How to automate crypto technical analysis: what a year of BTC alerts shows

To automate crypto technical analysis, write your setup as a yes-or-no rule, check it on every closed candle with a script or an alert tool, and send an alert only when the answer changes. We ran the textbook version of that rule on a year of hourly Bitcoin candles: it produced 553 new signals, about 11 a week, and 4 in 10 of them were gone within 3 hours.

The code is the easy part. The hard part is deciding which changes are worth a message.

The four steps

1. Get candles. Pull open, high, low, close and volume for each candle from your exchange's API or a market data provider. Use one exchange, one pair and one timeframe to start.
2. Write the rule as a test. "EMA 20 is above EMA 50 and RSI 14 is above 50" can be checked. "BTC looks strong" cannot.
3. Check it when the candle closes. Run the check once per candle, a minute after the close. A candle that is still open can show a signal that is gone at the close.
4. Alert on a change, not on a state. Send the symbol, the time, the price and the readings when the rule turns on. Stay quiet while it stays on.

What the rule did on a year of BTC

The rule: on each closed 1-hour BTC-USD candle, the setup is bullish when EMA 20 is above EMA 50 and RSI 14 is above 50, and bearish when both point down. Anything else is no setup. It is the rule most automation guides use as their first example.

From 25 October 2025 to 22 October 2026 (8,690 closed hours), BTC was in a bullish setup for 3,078 hours, a bearish setup for 3,234 hours and no setup for 2,378 hours. How many alerts you get depends on when you send them:

A script that checks every hour and has no memory of what it already sent gives you the first row: about 17 messages a day. Storing the last state and alerting only on a change brings that down to about 1.5 a day.

Short-lived signals are the bigger problem. Of all the setups in the year, 122 lasted one hour and 230 (41.5%) were over within 3 hours. Adding the 4-hour trend filter cut the number of alerts by about a third, but 43% of the remaining setups still ended within 3 hours. A filter makes the stream smaller. It does not make each signal more lasting.

One real day: 29 July 2026

BTC closed every hour between about $63,400 and $64,600 that day, less than 2% from low to high. The rule still started six setups:

If each alert had been a trade, that is six entries in a range a person watching the daily chart would call flat.

Is the signal a prediction?

Not on its own. We checked where BTC was 24 hours after each new setup:

• After a bullish setup started, BTC was higher 24 hours later 47.8% of the time (278 signals).
• After a bearish setup started, BTC was lower 24 hours later 45.3% of the time (274 signals).
• For any hour picked at random, BTC was higher 24 hours later 50.1% of the time.

That is no better than chance, before fees. Automating the analysis saves you from staring at charts. It does not create an edge. Test any rule on its own history, then on live prices with simulated money, before you let it place an order. Our guide to testing an AI trading setup before risking money shows how many trades that takes.

Three changes that make the alerts useful

• Remember the last state. Alert when the setup starts or flips. Stay quiet while it holds.
• Say why in the message. "BTC bullish at 11:00 UTC, close $64,437, RSI 56, EMA 20 above EMA 50" can be checked in a second. "BTC buy signal" cannot.
• Decide how long a signal must hold. If 4 in 10 setups end within 3 hours, waiting for a second closed candle, or a higher timeframe, removes some noise. Count what it costs you in late entries before you keep it.

Doing it without code

If you would rather not run a script, TradingWizard bots do the checking for you. Pick a market and an asset, set up a bot with a strategy such as Moving Average Cross, MACD or Fibonacci, and watch what it does. Each bot shows what it read, why it traded or waited, and when it checks next. Demo trades use simulated money. No real orders are sent.

Set up a Moving Average Cross bot on BTC in /wiz and compare its next decision with the rule above.

How we counted

Coinbase BTC-USD hourly candles from the public API, 23 October 2025 to 22 October 2026: 8,750 candles, 10 missing from the feed. The first 60 hours were used to warm up the averages. EMA 20 and EMA 50 use the standard smoothing; RSI 14 uses Wilder's smoothing. A setup counts once its 1-hour candle has closed. The 4-hour trend uses the last fully closed 4-hour candle, with EMA 20 above EMA 50 meaning up. The 24-hour check compares the close at the signal with the close 24 hours later and leaves out fees.

For alerts that run while you sleep, read how to build a crypto alert setup for 24/7 monitoring.