Forecast your sales in Google Sheets, no code

Hi, it's Ivo.

What you need first

1. A Google Sheet with your numbers by date. Sales, orders, visitors, covers: any number you count every day or every week.
2. A Google Cloud project with BigQuery. Connected Sheets asks for one the first time you connect. You set it up once.
3. History. The tool runs with as few as 3 data points, but that tells it almost nothing. For seasons (summer vs winter, December vs August) a year or two of history is better.

About the project: Google says that without a billing account you can work in the BigQuery sandbox, which doesn't ask for a credit card. I haven't confirmed that "Create a forecast" runs in the sandbox. If it asks you to turn on billing, that's the reason. Read the BigQuery pricing page before you connect a card.

The sheet template

Two columns. One row per day (or per week). Nothing else on the sheet.

Rules for the template:

1. Column A is a real date (Format > Number > Date), not text like "Mon 13".
2. Column B is a plain number: no currency sign, no "k", no notes in the cell.
3. No total rows, no blank rows in the middle, no merged cells.
4. If you were closed on a day, put 0 and write the reason somewhere else, so you remember why the line dips.
5. Keep the header row exactly one line: date and sales (or your own names).

The no-code steps in Google Sheets

1. Open the sheet. Go to Data > Data connectors > Connect to BigQuery. (One Google help page calls the menu Data > Data analytics. If you don't see Data connectors, look there.)
2. Pick your Google Cloud project. You can also import your sheet into BigQuery straight from Sheets.
3. In the connected sheet, click Advanced analytics > Create a forecast.
4. Date column: pick your date column and how to group it (day, week or month).
5. Forecast column: pick your number column, aggregated by Sum.
6. Enter horizon: how many periods ahead. Start small: 8 weeks, or 30 days.
7. Tick Include historical data and Include prediction levels. The default level is 95%.
8. Click Create.

You get a new sheet with the prediction for each period, a lower bound and an upper bound, plus a chart of your history with the prediction after it.

How to read the range

• The middle number is the model's best guess for that period.
• The lower and upper bound are the range. At 95%, the real number should land inside it most of the time.
• The band gets wider the further ahead you look. That's normal. Week 8 is less certain than week 1.
• Plan your stock and your shifts between the low line and the high line. The low line is what you're fairly sure to see. The high line is a busy week you'd want to be ready for.
• If the lower bound goes below zero, ignore that part. It means the history is too short or too noisy.

When not to trust it

1. Promotions. The simple version only sees your history. It doesn't know about next month's offer, so a promo week will jump out of the range.
2. New products. No history, no line to continue. Under 3 data points it won't run at all.
3. One-off events. A closure, a renovation, a new competitor across the street, changed opening hours. The model doesn't know they happened unless your numbers show it.
4. Very short history. A few weeks can't show a season.

A quick check before you rely on it: hide your last 8 weeks, run the forecast on the rest, and compare it with what really happened.

It's a planning tool. Google itself says the results are comparable to classic statistical methods like ARIMA. It is not financial advice, and it won't replace your own judgment or your accountant.

A note on the license

• TimesFM's code on GitHub is Apache-2.0. The README says the open version is not an officially supported Google product.
• The TimesFM 3.0 model weights you can download are licensed for non-commercial, non-production use only. Running the downloaded model for your business is not allowed.
• Business use of TimesFM is allowed through Google Cloud: BigQuery and Google Sheets (Connected Sheets). That's the path on this page.

As of October 2026. Menus change, so if a step looks different, check Google's help page.

Sources: support.google.com/docs/answer/16923039 · support.google.com/docs/answer/9702507 · docs.cloud.google.com/bigquery/docs/timesfm-model · docs.cloud.google.com/bigquery/docs/sandbox · github.com/google-research/timesfm