We ran a 12-month test of AI trading vs human traders: an AI-driven trading strategy against the results of an experienced human trading desk on the same solar portfolio in European spot power markets.
The AI strategy outperformed the human benchmark in 9 out of 12 months.
However, the three months where human traders performed better turned out to be the most important part of the analysis.
Because the real conclusion is not that AI replaces human traders. And it is not that human expertise can outperform automation in every scenario.
The reality is more nuanced — and commercially far more relevant.
Key takeaway: AI outperformed humans in 9 of 12 months with 8.4% net improvement, but human oversight remained critical during regime-change events.
The Comparison Framework
The analysis was conducted on a solar portfolio with significant merchant exposure across three European markets.
The human benchmark was based on actual realised revenue generated during the 18 months before the deployment of the AI trading strategy. During that period, the portfolio was managed by an experienced trading desk responsible for intraday optimisation, market positioning, and risk management.
The AI strategy initially operated in shadow mode for validation before moving into live deployment during the 12-month comparison period.
A note on method: the human benchmark is historical, not concurrent. The trading desk was measured over the 18 months before deployment, and the AI over the following 12 months, so differences in prices, volatility and weather between the two periods affect the comparison. The results show what the switch delivered for this portfolio; they are not a controlled side-by-side trial.
The objective was straightforward: Could AI consistently improve trading performance under real market conditions?
In most cases, the answer was yes. But the areas where it struggled are equally important.
Where the AI Strategy Outperformed
High-Frequency Intraday Optimisation
Human traders operate within natural bandwidth limitations. Managing multiple assets, monitoring forecasts, balancing market exposure, and reacting to intraday volatility across several markets simultaneously creates operational constraints, regardless of experience level.
The AI strategy continuously processed market signals and adjusted positions at 15-minute intervals when spreads justified action.
Over the 12-month period, this continuous optimisation generated approximately 6.3% more revenue than the human benchmark on the intraday trading component alone.
The performance advantage became most visible during periods of elevated price volatility, particularly when forecast deviations created large intraday pricing opportunities.
AI performed exceptionally well in environments requiring:
- Continuous recalculation
- Rapid execution
- High-frequency decision-making
- Simultaneous multi-market optimisation
These are areas where automated systems naturally outperform manual processes.
Overnight and Weekend Market Coverage
European spot markets operate continuously. Human trading desks do not.
Market conditions can shift materially overnight due to weather changes, grid events, or updated generation forecasts. While liquidity is often lower during these periods, pricing movements can still create meaningful trading opportunities.
The AI strategy continued monitoring and adjusting positions outside standard trading hours without interruption.
Over the full comparison period, overnight and off-hours optimisation contributed approximately 2.1% additional revenue compared to the human benchmark.
Achieving similar coverage manually would require significantly higher staffing costs and operational complexity.
Execution Consistency
One of the clearest advantages of the AI strategy was consistency.
Human decision-making is influenced by fatigue, stress, recency bias, and loss aversion, even within highly experienced trading teams.
The AI system applied the same execution logic consistently throughout the entire comparison period. As a result, performance volatility was materially lower.
During the 12-month analysis:
- The AI strategy’s worst-performing month was approximately 3.1% below expected benchmark levels
- The human benchmark’s worst-performing month underperformed by approximately 8.7%
The strongest months were relatively comparable. The major difference was downside control and execution discipline.
Cross-Market Correlation Analysis
The portfolio operated across three interconnected European markets with measurable pricing relationships.
The AI strategy systematically identified and exploited cross-market correlations using historical market data and continuous signal analysis.
Human traders understood these relationships conceptually, but AI was able to apply them faster, more consistently, and across a greater number of simultaneous variables.
This became another meaningful contributor to overall revenue improvement.
Where Human Traders Performed Better
The most important insight from the comparison came from the periods where human traders outperformed the AI strategy.
The human desk came out ahead in three calendar months:
- November
- February, where the underperformance was concentrated in the first half of the month
- The month that included an eight-day gas supply disruption, which drove that month’s shortfall
All three periods shared a common characteristic: They involved regime-change market conditions.
These were situations where normal statistical relationships broke down and historical assumptions became unreliable.
The Regime-Change Problem
The gas supply disruption period was particularly revealing.
The AI strategy had been trained on historical market conditions where gas supply remained relatively stable and pricing relationships followed established patterns.
The disruption introduced market behaviour outside those historical norms.
Experienced human traders were able to interpret geopolitical developments, reassess risk exposure, and apply qualitative judgment to changing market dynamics.
The AI system continued optimising within the boundaries of its training framework.
For that period, human decision-making produced better outcomes.
This highlights an important limitation of AI trading systems:
AI is fundamentally a pattern-recognition system. When markets enter genuinely unfamiliar territory — such as geopolitical shocks, structural regulatory changes, or extreme supply disruptions — human reasoning and contextual interpretation remain highly valuable.
Energy market transparency data from the ENTSO-E Transparency Platform shows how frequently market conditions shift across Europe, confirming why regime-change detection matters.
The 12-Month Result
Over the full comparison period, the AI strategy delivered a net performance improvement of approximately 8.4%. The main measured components were:
| Component | Effect vs human benchmark |
|---|---|
| Intraday optimisation | +6.3% |
| Overnight and weekend coverage | +2.1% |
| Net result | +8.4% |
The intraday and overnight gains together total 8.4%. Cross-market correlation trading also contributed and offset the 2.9% given back in the regime-change months, but that effect was not measured as a separate line, so it is not shown as a figure here.
For the portfolio analysed, representing approximately €4.2 million in annual merchant revenue, that equated to roughly €353,000 in additional annual revenue.
The results were commercially significant.
However, the approximately 2.9% underperformance during the three regime-change months was equally important because it highlighted the operational importance of governance and oversight.
The Right Operating Model
The conclusion from this analysis is not that AI should replace human traders. And it is not that manual trading is superior.
The strongest operating model combines both capabilities.
AI performs exceptionally well at:
- High-frequency optimisation
- Intraday execution
- Overnight monitoring
- Cross-market signal analysis
- Continuous recalculation
Human traders perform best in:
- Regime-change detection
- Strategic decision-making
- Qualitative market interpretation
- Governance and risk oversight
- Manual intervention during abnormal conditions
The most effective trading environments are built around supervised intelligence rather than fully autonomous execution.
This is exactly how ClearSpot’s agentic AI layer works for solar operations: AI agents detect anomalies and suggest actions, but humans retain approval control for critical decisions. For the risk side of this argument, read why I don’t trust fully autonomous solar trading.
Why Physical Asset Intelligence Matters
Trading performance is directly connected to generation accuracy.
If a trading system assumes a solar asset is operating at expected performance levels while the site is experiencing degradation, inverter issues, or equipment faults, imbalance risk increases immediately.
This is why physical asset intelligence plays a critical role in commercial optimisation.
Accurate operational monitoring, inspection workflows, and real-time asset visibility improve forecasting accuracy and strengthen trading performance across the portfolio. Utility-scale solar performance monitoring keeps generation forecasts honest, and a portfolio-wide asset management platform ties those signals to commercial decisions.
Commercial optimisation and operational intelligence cannot be separated.
For solar operators seeking AI that combines automation with human control, ClearSpot.ai provides agentic AI for solar O&M with human-in-the-loop approvals built in.
The Bottom Line
AI trading systems are already outperforming fully manual trading operations in several areas of energy market optimisation. That trend will continue.
However, the future of energy trading is unlikely to be fully autonomous.
The strongest model is supervised intelligence — combining AI-driven execution with human governance, contextual judgment, and strategic oversight.
The organisations that will outperform over the next decade will not simply automate faster. They will build operational structures where human expertise and machine intelligence work together effectively.
ACER’s Market Monitoring Report confirms increasing regulatory focus on trading governance and accountability in EU energy markets.
Frequently Asked Questions
Yes. Over the 12-month comparison, the AI trading strategy beat the historical human benchmark in 9 of 12 months and delivered a net revenue improvement of approximately 8.4%.
AI performed best in intraday optimisation, continuous market monitoring, cross-market analysis, overnight execution and high-frequency decision-making.
Human traders did better in three regime-change months: November, February and the month that included an eight-day gas supply disruption, when historical pricing relationships broke down.
Humans can interpret uncertainty, evaluate geopolitical developments and apply qualitative reasoning in situations outside the AI model’s training data.
Handling genuinely novel market conditions. AI is highly effective within known statistical patterns but can struggle during structural breaks or unprecedented events.
It compares 12 months of AI trading with the human desk’s results over the previous 18 months, not a side-by-side trial, so market differences between the two periods affect the numbers.
Trading accuracy depends on accurate generation forecasts. Poor asset visibility or undetected equipment faults create imbalance risk and reduce trading performance.