Average Can Mislead. The Median Shows What Most Bots Did

Last updated: October 10, 2026.

The average return adds up every bot's result and divides by the count, so one exceptional bot pulls it up for everyone. The median is the result of the bot in the middle: half did better, half did worse. When the average is much higher than the median, a few big winners are carrying the number, and the median is the better guide to what a typical bot earned. In H1 2026 platform data, the average ran 2.8 to 14 times above the median depending on bot type.

Most "bots made X%" headlines quote the average. That's not a lie. It just answers a different question from the one a new trader is asking.

Key facts at a glance

• Average answers: how much did all the bots make together, divided by the count?
• Median answers: what did a typical bot make?
• Outliers move the average and barely touch the median. One bot at +300% among a hundred moves the average from 1% to about 4%. The median stays at 1%.
• Leverage widens the gap. On one multi-exchange platform in H1 2026, COMBO bots averaged +15.77% while the median closed bot made +1.76%.
• The gap works both ways. A few large losses can drag the average below a positive median, as with spot DCA bots in the same period.

100 bots, one outlier

Take 100 bots. Ninety-nine of them made exactly +1%. One caught a move and made +300%.

A report that says "bots averaged almost +4%" is accurate. It describes a result that 99 of the 100 bots never got. If you launched one of these bots, the median tells you what to expect. The average tells you what the best one did, spread thin across everyone.

Now flip it. Ninety-nine bots made +1% and one was liquidated at −100%.

The average says the group lost money. The typical bot didn't.

What real bot data looks like

Platform data. Bitsgap bot performance research, H1 2026, closed bots on the platform.

Among COMBO bots that closed in profit, the average was +21.95% and the median +7.89%, a gap of 2.8x. Leverage is the pattern behind the biggest gaps. It multiplies the winners far more than it moves the typical result, so leveraged bot types show the widest spread between the two numbers.

The same thing happens with time. Across Q3 2026, the median bot ran 1.2 days before it was closed, while the average was 9.5 days. A small group of bots running for weeks pulls the average up eightfold.

A case study read both ways

In July 2026, one trader ran four bots through a falling market: three COMBO shorts at 10x and one spot DCA. Together they made $8,678 on $7,581. One of the four, the WLD short, produced about $7,098 of that.

Without the WLD bot, the other three made roughly $1,580. Still a good ten days, and a completely different headline. An average of four bots described a result no single bot produced. The full breakdown is in +$8,678 in a Falling Market.

When the average is fine

Use the average when results are tightly grouped. If the average and the median are within about 1.5x of each other, both numbers say roughly the same thing. Spot grid bots in a steady range often look like this.

Rely on the median when:

• the average is more than twice the median;
• the sample includes leveraged bots;
• the report is about a short period or a handful of bots;
• a single result is described as "up to X%."

Seven questions to ask any bot results report

1. Is there a median next to the average? If only the average is shown, assume a few outliers are doing the work.
2. How many bots are in the sample? A result from 12 bots says little. One from 16,734 says more.
3. Real funds, demo, or both? Demo bots can run settings nobody would fund. A report should say which it counts.
4. Closed bots only, or open ones too? Open bots carry unrealized PnL that can still swing.
5. Were outliers or tiny deposits removed? Cleaning is fine if it's disclosed.
6. What did the market do in that period? A short bot in a crash and a long bot in a rally both look brilliant.
7. What share closed at a loss? A win rate without the size of the losses is half a picture.

A report that answers all seven is one you can plan around. What Actually Worked in 2026? and the September 2026 numbers show both figures side by side.

How to use this on your own bot

Run the setup you're considering through a backtest, then look at it the same way:

• Judge it by the typical trade, not the best trade.
• If one trade or one coin makes most of the profit, the setup depends on luck repeating.
• Run several small bots instead of one large one, and compare their median with their average after a few weeks.

Test your setup and read it like a report: median first.

Frequently asked questions

What is the difference between median and average return? The average adds every result and divides by the count, so large outliers move it. The median is the middle result when all results are sorted: half did better, half did worse. For trading bots, the median is closer to what a typical bot earned.

Why do trading bot platforms show average returns? The average is easy to compute and usually looks better, because a few large wins pull it up. It's accurate for the group as a whole but overstates what most individual bots made. Reports that show the median alongside it are easier to plan around.

Why is the average so much higher than the median for leveraged bots? Leverage multiplies large winning moves far more than it changes the typical result. In H1 2026 data from one platform, COMBO bots averaged +15.77% while the median was +1.76%, and DCA Futures averaged +9.09% against a +0.64% median.

Can the average return be negative while most bots made money? Yes. A few large losses can pull the average below zero while the median stays positive. In H1 2026, spot DCA bots on one platform had an average of −0.10% and a median of +0.01%.

Which number should I use to set expectations for my bot? The median, especially when the average is more than twice as high, the bots use leverage, or the sample is small. Use the average only when the two numbers are close.