Forecast Your Numbers with Google’s TimesFM
Pick an example or paste your own sales, patients, stock or cash figures. See the weeks ahead, with a likely range and an accuracy check.
- Free, no sign-up
- Your data is not stored
- Results in seconds
- 1. Add your numbersPick an example, paste two columns (date, value) or drop a CSV file.
- 2. Choose how far aheadTap 1 week, 4 weeks or 3 months, or type any number of steps.
- 3. Read the resultThe cyan line is the forecast. The shaded band is where 8 in 10 values should land.
- History
- TimesFM forecast
- Likely range (80%)
- Simple repeat of last cycle
- Actual (hidden)
See the numbers
How to read your forecast
- 1. The cyan line
- The most likely value for each future day, week or month.
- 2. The shaded band
- 8 times in 10 the real number should fall inside it. A wide band means the future is less certain, so plan with a buffer.
- 3. The dashed line
- A simple repeat of your last cycle, the way most spreadsheets guess. Use it to judge whether the AI adds value.
- Accuracy check
- Tick it to hide your latest rows. The page forecasts them and shows the average error for both methods.
Tips for better forecasts
- Use one row per day, week or month, with no gaps
- Give at least two full cycles: 8 weeks of daily data or 2 years of monthly data
- Keep one kind of number per sheet, for example sales of one shop
- Remove totals and blank rows before pasting
- Forecast no further ahead than about a quarter of your history
- Re-run each week as new numbers come in
Forecasts we can build into your business
TimesFM is a forecasting model from Google Research, trained on a very large collection of real and synthetic time series. It reads your history and predicts what comes next, without training on your data.
Retail and distribution
Daily sales and stock per shop or product, so reorders arrive before shelves empty.
Hospitals and clinics
Patient numbers by day and department for rosters and beds.
Restaurants and hotels
Covers and room nights to plan food orders and shifts.
Finance teams
Collections and payments to see cash gaps weeks early.
Marketing
Visits and enquiries to set ad budgets by season.
Factories and buildings
Power, water and machine readings to plan maintenance and cut peak bills.
Frequently asked questions
What is TimesFM?
TimesFM is a time-series foundation model from Google Research. It forecasts numbers over time, such as sales or visits, without being trained on your data first.
Is my data stored?
No. Pasted or uploaded data is read in your browser and sent only to our forecast server to compute the result. It is not saved or used for training.
Which file format works?
Any CSV or spreadsheet copy with a date column and a number column, or just one column of numbers. Headers and extra columns are skipped; the last number on each row is used.
How much history do I need?
At least 16 rows. For weekly or seasonal patterns, give two or more full cycles, for example two years of monthly figures or eight weeks of daily figures.
How accurate is it?
It depends on your data. Tick the accuracy check: the demo hides your latest rows, forecasts them, and shows the average error next to a simple repeat of the last cycle.
Can you run this on our own server?
Yes. We can set up forecasting on your hardware or cloud and connect it to your sheets, ERP or database. See Native AI development.