Automation is the last step, not the first.

01Step 1: write the rule down

Before a platform, before a broker, before any code: state the strategy in writing. A complete specification covers the entry condition, how the instrument or strike is chosen, the position size, the stop-loss and where it lives, the exit condition, and what happens when something goes wrong.

That last category is the one people skip and the one that costs money. What should the system do when a fill does not confirm, when the connection drops mid-position, when an order sits unfilled, when the market gaps through the stop? Those are not edge cases. Over a year of trading they are certainties.

The test of a written rule is simple: if two people read it and would place different orders, it is not finished. A rule that says "enter when momentum looks strong" cannot be tested, because in hindsight it always resolves in your favour.

02Step 2: backtest it honestly

A backtest is not evidence that a strategy works. It is evidence that a strategy would have worked, which is a weaker claim than it sounds and is routinely weaker still than it is presented.

Three things separate a useful backtest from a flattering one. Costs have to be applied on both sides of every trade, not just once: slippage on entry and on exit, brokerage, statutory charges, and the cost of any hedge. Expiry calendars have to be the real ones, because option strategies behave completely differently depending on which day the contract dies. And the data has to include periods the strategy was not designed against, because a rule tuned on the same stretch it is tested on is not being tested at all.

We go through this in detail, including the test that shows whether your selection process was picking signal or noise, in backtesting that tells the truth.

03Step 3: no-code or custom

This is the fork that decides most of what follows, and there is no universally correct answer.

Start no-code if your strategy is a recognisable structure with standard entries, exits and stops. Platforms handle that well, you will be testing within the hour, and you will learn what you actually want far faster than by specifying it in the abstract. For a large number of traders this is the destination, not a waypoint.

Move to custom code when you hit a wall the builder cannot express. In practice the wall is usually one of three things: an entry condition that depends on something the platform does not expose, execution behaviour you need control over such as how stops are placed with the broker or how re-entries are handled, or the simple fact that you want to own the system rather than rent access to it. We compare what each type of platform can and cannot do in algo trading platforms in India.

04Step 4: clear the regulatory setup

This step moved from a formality to a real dependency, and it is the one most guides written before 2026 get wrong.

Since 1 April 2026, SEBI's retail algo framework is fully in force. Every algo order carries an Algo ID issued by the exchange for a strategy your broker has registered, and API access is expected from a static IP address whitelisted in advance. The practical consequences are that your algo needs somewhere fixed to run from rather than a laptop on home broadband, and that you should confirm your chosen broker is still onboarding new retail API clients before you commit to it.

What has not changed: automating your own account with your own capital does not make you a regulated entity, and you do not need SEBI registration to do it. The full detail is in SEBI algo trading rules.

05Step 5: forward test

Run the finished system against live market data with no money behind it. Every assumption in the backtest is now exposed to a market that does not care about them.

What forward testing catches is not usually the signal. It is everything around the signal: orders that fill at a worse price than modelled, a feed that stalls for ninety seconds, a stop that was placed locally instead of with the broker and therefore did not exist when the connection dropped. A month of forward testing is cheap. Discovering these things with capital deployed is not.

This is also the point where you learn whether you can actually watch the thing trade without interfering, which is a genuine requirement and not a character test you are guaranteed to pass.

06Step 6: go live small

Start at the smallest size the instrument permits. One lot. Then leave it alone until live results and test results agree over a meaningful number of sessions.

Size is the last variable to change and the easiest to change too early. A strategy that behaves as expected at one lot for two months has told you something. The same strategy scaled up after two good weeks has told you nothing, and has put real money behind the part of the sample that was luck.

07What a year actually costs

Trading capital is separate from running costs, and running costs are what people fail to plan for. Rough annual figures for a single retail setup:

Call it 20,000 to 40,000 rupees a year for infrastructure before any development work. The number matters because it sets a floor on the capital that makes sense: running a 30,000 rupee annual cost against a 50,000 rupee account is a losing arrangement before the strategy places its first order.

08Five failures that end most attempts

• Testing without costs. The most common single reason a live system does not resemble its backtest. Gross and net are different strategies.
• Treating execution as an afterthought. The signal gets months of attention and the order handling gets an evening. Live losses come overwhelmingly from the second.
• Optimising until it looks good. Enough parameter combinations and something will look excellent by chance alone. If you searched, you have to test whether what you found survives on data you did not search over.
• Scaling on a good month. Size increases should follow evidence, and one month is not evidence.
• Interfering. Overriding the system during a drawdown converts a tested strategy into an untested one, usually at the worst moment.

09Questions people ask