unusually easy to misread, though. A model holds several versions of the same number, actuals stop and forecast starts at a switchover period, and an account’s time behavior decides whether a quarter is a sum or a closing balance. A connection that returns the wrong version still returns a number that looks right.

How each tool is reached

The three tools expose their data differently, and the difference decides whose access applies to an answer.

ToolAPIWhose access
AnaplanIntegration API v2 and transactional APIs for lists, views and cellsEach token belongs to an Anaplan user, so model roles and selective access apply to what the token reads
Workday Adaptive PlanningAPI methods such as exportData, exportLevels, exportAccounts and exportVersionsEach call carries an Adaptive Planning user’s credentials, and results respect that user’s permissions
PigmentREST API export endpoints for views, metrics, lists and tablesEach user’s Pigment roles and access rights decide which views, metrics and scenarios they can read

Why whose access matters here

With Anaplan, Adaptive Planning and Pigment, a budget owner who signs in with their own account (or your SSO) sees their own cost centers or levels, the same as in the planning tool. That is the natural fit for questions from budget owners across the company.

The same holds for sensitive data. Draft scenarios and workforce data stay with the users the planning tool already allows to see them, because Nexatron inherits those existing permissions.

Settle these before the first question

Five choices decide whether an answer matches the number in the board pack. None of them can be inferred reliably from the wording of a question.

  • Version or scenarioWhich budget, which forecast and which scenario is the plan of record. Anaplan versions, Adaptive Planning versions and Pigment scenarios all hold several at once.
  • SwitchoverWhere actuals end and forecast begins. A full-year number combines both, and the boundary moves at every close.
  • Time behaviorWhether an account sums over time, like opex, or takes the period-end balance, like headcount or cash.
  • Variance signWhether favorable is positive for both revenue and expense, and whether variance is plan minus actual or actual minus plan.
  • LevelWhich level of the cost center or department hierarchy to report at, and whether totals include children.

Read through the model’s own views

Model builders already maintain saved views and exports for downstream use. Reading those views, and the line items or metrics behind them, keeps an assistant’s answer aligned with what FP&A publishes. Querying raw cells without the model’s logic invites a second, slightly different calculation of the same number.

Cross-system questions are where planning data is most useful: plan against actuals posted in the ERP, planned hires against people in the HR system. Those need the planning tool and the system of record in one answer, with the query for each shown, and an agreed mapping between accounts, levels and departments.

Which number doesthe question mean?Before the switchoverActualsAs loaded into the model, or asposted in the ERPAfter the switchoverForecast versionThe current version of recordFull yearActuals plus forecastSay where the boundary sitsAgainst budgetBudget versionLocked, as approvedWhich number does the question mean?Before the switchoverActualsAs loaded into the model, or as posted inthe ERPAfter the switchoverForecast versionThe current version of recordFull yearActuals plus forecastSay where the boundary sitsAgainst budgetBudget versionLocked, as approved
Most planning disagreements come from picking a different branch of this tree.

Keep agents read-only on plans

Planning tools also accept imports, and some MCP servers expose write tools. A forecast submission is a decision someone owns, and an agent that can load data into a model needs its own approval, its own audit trail and a person who confirms each change. For questions, read-only tools are enough, and they keep the review simple.

Test it the way you would test any connector. Two people with different model access ask the same question, and each should see only their own levels or cost centers, or a refusal. Then ask for a change and confirm nothing is written to the model.

Three questions to test first

Pick three questions FP&A already answers by hand, and write down the expected number before asking. Good candidates are actuals against the current forecast for one department in the last closed month, what changed in total opex between the last two forecast versions, and which cost centers have not submitted for the current cycle. Each touches a different part of the model: versions, the switchover and a status flag.

If the assistant matches all three, widen the group of people allowed to ask. If it misses, the difference almost always points at a definition from the list above, and fixing the definition fixes every later answer that uses it.

How Nexatron handles planning tools

Nexatron connects to Anaplan, Pigment and Workday Adaptive Planning through their APIs with user-level access: users sign in with their own account (or your SSO), and Nexatron inherits their existing model roles and access rights. People ask in the Nexatron app or from Claude, ChatGPT and Microsoft Copilot through Nexatron’s hosted MCP server. Questions only read the model.

Each answer says which version, scenario or reading it used, with the others one click away, and shows its query. It reads Verified only once a named person on your team has reviewed the logic. A question Nexatron cannot answer under your rules is declined with the reason and what to do next. The Anaplan, Pigment and Adaptive Planning use-case pages list example questions, and the Workday Adaptive Planning and NetSuite page covers plan against posted actuals.

Claude is a trademark of Anthropic, PBC.

Bring three business questions

Explore your sources, definitions, and review criteria with our team. Please keep credentials and sensitive records out of the inquiry.