TL;dr: In its September Workspace Drop, Google added Sheets canvas, which turns a spreadsheet into an interactive mini-app with one click. You describe what you want in plain language, Gemini builds the layout, and it lives as a tab in the workbook, synced to the sheet in real time. Google's own example was a financial scenario planner, so I built one to test it: a mortgage-rate scenario planner ahead of the Bank of Canada's October 28 decision. The setup, the prompting, and the honest limits are below, and there is a template at the end with the full working model.

Let me set the scene, because every finance person has lived this. You build a scenario model in a grid, careful with the inputs, everything labeled, and then you send it to a stakeholder who scrolls straight past the input cells and asks, "so what happens if rates go up?" The model is right there and they cannot see it, because a grid of cells is a builder's interface and they are not the builder. We have all tried to solve this with cleaner formatting, with instructions in bright yellow cells, with a separate summary tab, and none of it really works, because the problem was never the formatting. The problem is that the person making the decision does not want to operate your model. They want to move a lever and see the answer.

That is what canvas is for. It sits in your workbook as a tab, next to your sheets, and it is a small interactive app wired to your cells. You prompt Gemini in plain language, something like "build a scenario planner with sliders for the mortgage rate under three Bank of Canada scenarios," and it generates the layout with controls bound to your input cells. When someone moves a slider, the cell changes, your formulas recalculate, and the results update. It shares like any other sheet, which means you can send the same workbook to the stakeholder and they get the levers while you keep the grid.

Here is the worked example, because the specifics matter more than the concept. The template at the end has all of this built, so download it and follow along. The model itself is simple on purpose. There is an inputs block with three named cells: Balance for the outstanding mortgage balance, CurrentRate for the current variable rate, and AmortYears for the remaining amortization. Naming the cells is the single most important setup step, and I will come back to why. Below the inputs sits the scenario grid with four columns: your current rate, the Bank of Canada holding, a 25 basis point hike, and a 50 basis point hike. Each column shows the effective rate, the monthly payment from a standard PMT formula, the total interest over the remaining amortization, and the change in monthly payment versus today. Nothing exotic, just the math every analyst already knows how to build.

Now the canvas part. With the model built, you open a new canvas tab and prompt it. The prompt that worked for me was specific about the bindings: "Build a scenario planner with a slider for the mortgage rate from 3% to 7% in quarter-point steps, bound to the CurrentRate cell, and show the monthly payment and total interest from the scenario grid for the hold, plus 25, and plus 50 scenarios." Being explicit about the cell names is what keeps Gemini from guessing. When I first prompted it loosely, it built a beautiful planner wired to the wrong range, because it picked up a formatted total row instead of the inputs. The fix was boring and effective: name your input cells, reference them by name in the prompt, and check the bindings before you show it to anyone. If it misreads a range, do not re-prompt five times hoping for better luck. Delete the control, name the range, and prompt once with the name in it.

The honest limits, because this is where practitioner judgment comes in. The canvas is a view layer, and your formulas still live in the grid, which is exactly as it should be. But that split creates real questions you need to answer before this goes anywhere near a decision. First, when someone moves a slider in the canvas, it writes back to the cell, so your inputs are being edited by people who never see the grid, and you need to decide whether that is acceptable for your process or whether the canvas should be a copy while the model of record stays locked. Second, the audit trail gets murkier: version history shows the cell changed, but it does not show which canvas control changed it or why, so if your team has review standards, the canvas does not automatically meet them. Third, anything Gemini generates can misbind, as I found out, which means the canvas tab needs the same kind of checking you would give a junior analyst's first draft: trace every control to its cell before you trust it. None of this is a reason not to use it. It is a reason not to confuse the interface with the model.

So does it earn a place? From where I sit, yes, but in a specific seat. Canvas is the best stakeholder interface I have seen for a spreadsheet model, because it finally separates the levers from the machinery without duplicating the file or rebuilding anything. But the grid remains the system of record, with the formulas, the version history, and the review trail your controller expects. Use the canvas to let people touch the model. Keep the model where your controls are.

The template is below. It has the full mortgage-rate scenario planner: named input cells, the four-scenario grid, and notes on each step so you can build the canvas tab yourself in about ten minutes. Copy it, wire up your own numbers, and run it past the toughest reviewer on your team before it goes anywhere.

Template: template: Mortgage rate scenario planner (Google Sheets, make a copy to use it)

- Francois