By Felix – founder of unCoded, trading crypto since 2016.

I started running crypto trading bots in 2020.

Before that, I'd been trading crypto manually since 2016 – buying Bitcoin at random moments, holding through crashes, and convincing myself I was a long-term investor when really I just didn't want to sell at a loss.

When I switched to bots, I thought automation would solve the problems manual trading created.

It solved some. It created worse ones.

Five years later, most of what I now consider obvious about bot trading I had to learn by losing money. This is the article I wish someone had handed me at the start.

1. The marketing numbers are fake. All of them.

Every bot platform I evaluated showed strategies returning 40%, 80%, 150% per year. Community forums amplified these numbers. YouTube videos demonstrated them. Strategy marketplaces sold templates with impressive historical performance.

Almost none of it was real.

Not always fraudulent. Sometimes just the accumulated effect of honest mistakes – backtests evaluated at candle close that missed wick-triggered stops, Sharpe ratios calculated with wrong annualization factors, fees computed in quote currency when they were actually paid in BNB, strategies tested on a single chart during a bull market and sold as general-purpose.

The platforms weren't lying. They'd just built measurement systems that systematically overstated performance, and they had no incentive to fix what was working for customer acquisition.

It took me eighteen months and roughly €4,000 in losses to internalize one sentence: any trading number you didn't personally verify is approximately meaningless.

Today I ignore marketing numbers completely. I evaluate platforms on architecture, custody, and pricing. I test strategies with small capital before scaling. And I assume every backtest has hidden optimism until I've proved otherwise.

2. Subscription fees eat smaller portfolios alive

I ran Cryptohopper for five years at $130 per month. That's $1,560 per year.

On a €20,000 portfolio – where I started – that's a 7.8% annual cost before I've earned a single cent. To break even, my strategies had to generate 7.8% annually. Add 15% German capital gains tax on short-term holdings, and I needed roughly 10% just to match keeping my money in a savings account.

I didn't consistently generate 10% annually. Very few retail strategies do. Across five years, my net return after Cryptohopper fees was worse than if I'd simply held Bitcoin.

This is the math nobody runs before subscribing. The platform advertises a $19/month starter tier and implies any profit covers it. But the features you actually need live in the $60 or $130 tier, and those tiers require returns most retail strategies don't produce.

The brutal version: subscription platforms charge the same whether you win or lose. That's not a bug. It's the business model. Your losses don't affect their revenue – only yours.

Now I run the break-even math before any subscription. If a bot costs $60/month, I need $720 per year just to stand still. On a $10,000 portfolio, that's 7.2% before profit. Any platform whose break-even exceeds my strategy's realistic expected return is a net-negative decision.

3. Backtests lie. Yours probably does too.

Early on, I trusted backtests. A configuration showing 25% annual on two years of data? Deploy it. Then I'd watch live performance come in at 5%. Or flat. Or negative.

The reason is specific and almost universal: most retail backtests have predictable failure modes, and most retail platforms have several of them.

Candle-close evaluation is the worst. A backtest that only checks strategy logic at each candle's close misses everything that happens during the candle. A stop loss that triggers on a wick down to $95,000 and recovers to $96,500 by close shows in backtest as "stop never hit." In live trading, that stop executed and you ate the loss. Multiply across hundreds of trades and your backtest numbers stop resembling reality.

Fee calculation in the wrong asset. If fees are paid in BNB and backtests calculate them as percentages of USDT trade value without conversion, the math is systematically wrong. The error compounds with every trade.

Survivorship bias in chart selection. Strategies backtested on BTC/USDT from 2021-2023 look brilliant because BTC survived. The same strategy on DOGE/USDT over the same period might have been catastrophic, but nobody publishes those results.

Sharpe ratio annualization errors. A strategy on 1-minute bars needs √525,600 annualization, not √365. Platforms using √365 for all timeframes produce fake Sharpe ratios that look great and mean nothing.

I now only trust backtests run on 1-second base-candle data with intracandle evaluation, correct per-timeframe annualization, proper fee calculation in the asset fees are paid in, and multi-chart validation. Anything less is entertainment, not measurement.

4. Tax software costs €400 a year and nobody mentions it

My first tax year with Cryptohopper: 11,000 transactions. The CSV export ran 847 pages.

I spent three weekends trying to reconcile it manually. I gave up and paid €380 for CoinTracking. It worked – after another full day cleaning up mislabeled trades.

I've since tried Koinly, CoinLedger, TokenTax. Prices for trader-tier plans ranged from €250 to €600 per year. All of them had quirks. One couldn't handle BNB fees. Another misclassified staking rewards as trades. A third kept duplicating transaction types. Every year I spent between 8 and 20 hours manually fixing what the software got wrong.

This cost is in nobody's marketing material. It's not on any pricing page. But it's real, it's annual, and it scales directly with your bot's activity.

Budget €300-600 per year for tax software as a fixed operating cost. Factor it into break-even calculations. And choose bots that generate clean CSV exports with proper timestamps, fee attribution, and transaction categorization – a clean export versus a bad one is another 10 hours per tax year.

Nobody talks about this because it's boring. It's also expensive.

5. Your psychology is the biggest risk factor in your portfolio

The promise of automated trading is that it removes emotion.

The reality is that automation moves emotion to a different decision point.

A manual trader feels emotion on every trade. A bot trader feels emotion while watching the bot trade. The 15% drawdown on your screen feels psychologically identical to taking a 15% loss manually – except now you're watching it happen in real time, unable to intervene without stopping the strategy entirely.

I've turned off working bots during drawdowns because I couldn't stand watching them get worse. Every single time, it was wrong. The bot would have recovered within weeks. By stopping it, I locked in losses and missed the recovery. One year I calculated the cost of these emotional interventions: €3,200 in foregone returns across four incidents.

I've also left bots running too long when they clearly shouldn't have been. Market regime changes, configurations that stopped making sense, exchange events requiring reconfiguration. Emotional attachment to "the setup that worked" kept me from intervening when I should have.

Automated trading doesn't solve trader psychology. It relocates it. The decisions you make about the bot – when to pause, when to reconfigure, when to add capital, when to pull it out – are still emotional decisions, and they're often more expensive than the individual trades the bot is making.

What works: pre-commit to specific intervention rules in writing, before emotional situations arise. "If drawdown exceeds 20%, pause and reassess" is something I can follow under stress. "Feel like stopping the bot right now" is not a rule. It's panic with better grammar.

6. Pricing structure matters more than any feature list

This is the insight I resisted longest. It's also the one that made me build unCoded.

For years I compared platforms on features. TradingView webhooks? Concurrent order limits? Strategy indicators? I benchmarked every alternative capability-by-capability.

What I missed: pricing structure determines what a platform actually optimizes for.

Subscription platforms optimize for retention. They need users to keep paying regardless of trading outcomes. This means engagement features – social elements, strategy marketplaces, AI assistants, copy trading – get prioritized over features that actually improve user profitability. The platform's revenue is decoupled from whether users make money. There's no structural pressure to deliver honest backtesting, conservative risk management, or crash-safe defaults.

Profit-sharing platforms optimize differently. When the platform only makes money if users make money, profitability features beat engagement features. Honest backtesting becomes essential, because misleading users into losing strategies hurts platform revenue. Crash-safe defaults matter, because catastrophic losses kill the profit stream. Feature depth is driven by what actually generates returns.

This is structural, not moral. Subscription platforms aren't evil. Profit-sharing platforms aren't saintly. But incentive structures compound over years of product decisions. A platform that needs you to keep paying whether you win or lose ends up different from one that only earns when you do.

After five years, the conclusion became unavoidable: subscription incentives produce products optimized for the wrong thing.

Now I evaluate platforms on pricing structure first, features second. A feature-rich subscription platform with misaligned incentives is worse than a feature-limited profit-sharing platform with aligned incentives – because aligned platforms get better over time, and misaligned ones get worse.

The summary

Six lessons, learned expensively:

Marketing numbers are fake. Subscription costs stack up. Backtests lie in predictable ways. Tax software is a fixed operating cost. Your psychology is the biggest risk in your portfolio. Pricing structure shapes what a platform becomes.

If you're starting with crypto bots, internalize these before you deploy meaningful capital. The alternative is learning them the way I did – slowly, expensively, over years of suboptimal outcomes.

I built unCoded after five years of running other platforms because I couldn't find one that addressed all six issues structurally. That doesn't make unCoded right for everyone. It makes it right for people who've learned the same lessons and want a platform designed with them in mind from the start.

If you're not there yet, the best advice is the most boring one: start small, monitor carefully, and spend more time understanding what your tools actually do than hoping they'll make you money.

The hoping is the expensive part.

Felix is the founder of unCoded — a self-hosted, non-custodial crypto Spot trading bot with profit-sharing pricing. Documentation at uncoded.ch/docs. ArrowTrade AG, Switzerland.