GA
Close Agent
Runs the month-end checklist. Chases what is late. Records who signed.
3 days1 (modeled outcome) shorter month-end close
Finance agent
GAPlanning Agent
The Planning Agent is a governed AI agent that explains every variance. Answers plan questions in plain language.
How it works
What the Planning Agent does
Attributes budget-to-actual variances to price, volume, mix, FX, and timing, drafts the commentary in your format, and lets finance interrogate the plan conversationally.
- 01
Connect plan and actuals
The agent reads budget and forecast versions from your planning system and actuals from the GL or warehouse, along with driver data: headcount, volume, price, FX rates.
- 02
Compute and attribute
On close, it computes variances by cost center, account, and driver, then attributes each to price, volume, mix, FX, timing, or one-time items, with the math shown.
- 03
Draft the commentary
Commentary follows your pack's structure and thresholds. Each paragraph carries its supporting table. Analysts edit, sign, and publish.
- 04
Explore and scenario
Finance asks follow-ups in Slack, Teams, or the workspace. Scenarios run on hiring plans, pricing, and timing. Promoting a scenario to the plan of record requires a planner's approval.
app.meridian.example/agents/planning/q3-variance
Q3 2026 · actual vs plan
Commentary drafted
| Line | Actual | Plan | Variance |
|---|---|---|---|
| Revenue | 12,481,220.00 | 12,940,000.00 | −3.5% |
| Cost of revenue | 4,102,880.00 | 4,268,000.00 | −3.9% |
| Sales & marketing | 2,884,140.00 | 2,640,000.00 | +9.2% |
| R&D | 3,190,500.00 | 3,220,000.00 | −0.9% |
| G&A | 1,208,760.00 | 1,260,000.00 | −4.1% |
Draft commentary. Sales and marketing ran 9.2% over plan, driven by the pulled-forward Q4 conference spend ($184,400.00) and two contractor extensions. Revenue landed 3.5% under plan on slower enterprise close rates; pipeline coverage for Q4 is 3.1×.
The agent contract
Six terms, applied to this agent
Every Meridian agent runs under the same six terms, enforced at the Gateway below the agent. Here is what each one means for the Planning Agent.
Scoped to one workflow
Explains variances and answers plan questions. It never writes to the plan of record.
Reads only permitted data
Inherits row-level security from the planning system. Users see only their cost centers.
Every action logged
Every number carries the query that produced it. Commentary is tagged agent-drafted until signed.
Consequential actions require human approval
Scenarios stay drafts until a planner promotes them. Analysts sign commentary before it publishes.
Measurable outcome
Analyst hours on the pack and variances explained above threshold reported monthly.
Runs on your chosen model
Pin sensitive plans to a specific model provider or region per workspace.
Scope
Data, actions, and approvals
What the Planning Agent reads, what it does on its own, and where a person signs.
Data it reads
- Budget, forecast, and plan versions from the planning system
- Actuals from the general ledger and data warehouse
- Drivers: headcount and hiring plan, volume, price, FX rates
- Cost center and account hierarchies
- Prior-period commentary and reporting-pack templates
- Row-level security from the planning system
Actions it takes
- Compute variances by cost center, account, region, and driver
- Attribute each variance to price, volume, mix, FX, timing, or one-time items
- Draft commentary in your reporting-pack format with supporting tables
- Answer plan and actuals questions in plain language with the query shown
- Run what-if scenarios on headcount, pricing, and timing assumptions
- Flag forecast lines with trends outside their historical range
What needs approval
- Never writes to the plan of record; scenarios stay drafts until a planner promotes them
- Every number carries the query that produced it, inspectable by the reader
- Commentary is tagged as agent-drafted until an analyst signs it
- Inherits row-level security from the planning system; users see only their cost centers
- Model choice is set per workspace; sensitive plans can be pinned to a specific model or region
Outcomes
Numbers the Planning Agent is accountable for
Modeled outcomes from design-partner deployments. The Registry reports your actuals from the first week an agent is live.
Minutes
to first-draft variance commentary, down from days1
−60%
analyst hours on the monthly reporting pack1
100%
of variances above threshold with driver attribution
1 query
behind every number, inspectable by the reader
- 1.Modeled outcome from design-partner deployments. ↩
“The commentary draft lands the morning after close with the driver table under each paragraph. My analysts edit and sign. They stopped compiling.”
In production at
Integrations
Systems it works with
Connected through prebuilt connectors and zero-copy warehouse access in Data Fabric. The agent acts as the user and inherits each system’s permissions on every read.
- Planning and FP&A platforms
- ERP
- Snowflake
- Databricks
- BigQuery
- Excel
- Slack
Credits
What the Planning Agent costs
Included on every plan: Starter with 5,000 credits per month, Growth with 30,000, and Enterprise with custom pools. Credits are consumed per completed action, never per message. Usage beyond the plan is billed at $0.12 per credit.
| Action | Credits |
|---|---|
| Variance commentary, per reporting unit | 3 |
| Conversational query answered | 1 |
| Scenario modeled | 5 |
Questions
Questions about the Planning Agent
Answered the way buyers ask them. Each answer is also published as structured data.
It is the automated computation and explanation of differences between planned and actual results. Meridian's Planning Agent computes variances at whatever grain your hierarchy supports, attributes each to its drivers, and writes the explanation in the format your reporting pack already uses. Analysts review and sign rather than compile.
It decomposes the variance using the drivers you plan on. A revenue variance is split into price, volume, mix, and FX. An opex variance is split into headcount, rate, timing, and one-time items. The decomposition is arithmetic you can inspect, not a narrative guess.
Not on its own. It can build scenarios and show their effect. Promoting a scenario to a forecast version requires a planner with the right permission to approve it in the planning system.
Any planning platform with an API or a warehouse export, and any GL or warehouse for actuals. Most design partners connected through Snowflake or Databricks so the agent reads the same governed tables as their BI tools.
No. Your data is never used to train Meridian's models or any third-party model. Model choice is yours, and you can pin a workspace to a specific provider and region.
Related agents
Agents most often deployed alongside the Planning Agent, on the same data and the same approval policy.
Get started
Put the Planning Agent to work.
Start with one workflow, one approver, and one number to move. Most design partners were live in five weeks.