Latest episode
SensibleAI agents on a OneStream model we have checked can feed them.
SensibleAI is the forecasting and agent layer OneStream (EPM software) runs on its own platform. Constancia, an EPM consultancy, surveys the model underneath before a budget goes to the layer on top.
OneStream at a glance
- Covers
- Planning, financial close and consolidation, reporting and analytics.
- Data in
- Direct integration to Oracle, SAP, Microsoft, IBM, Infor, Workday, Databricks and Snowflake.
- Our record
- 58 implementations over 35 years.
SensibleAI agents on OneStream read the same entities, accounts and periods the group's close already uses, and they act on what they read. Constancia establishes whether a group's model can carry them before any agent is switched on.
- Reads
- The consolidated model as it runs, with every figure traced to the ledger balance it came from.
- Records
- Which definitions have a named owner today and which the roadmap still has to settle.
- Then
- A fixed price for the work that makes the model ready, set before the agent layer is bought.
SensibleAI works on a traceable model, and our survey says whether yours is one.
SensibleAI puts forecasting and agents on the OneStream model a group already closes on, for a finance team whose board has set an AI date. The agents read the model directly, so what the model holds decides what the agents can do. The survey is the piece of work Constancia sells here, and the agent build follows it.
SensibleAI reads 8 direct integrations, from Oracle and SAP to Databricks and Snowflake, and an agent works on whatever those integrations load. Constancia, an EPM consultancy, reads the same loads in the survey and lists what each one will release.
Where 3 entity teams finish their numbers in a workbook after the extract, the agent reads the extract and misses the step that changed it. The variance commentary it drafts reads well and cites a margin the group controller cannot find in any ledger.
An agent needs 5 things from the OneStream model before its output can be checked.
The 5 are the properties the survey tests, entity by entity. A model with all 5 carries an agent, and a model missing 1 of them produces a confident answer nobody can defend.
- 01
A figure that traces to a ledger
Every consolidated figure links back to the ledger balance it came from, with no manual step in between.
- 02
A definition with an owner
Revenue, gross margin and headcount each have 1 written definition and 1 person who signs for it.
- 03
A plan on the same dimensions as the actuals
The forecast and the close share entities, accounts and periods, so a variance is a subtraction and needs no mapping table.
- 04
A close that repeats
The same process produces the same result whoever runs it, so the agent learns 1 pattern and not 12 variations of it.
- 05
A record of past decisions
Last quarter's mapping choices and adjustments are written where the agent can read them, so it makes the same call again.
None of the 5 asks for a perfect model, and the survey lists which of them a group's model meets today.
4 things hold in every group where SensibleAI pays back.
- 01
The close already runs on OneStream
The consolidation lives on the platform, so the agent reads the model itself and no copy of it.
- 02
A date on the board's AI request
The board has asked for a forecast or a commentary by a named meeting, and that date sets the sequence.
- 03
Drivers owned outside finance
The sales director stands behind the pipeline number and HR behind the headcount, so the forecast can be challenged.
- 04
A person to sign the output
A named person in group finance signs anything the agent produces that reaches the board pack.
A group whose close still ends in a workbook does the model work first, and the survey sets the price for it.
What are common AI agents?
The common AI agents in finance propose mappings, check reconciliations, draft commentary and watch the close calendar. Each one repeats a task whose answer a person can verify against the ledger.
- 01
The mapping agent
A new entity's chart of accounts arrives and the agent proposes the map to the group chart, line by line, for an accountant to accept.
- 02
The reconciliation agent
The agent checks each loaded balance against its source ledger every period and lists the ones that fail.
- 03
The commentary agent
A first draft of the variance narrative comes from the agent, and the FP&A analyst rewrites it before the CFO sees it.
- 04
The calendar agent
The agent watches the close calendar and flags the day an entity's load slips, before the group controller notices.
Each of the 4 reads the model the close runs on, which is why the 5 properties above come first.
3 pages carry the rest of the argument about agents in the close.
Agentic AI for finance
The same 4 things an agent has to read, described without reference to any platform.
Read about agentic AIOneStream implementation
The build that puts the close on the model the agents will later read.
Read about implementationOneStream managed services
The running of the model after go-live, including the releases that change what an agent can see.
Read about managed services
See more on our
Subscribe on YouTube
Every Sunday, 6am
The Sunday Close.
Our weekly newsletter on AI in finance.
Yours to enjoy with your Sunday morning coffee.
- AI in finance observations from Alex (CEO) and Brad (COO)
- Newest video from our YouTube channel
- AI in finance trends we have spotted each week
Tell us the date you are working to.
The people who would do the work read every message and reply inside 2 working days.