AI Trading EA Results: +15.6% This Month, 47% Win Rate for the Year

My AI trading EA is up 15.60% this month. Thirty-seven trades, 67% of them winners.

Now let me spend the rest of this article talking you out of caring about that number, because a good month is the single most misleading thing a vendor can put in front of you. I would rather hand you the whole picture than the flattering corner of it.

Then I will give you the drawdown, the loss profile, and the numbers you should actually plan around.

Every number from the account, not the good ones

This is a real account, in USD, running live and forward tested in public. All figures below are the account as it stood on 19 August 2026. Numbers move daily, so treat these as a dated snapshot and check the live Myfxbook account before you decide anything.

The headline figures: total gain since going live +54.01%, maximum drawdown 16.25%, 353 trades. Balance $10,852.77, with the account's highest point at $11,010.01 earlier this month.

This month: +15.60%, 37 trades, 67% win rate.

This year: +56.94%, 340 trades, 47% win rate. Read that against the monthly figure and you already have the point of this article.

The trade-level numbers, which matter more than any of the above: profit factor 1.28. Average win $103.58, average loss $75.42. Expectancy $10.78 per trade. Best trade $587.90, worst trade $170.04 loss. Average trade length 4 hours 14 minutes. Longs won 72 of 148 (48%), shorts won 98 of 205 (47%). Sharpe ratio 0.24.

What those numbers actually mean

Most people scan a stats page for the gain and the win rate. Those are the two least informative fields on it. Here is how to read the rest, and this is worth knowing whatever EA you are evaluating, mine included.

Profit factor 1.28

Gross profit divided by gross loss. For every dollar the system lost, it made $1.28. That is a real edge and it is a modest one. Anyone showing you a profit factor of 4 on a live multi-hundred-trade forward test is showing you either a very short sample or something that has not met a bad market yet. A 1.28 across 353 trades is the kind of number that survives, and it is also the kind of number that means losing stretches are normal and frequent, not exceptional.

Win rate 47% for the year, 67% this month

That gap is the whole lesson. Thirty-seven trades is not a sample, it is a mood. The year's 47% across 340 trades is much closer to what you should expect to live with, and it is fine, because the system is not built on winning often. It wins $103.58 on average and loses $75.42, so the average winner is about 1.37 times the average loser. That ratio, not the hit rate, is where the profit comes from.

If you buy this expecting the 67% month to be the norm, you will be disappointed in a completely predictable way, and I would rather that disappointment happen now, for free.

Expectancy $10.78

The average outcome of a single trade. This is the number that answers the only question that matters, and it also tells you what your account size needs to be for the strategy to be worth running at all. At roughly $10 per trade on this configuration, the math only works on an account where that is a meaningful proportion of your risk unit.

Maximum drawdown 16.25%

The deepest peak-to-valley fall the account has taken. Assume the future contains a worse one, because the worst drawdown in any track record is always simply the worst one so far. The practical question is not whether you like 16.25%. It is whether you would have kept the EA running through it, because everyone who turned it off at the bottom converted a drawdown into a loss.

The number nobody publishes: a Z-score of -4.56

I have never seen a vendor volunteer this one. The Z-score measures whether the sequence of wins and losses is random. At -4.56 with 99.99% confidence, this system's results are decidedly not random in sequence: they cluster. Winners tend to arrive near other winners, and losers arrive near other losers.

The practical translation matters a lot. The bad stretches will not be politely distributed through your month. They will come in clumps, several losses in a row, and that is exactly the pattern that makes people intervene, override, or switch a system off one trade before the run turns. If you know the clustering is structural rather than a sign of breakage, you are far more likely to sit through it. That is the whole reason I am telling you.

Why a good month is the least useful number here

Three reasons, and they apply to every performance claim you will ever read.

It is the number most easily selected. Any system with a real edge and normal variance produces some excellent months. Publishing one proves the system had a good month. That is all it proves.

It is the least predictive. Short samples are dominated by variance. The 340-trade year tells you far more about what next month might look like than this month does.

It sets the wrong expectation at exactly the wrong moment. A buyer who arrives after a 15% month starts their own first month at a mental baseline of 15%, and a perfectly normal flat or negative first month then reads as failure. That buyer refunds and leaves a bad review, and neither of us wanted that outcome.

So: the month was good. The year is the number to plan around.

What is actually running

DoIt Alpha Pulse AI sends structured market context to current external AI models and uses their analysis as the trading logic, inside risk rules the AI cannot override. You choose the autonomy level: full, where it opens, manages and closes; moderate, where it opens and closes but does not modify stops and targets; or conservative, where you keep tighter manual control. It ships with presets as starting points, and you can write your own strategy logic in plain language.

The risk layer is the part I would look at hardest if I were buying: confidence thresholds that filter weaker setups, ATR-based dynamic stops, exposure control, daily drawdown protection, limits on how many ideas it will take, and a minimum reward-to-risk filter. No martingale and no grid.

Before you buy: the things that disqualify people

• You need an API key from your AI provider, and API usage costs money. That cost is real, ongoing, and controlled by your configuration. On a small account it can outweigh the edge entirely.
• You cannot backtest this in the strategy tester. The EA depends on live AI analysis, so historical simulation is not available. Forward testing is the only honest evaluation, which is why the account above is public.
• Results vary with your broker, spreads, execution, chosen preset, AI provider and risk settings. The account above is one configuration on one broker. Yours will differ.
• If you expect no losing weeks, or you plan to rewrite the prompt every time a trade loses, do not buy this. The clustering section above tells you exactly what you have signed up for.

Want to see it before you commit? The forward-tested account is public and updated automatically, drawdowns included. See DoIt Alpha Pulse AI and the live track record.

The close

The most useful thing I can tell you about a 15.60% month is that it is not the number to plan your account around. Plan around 1.28, around 47%, around $10.78 a trade, and around a drawdown deeper than 16.25% arriving at some point, because it always does.

If those numbers look like something you can live with for a year, the good months will take care of themselves. If they do not, no single month should change your mind, and any vendor who leads with one is hoping it will.

I publish breakdowns like this every week. Real numbers, including the ugly ones. Join the newsletter.

Trading involves substantial risk of loss. Past performance does not guarantee future results. All figures above are the state of a live account on 19 August 2026 and will have changed since: verify on the live Myfxbook account. API costs are paid to the AI provider and are controlled by your configuration.