Why Gemini in Google Sheets Still Can't Run Your Production Budget

That gap is not a criticism of Gemini. It is a description of what happens when an AI tool is asked to compensate for something the underlying file was never designed to provide.

This post is for production teams already inside Google Workspace. The reasons to be there are real: your whole organisation lives in it, collaborators can access files without a new login, and the cost per seat is hard to argue with. The question is not whether Google Workspace is a sensible place to run a production business. The question is whether adding Gemini to a spreadsheet turns that spreadsheet into a financial system. It does not.

What Does Gemini in Sheets Actually Do?

Gemini has read access to your file. It can surface values, describe what it sees, and answer questions about cell content. It can generate formulas, write scripts via Apps Script, and summarise ranges of data.

This is useful for many tasks. It is not sufficient for production finance. The reason is not that Gemini reads spreadsheets poorly. The reason is that a spreadsheet is not a financial system, and adding an AI to it does not change that. The AI inherits the spreadsheet’s structural limitations.

Does Gemini Know Whether Your Budget Adds Up?

A production budget is not a neutral collection of numbers. Every line item holds a position in a calculation chain. In DACH productions, that chain is already complex: a single crew member’s cost splits across direct fee, Arbeitgeberanteil (AGA, roughly 21%), KSK-Abgabe (5.0% for eligible freelancers), and Überstundenzuschläge. Employment classification — whether a crew member is angestellt or selbstständig — determines which Sozialabgaben components apply. A system that treats all crew the same way produces the wrong total.

Those relationships exist in a spreadsheet as formulas a human wrote. Formula chains break silently. A bad paste, a misaligned reference, a copied section that did not carry the right offsets — any of these can disconnect a source value from the total it was supposed to feed. The top-sheet number does not update. Nothing flags it.

Gemini reads what is in the cells. It has no way to verify that the relationships between those cells are intact. Whether a given cell reflects the correct, current calculation is unknowable to the AI, because the file has no schema — no enforced definition of how its values relate to each other. That is not a Gemini problem. It is a file problem.

Can Gemini Tell You Which Version Is the Current Budget?

Production budgets move through states: draft, submitted, locked, approved, superseded. A figure that is correct in a draft may be wrong after a shooting day is added. A line pending sign-off is materially different from one that has been authorised.

Google Sheets has no concept of document state as a financial concept. What it has is version history: a log of individual cell changes. That log does not record that a budget transitioned from draft to approved on a specific date. It does not record which file or tab represents the live document versus a superseded revision.

When two versions of a budget exist in different files or tabs, an AI reading the open file has no basis to determine whether it is the current document. It sees the file in front of it. The question “is this the approved version?” requires information the file does not hold. That is a data model problem. The AI can only surface what the file gives it.

Is File Access the Same as Financial Authorisation?

In Google Sheets, access is file-level: view, comment, or edit. A department head may have edit access to update their own section. A production accountant has edit access to work across sections. These are file permissions. They are not financial authorisations.

When a line item changes but the authorisation trail is absent, no audit query can distinguish a sanctioned edit from an unilateral one. The file access log records who changed a value and when. It does not record whether that change was approved by a line producer, authorised within the scope of a cost report, or aligned with a purchase order.

In production finance, authorisation is a first-class concept. A change to an actuals line is not the same as a change to a budget note. It requires a different class of permission and an audit record that downstream reporting can rely on. File-level access control does not model this distinction. Gemini reading a shared file inherits none of it.

What Does Production Finance Infrastructure Actually Require?

The three failure modes above are not edge cases. They describe the daily reality of any production with more than one person touching numbers. They are not solved by a better prompt or a more capable model. They are solved by a different data model.

Production finance infrastructure that addresses these failure modes at the architecture level requires three things.

First, real-time multi-user collaboration tied to role-based access that maps to actual production roles — not to file permission tiers. Every collaborator’s access should reflect their function, and the system should enforce what each role can do with each class of financial record.

Second, document state built into the data itself. The system needs to know that a budget has states, that transitions between those states are meaningful events, and that a figure in a draft budget is not the same object as the same figure in an approved budget. This is a data model requirement, not a tagging convention.

Third, financial authorisation as a concept distinct from file access. Reading a line item is not the same as changing it in a financially binding context. The system should enforce this at the record level.

Splinde is built around these requirements. Real-time multi-user collaboration with role-based access has been live since January 2026. Admin and non-admin role permissions went live in February 2026. Scenario Management — the ability to model and compare budget versions as structured data, not as separate files or tabs — has been live since March 2026. Core product capabilities including AGA, KSK, and Overtime AddOns handle DACH employment classification directly in the calculation layer, so the system applies the right Sozialabgaben components based on crew classification rather than relying on a formula a user configured.

Approval workflows and PO matching are part of the architecture Splinde is built around. We will share capability detail as those features become available for production teams to use directly.

Is Gemini the Problem, or Is the File?

The appeal of Gemini in Sheets is genuine. It is already in your ecosystem, requires no migration, and gives your team something that looks like intelligence on top of a tool they already know.

The problem is structural. Gemini is a layer built on top of a file. The file still has no schema. The file still has no state. The file still has no concept of financial authorisation. Adding an AI layer does not address those absences — it inherits them.

We covered the structural argument in depth in our analysis of how LLMs interact with spreadsheet environments: LLMs vs. Spreadsheets: A Deep Analysis. The core conclusion applies here: the model you point the AI at is the constraint, not the AI itself. Gemini is not the problem. A file being asked to behave like a financial system is the problem.

Production finance needs a data model that was designed for production finance. That is what changes the result — whether or not there is an AI layer on top of it.

> Ready to see what production finance infrastructure actually looks like?

> Book a 30-minute demo and see how Splinde handles these scenarios at the data model level — with role-based access, structured records, and budget state that actually tracks.

> Book a Demo Start Free

FAQ

Can Gemini in Google Sheets manage a production budget?

Gemini can read and summarise data in a Google Sheet, generate formulas, and answer questions about cell values. It cannot verify that the budget’s internal calculation chain is intact, distinguish between draft and approved versions of a budget, or enforce financial authorisation on edits. These are structural limitations of the file-based system, not limitations of Gemini itself.

What does a production finance system need that Google Sheets does not provide?

Three things: a schema that defines the relationships between line items so totals are always derived correctly; a workflow model that tracks document state (draft, approved, locked, superseded) as a first-class data concept; and financial authorisation logic that goes beyond file-level access permissions. A spreadsheet has none of these by default, and adding Gemini does not supply them.

Why does DACH production finance add complexity that spreadsheets handle badly?

DACH productions must apply different Sozialabgaben components depending on whether crew are angestellt or selbstständig. The Arbeitgeberanteil (AGA, roughly 21%) applies to employed staff; the KSK-Abgabe (5.0%) applies to eligible freelancers. A spreadsheet can encode these rules as formulas, but those formulas can be broken by a copy-paste or a manual override with no warning. A purpose-built system applies the correct calculation based on the crew classification record, not based on a formula someone wrote and someone else can accidentally overwrite.

Why is real-time collaboration in Google Sheets not enough for production finance?

Google Sheets real-time collaboration allows multiple users to edit the same file simultaneously. But it has no concept of which edits were financially authorised, which version of the budget is the current approved document, or whether a change to an actuals line required sign-off before downstream records could depend on it. Production finance requires these distinctions at the data model level, not at the file access level.