Excel

Automate native Excel reports from the data that matters

OfficeMaker turns structured operational data into repeatable native Excel reports and reporting workbooks. The workbook pattern — sheets, tables, formulas and formats — stays stable while the figures refresh. Filter and calculate before any model writes a narrative. The download is a real .xlsx that a reviewer can audit in Microsoft Excel.

The outcome

A recurring management pack, operational tracker or finance extract is a poor candidate for a fresh prompt every Monday. Define the workbook once against the live Excel schema, then run it when the source has new numbers. Each run still produces an editable .xlsx on the same schema-led path as the AI Excel generator.

The architecture is schema → structured JSON → validation → OfficeMaker middleware → native Office file. The model proposes structure and, when you ask for it, interpretation. Middleware builds the file. JSON is a representation language models are extensively trained on, which keeps Office Open XML and rendering code out of the prompt. That can improve predictability. It does not guarantee a particular report quality.

Typical triggers

On an organisation tenant, a report can start from a scheduled job, an API or webhook call, a website form, an uploaded spreadsheet, or a business workflow. An entitled Salesforce connection can supply CRM fields. An assistant can call the same generation through MCP or REST. The public free MCP server can create a new workbook from schema-valid JSON. It cannot register the schedule or the organisation workflow.

  • Scheduled job on an organisation tenant
  • API or webhook call
  • CRM event where Salesforce generation is entitled
  • Form submission
  • Uploaded spreadsheet used as a source, then queried on the paid document service
  • MCP or REST create from an application or assistant

Filter first, reason second

Call a calculation “AI” only when a model is actually interpreting something. The reporting sequence is source data, then query and filter, then calculate and aggregate, then a relevant structured result, then language-model interpretation if the narrative needs it, then Excel report generation.

Source data → query and filter → calculate and aggregate → relevant structured result → model interpretation if required → Excel report.

The model should not read hundreds of thousands of irrelevant rows to produce a weekly pack. Deterministic work belongs in the source system, in Excel formulas, or in OfficeMaker Excel search on the authenticated document service. Excel search converts the workbook and can list tables, sample rows, query, aggregate and return compact JSON or CSV. Keyed updates compile to edit and merge operations and can require approval before they are committed. Those verbs are paid. They are not on the free MCP server.

Sending only the relevant aggregate uses fewer tokens than placing the full workbook in context, because the unused cells are never serialised into the prompt. The difference scales with what you left out. OfficeMaker does not promise a fixed percentage reduction. A separate, dated illustration of middleware versus a code-and-screenshot loop is linked below and is specific to that PowerPoint workflow.

Filter first, reason second.

What the native workbook can contain

Returning a real workbook means the next analyst inherits formulas rather than a picture of numbers. Where the live Excel schema supports them, include multiple sheets, tables, formulas, charts, number formats, named ranges, freeze panes, data validation and conditional formatting. Use formula bodies without a leading equals sign, and prefer A1 or named ranges. Keep sheet names easy to reference.

Automated numbers are only as good as the inputs and the formulas you defined. OfficeMaker does not certify management reports.

Repeatability

The useful property of report automation is a same structure with refreshed data. The automation stores the workbook pattern and the input contract. A schedule or an API call supplies the next period. Reviewers see a consistent file instead of a redesigned grid. Organisation run history ties an output back to the inputs for that run.

Widen from one report to a pack only after the first workbook has been checked in Excel, including formats and any narrative sheet the model drafted from the aggregates.

Existing workbooks

Creating a new .xlsx from structured JSON is available on the free sandbox and on paid hosts. Querying, filtering and updating an existing workbook is the authenticated document service: convert the file, then Excel search or a document slice, and merge or edit when the change must be written back. The free MCP server stops at create, preview, schema and gap analysis.

If the source of truth is a database rather than a workbook, filter there and pass the aggregate into create. You do not need to round-trip a spreadsheet for the filter to happen outside the model.

Evidence behind this

Native Office files

Claim. OfficeMaker creates native Word, Excel and PowerPoint files.

Mechanism. The model supplies schema-valid structured state. OfficeMaker middleware validates that state and constructs the .docx, .xlsx or .pptx file.

Evidence. Document generation API

Limitation. Create is available on the public free sandbox. Merge, edit and conversion of existing files are paid document-service verbs.

Deterministic Excel query

Claim. An existing workbook can be filtered before a model sees row-level detail.

Mechanism. Paid document-service Excel search scans the converted workbook and returns compact results. Writes compile to existing edit or merge operations and can require an approval token.

Evidence. Query Excel before the LLM

Limitation. This query and update surface is not on the public free MCP server. Token impact depends on how much data the query removes.

Middleware and token use

Claim. Doing file work outside the model can reduce unnecessary token use.

Mechanism. Parsing, filtering, calculation, validation and file construction happen in OfficeMaker middleware, so the model works with relevant structured state rather than a rendering program or a screenshot loop.

Evidence. Token-efficiency workflow write-up (15 April 2026)

Limitation. The published figures are reasoned estimates for one PowerPoint workflow, about 60,000–90,000 tokens for a Python/PptxGenJS path with image review versus about 20,000–30,000 for OfficeMaker on that example. Savings vary by document size, workflow and model. They are not a billing guarantee.

Explore corporate deployment · AI Excel generator

Questions

Can OfficeMaker refresh Excel reports automatically?
Yes, on organisation tenants. Automations can generate Excel workbooks from a schedule, web form or API while keeping the same structured layout. The public free MCP server cannot register those scheduled jobs.
Can the free sandbox query an existing workbook?
No. The public free MCP server creates a new file and can discover schema and run gap analysis. Query, filter, merge and edit of an existing workbook are authenticated document-service verbs.