You are measuring churn wrong
Your firm tracks churn rate. The number appears on a dashboard, gets reported quarterly, and triggers a conversation about marketing spend. Churn rate is a lagging indicator. By the time it moves, the trader is already gone, the account closed, the CAC burned.
Measuring churn rate is like checking the temperature of a building after it burns down. Accurate, but useless.
Five metrics, four from trading data and one from support tickets, can show a trader pulling away before they quit. A firm that tracks them gives itself the chance to intervene while the trader is still active, still funded, still reachable.
Metric 1: Session frequency decline
A trader who logged in every market open for six weeks and now trades twice a week might be busy, or might be disengaging. The drop is worth a look.
We list session frequency first because a drop in it can show before any money is lost. Consistent traders build routines around market opens, specific sessions, and personal schedules. When that routine breaks, something has changed: a losing streak, frustration with the platform, a shift in confidence. The cause matters less than the pattern.
The signal is relative, not absolute. A trader who always traded twice a week is fine at twice a week. A trader who dropped from daily to twice a week is the one to flag. The metric requires a baseline per trader, calculated from their first 30 days of activity.
How to measure it
Track login frequency and active trading sessions per week. Compare the current 7-day window against the trader's personal 30-day rolling average. Use a decline of 40% or more, held for two consecutive weeks, as a starting flag, and calibrate it against your own churn records before you trust it.
Do not confuse this with seasonal patterns or market closures. Normalise for holidays and low-volatility periods. The signal is trader-specific deviation from their own baseline, not deviation from the firm's average.
Metric 2: Position size volatility
Consistent traders trade consistent sizes. They have a plan, a risk model, and a position sizing framework. When position sizes start swinging between 0.5x and 4x normal, the plan may have left the building.
Position size volatility measures the standard deviation of a trader's recent position sizes against their historical norm. A trader who risks 1% per trade and enters a position at 4% may have changed strategy, but after a loss it can be an emotional trade, and that is the case the metric is meant to catch.
Two patterns are worth separating. The first is a single large spike: one trade at 3x to 5x normal size, after a loss, which can be a revenge trade, an attempt to recover the money in a single position. The second is oscillation: sizes jumping between 0.5x and 3x normal with no consistent pattern, which can mean the trader has lost confidence in their sizing model.
How to measure it
Calculate the coefficient of variation (standard deviation divided by mean) of position sizes over a rolling 20-trade window. Compare against the same metric from the trader's first 50 trades. Treat a CV increase of 50% or more as possible emotional sizing. A sustained increase over 10 or more trades can mean the trader has abandoned their risk model.
Bigger sizes drain an account faster. A trader who tilts at normal size may lose over weeks. A trader who tilts at 4x size can blow a funded account in a single session. The metric is meant to flag both the risk of churn and how hard the exit could be.
Metric 3: Time between loss and next trade
This is the revenge trading indicator. It measures whether a trader pauses after a loss or re-enters the market within seconds.
A disciplined trader takes a loss, reviews the setup, checks their daily P&L against their plan, and waits for the next valid signal. This process takes minutes. Sometimes it takes the rest of the session. The gap between loss and next trade can serve as a rough proxy for impulse control under stress.
A re-entry within 60 seconds of a loss leaves little time for analysis. A firm can check in its own trade data whether those trades carry a lower win rate, a larger size and a shorter holding time than the trader's baseline.
How to measure it
Record the timestamp of every closed trade and the timestamp of the next opened trade. Filter for cases where the closed trade was a loss. Calculate the gap in seconds. Flag any instance where the gap is under 60 seconds and the subsequent position size exceeds the trader's 20-trade rolling average by 20% or more.
A single instance is a data point. Three instances in a 5-day window is a pattern. Treat five or more in a 10-day window as a recurring tilt pattern, and reach the trader.
Metric 4: Drawdown velocity
Hitting the drawdown limit is one measure. How fast a trader approaches it is another.
A trader who loses 5% over 30 trading days is having a rough month. A trader who loses 5% in 90 minutes is in crisis. The drawdown limit is the same. The trajectory is not. Velocity tells you whether the trader is grinding down or falling off a cliff.
Drawdown velocity matters because the psychological response can be different. A slow drawdown gives the trader time to adjust, reduce size, and reassess. A fast drawdown leaves no such time. In Arnsten's 2009 review, even mild acute stress that a person cannot control impairs working memory and top-down control; that a fast drawdown does this to a trader is our inference. A trader who has lost 5% in 90 minutes can lose more in the next 30.
How to measure it
Calculate the rate of P&L decline per hour during active trading sessions. Establish a per-trader baseline from their first 30 days. Flag when the current session's drawdown velocity exceeds 3x the trader's historical average.
The intervention window on fast drawdowns is narrow. At 3x normal velocity, a trader approaching the daily loss limit can hit it within the session. If the firm's retention strategy activates after the limit is breached, the strategy activates too late.
By then the worst moment may already have happened.
Track velocity at the session level, not the daily level. A trader who loses 3% in the morning, breaks for lunch, and loses 2% in the afternoon has a different experience than one who loses 5% in a continuous 90-minute spiral.
Metric 5: Support ticket sentiment shift
The first four metrics live in trading data. This one lives in customer support. Less precise, but it captures a dimension the others miss: the trader's relationship with the firm.
Traders who are learning submit tickets that sound like this: "How do I set a trailing stop?" "Where can I find my daily P&L report?" "What are the rules for scaling into positions?" The language is curious, forward-looking, and solution-oriented.
Traders who are leaving submit tickets that sound like this: "Why did my trade get stopped out early?" "Why was my payout delayed?" "Your platform cost me money." That is the shift to look for: from "how do I" to "why did you", from learning to blaming, from building a relationship with the platform to building a case for leaving.
How to measure it
This requires basic sentiment analysis on support ticket text, which some helpdesk platforms include as standard. Tag tickets as positive, neutral, or negative. Track the ratio per trader over time. Treat a shift from neutral or positive to negative, alongside any of the four trading metrics above, as a churn warning.
The sentiment shift can come before the trading changes. A firm can check its own records: do traders who turn negative in support trade with less discipline afterwards?
Building a composite risk score
Each of these five metrics is meant to catch a different facet of disengagement, and each is a proxy: session frequency for commitment, position size volatility for emotional regulation, the loss-to-next-trade gap for impulse control, drawdown velocity for crisis trajectory, and support sentiment for the relationship.
No single metric is a reliable predictor in isolation. A trader might reduce session frequency because of a holiday. Position sizes might spike because of a legitimate strategy change. The composite is meant to carry the signal.
A simple weighted scoring model is a reasonable start. Assign each metric a score from 0 (no signal) to 10 (strong signal). Weight them according to their predictive power in your specific trader population. Start with equal weights and adjust based on historical churn data.
A sample framework:
Session frequency decline (40%+): 0-10 × 0.20 = 0-2.0
Position size CV increase (50%+): 0-10 × 0.25 = 0-2.5
Loss-to-next-trade gap (<60s): 0-10 × 0.25 = 0-2.5
Drawdown velocity (3x baseline): 0-10 × 0.20 = 0-2.0
Support sentiment shift: 0-10 × 0.10 = 0-1.0
────────────────────────────────────────────────────────────
Composite churn risk score: 0-10.0
Set thresholds that trigger escalating responses. A score of 3 to 5 might trigger an automated check-in. A score of 5 to 7 might trigger a more direct intervention. Above 7, treat the trader as high risk and act.
What to do when the score spikes
Detection alone is monitoring. Pairing detection with a response turns it into intervention.
When a composite score crosses the threshold, the firm has options. The worst option is nothing. The second worst is a generic email. The trader may be in distress or pulling away, and a templated "We noticed you haven't logged in" message tells them they are a number in a database.
We would hold any intervention to three rules: it arrives while the behaviour is happening or within hours, it names the actual pattern rather than offering generic encouragement, and it helps the trader return to their own plan rather than prescribing a new one.
This is performance coaching, not financial advice. No trade recommendations. No position sizing suggestions. A prompt to pause, review, and re-engage with the process the trader already has.
~75% of retail traders quit within 90 days. The data to watch for those warning signs sits in the systems prop firms and brokers run today. A detection layer paired with real-time coaching (coaching, not financial advice) is built to reach traders while there is still time to keep them.
The metrics are knowable. The patterns are detectable. The question is whether you build the system to catch them before the trader is gone.
