The numbers stop being a copy
Every decision rests on a spreadsheet, and nobody can say where its figures came from or whether they are current. Here a cell names a KPI, and the answer to both questions is in the cell.
Model the decision where the numbers already live
Select the range a planner already maintains, say how to read it, and run a decision model on it. The result comes back as a run with an id, not a number pasted into a slide.
A cell can name a KPI
Write =KPI("attrition_rate") and the number comes from the platform, not from somebody's copy and paste. PROJECT, GOAL, RUN and SCHEDULED work the same way.
Every number can say where it came from
Select a cell and see what fed it, when it last changed, and who changed it. The blame trail and the error trace are part of the grid, not a forensic exercise.
It lives inside the governance
Workbooks carry the same Draft to Official lifecycle as everything else, so only what has been formalised is quotable, and the AI can only cite what is Official.
It is a real spreadsheet first
Governance is worth nothing if the tool cannot do the work. This is what is in it.
A real calculation engine
If the formulas are not there, nothing else matters. They are, including the ones people actually leave a tool over.
- 187 functions across maths, statistics, text, dates, logic, lookup, finance, and information
- Dynamic arrays that spill, with the spill range tracked so a formula cannot quietly overwrite work
- Named ranges and a name manager, so a model reads in words rather than in coordinates
- Automatic, manual and iterative calculation modes, with circular references resolved rather than refused
- Incremental recalculation on a dependency graph, so a large model does not recompute itself on every keystroke
- A formula bar with function search, argument help, and an insert dialog for the ones nobody remembers
Numbers that come from somewhere
The reason a spreadsheet goes stale is that every figure in it was typed once and never checked again.
- Platform functions that read a KPI, a project, a goal, a model run, or a scheduled value straight into a cell
- Bind a range to canonical HR and finance records, filtered the way you need them
- Cell lineage and blame, so any figure can be traced back to what produced it
- An error trace that explains a broken formula by walking its inputs rather than printing a code
- A lint panel that finds the things that silently break models: inconsistent formulas, hardcoded values inside ranges, references off the edge of data
Analysis that belongs to the decision
The analysis people currently do in a spreadsheet and then describe in an email, done where the decision record can see it.
- Run a decision model on a range: the numbers a planner already maintains become model inputs without being retyped
- Goal seek, sensitivity analysis, and scenario comparison built into the grid
- Monte Carlo simulation over a range, with the run recorded like any other model run
- Pivot tables and charts, with the chart following its source range
- Run history, so the model runs that came out of this workbook are listed against it
The parts that make it usable at work
A modelling tool nobody can share is a personal habit, not a system of record.
- Real time collaboration on the same workbook, with presence and a live connection indicator
- Comments anchored to a cell, and threads that resolve
- Folders, a template library, and filing by drag
- Version history and named versions, with a diff you can read
- Excel import and export that round trips, so nobody is trapped
- Conditional formatting, find and replace, borders, freeze panes, and print layout
- Ask the assistant to build part of a sheet, inside the same permissions and budget as every other AI call
Wired into everything else
A workbook is not an island here. It reads the platform, and the platform reads it.
KPIs
A cell reads the metric directly, and updates when it does.
Models
Run a model on a range and keep the result against the workbook.
Canvas
Put a workbook on a board, or bind a shape to one of its cells.
Agents
Everything here is reachable over MCP, under the same permissions.
Auditable, because a model that matters will be questioned
Draft to Official lifecycle, per workbook sharing, version history, cell level blame, retention and legal hold. The AI can ground on a workbook only once somebody has made it Official.