How much will we sell next month? Check the plain guess first
Updated October 2026
A forecast gives you one number for next month. Before you believe it, check whether just saying 'the same as last month' would have done better.
A forecast gives you one number for next month. Before you trust it, ask a simple thing. Would just saying "the same as last month" have done better?
We ran that test on US sales of a USD-reporting eCommerce apparel brand. We hid the last 12 months from a standard forecast. Then we asked the forecast, and three plain guesses, to predict those 12 months. The forecast missed by 27.22 percent on average. Saying "the same as last month" missed by 18.56 percent. The forecast came last.
Our position: a forecast earns trust only by beating the plain guess. Here, it did not.
That is one brand and one test. It is not a record, and it is not a rule about forecasting.
The question and the data
The question is the one every owner asks: how much will we sell next month?
We used units sold by one brand. The data has daily rows, so we added them up into whole calendar months. That gave 46 months, from May 2020 to February 2024. We learned from the first 34 months. We held back the last 12 to score the guesses.
Think of it like a quiz with the answers covered. Each method sees the first 34 months. Then it writes down its guess for the next 12. After that, we lift the cover.
One thing to know. 2020 and 2021 sit inside the pandemic. The forecast has no idea what a pandemic is. It only sees the numbers.
Four ways to guess next month
Three guesses need no model at all. You could do them on a napkin.
- Same as last month. Next month will match the last month you saw.
- Same month last year. Next month will match the same month one year ago.
- Average of the last 3 months. Next month will match the recent average.
The fourth is the forecast. It is a seasonal trend model, called Holt-Winters. It tracks where sales sit now, which way they are moving, and a yearly shape that repeats.
How they did
On the 12 held-back months, we measured the average miss as a percent of what really sold. Lower is better.
| Guess | Average miss (percent) | Average miss (units) |
|---|---|---|
| Same as last month | 18.56 | 1,487 |
| Same month last year | 21.51 | 1,833 |
| Average of the last 3 months | 24.51 | 1,912 |
| The forecast (Holt-Winters) | 27.22 | 2,215 |
The forecast had the biggest miss of the four. The tool itself called its answer "low-confidence". We agree, and we report it as a loss.
Basis for every figure: one brand, units per month, 46 months, 12 held back, one run.
What the forecast said for next month
We then fitted the forecast on all 46 months. For March 2024 it said about 7,143 units. The last real month, February 2024, was 6,528. Because this forecast lost its test on this brand, treat that number as one view, not a plan.
It also gave a range. We expect between 4,527 and 9,759 units. We would be surprised to land outside that. The range is wide, about 2,600 units either side of the middle. The range is a fair warning, but it is not the whole story. Real sales can still surprise you.
The forecast found a yearly shape. January was the highest month and June was the lowest. From lowest to highest, the swing was about 5,087 units. It also found sales sliding down by about 336 units each month. A forecast is a guess built from history. It is not a promise.
What this means for you
- Use the range, not the single number. One number looks sure. A range shows how unsure it is.
- Run the plain guess beside any forecast. Keep whichever one misses least on months you held back.
- Go easy on the yearly shape when history is short. Here the forecast had only 34 months to learn from. A yearly shape can be hard to pin down from so few.
- Do not stop forecasting. Just make it prove itself first.
Other ways to guess next month
Take next month's sales. You can say it will match last month, match the same month last year, match the recent average, or let a seasonal trend model project it. Each answers a different question, and the right one depends on your series. On this brand's one test, the plain guesses had the lower error.
- The same as last month. Right when: Your series is steady with little yearly shape. It believes that next month will look like this month, and nothing else matters.
- The same month last year. Right when: Your series has a strong yearly shape that repeats. It believes that the yearly shape repeats exactly, and the level has not moved.
- The average of recent months. Right when: Your series is noisy with no trend. It believes that the last few months are all equally good guides, and a month does not matter more because it is newer.
- A seasonal trend model (Holt-Winters). Right when: A trend and a yearly shape persist, and you have enough history to learn them. It believes that the trend and the yearly shape it learned will carry on. With only 34 months to learn from, a yearly shape may be hard to pin down.
How to do it on your own file
- Get one row per month. If you have daily or weekly rows, add them up into whole months. Leave out a first or last month that is only partly there.
- Use one number column, such as units sold. Do not add money in different currencies.
- Pick your plain guess first. "The same as last month" is a fine start.
- Hide the last 12 months. Learn from the rest.
- Score every guess on the 12 hidden months. Keep the one with the smallest miss.
Or let the free tool do the steps for you. It shows the forecast, its range, and the plain guesses side by side.
Questions people ask
Why did the plain guess win?
We did not test why, so we do not claim a cause. It is one brand and one test of 12 months. Another brand could go the other way.
Does this mean forecasts do not work?
No. This is one brand and one test of 12 months. Another brand could see a different result.
Is "the same as last month" always the best plain guess?
No. It suits a steady series. "The same month last year" suits a strong yearly shape that repeats. "The average of recent months" suits a bumpy series with no trend.
How many months do I need?
More is better. Our example had 46, with 34 to learn from. With monthly rows the tool needs at least 29 months to fit a yearly shape. With fewer it falls back to simpler models.
What is the "average miss"?
It is how far off each guess was, as a percent of what really sold, averaged over the 12 held-back months.
Should I trust the range?
Use it as a guide, not the whole story. It is wide here, and real sales can still land outside it.
Your turn
Does your forecast beat the plain guess? Upload a monthly file to the free forecasting tool and find out. Or email your file to [email protected].
Sources
Data: Conjura Multi-Region MMM Dataset (Anderson 2024), figshare. CC BY 4.0. Changes: one anonymised series, US sales of a USD-reporting eCommerce apparel brand, summed to complete calendar months, units as the outcome.
How we ran it
Three things to know about this reading.
Months held back
We hid the last 12 months from the forecast, learned from the first 34, and compared each method with what really sold. The forecast missed by an average of 27.22 percent of units. In units that is about 2,215 a month. The tool calls this low-confidence.
The plain guesses
We scored three plain guesses on the same 12 months. 'The same as last month' missed by 18.56 percent. 'The same month last year' missed by 21.51 percent. 'The average of the last 3 months' missed by 24.51 percent. The forecast had the largest miss of the four.
One test, one brand
This is one brand and one 12-month test, not a track record. 2020 and 2021 sit inside the pandemic, and the forecast has no notion of it. A forecast is a guess from history, not a promise.