Reduce preventable attrition
Early attrition converts acquisition expenditure into unrecovered cost and forfeits all future revenue from the account. The system surfaces the behavioural patterns that precede account closure before they complete.
Behavioural intelligence for brokers and trading platforms
The layer that catches the pattern before the cost.
Our models learn how each of your participants makes decisions under pressure, detects the patterns that precede their costly mistakes, and surfaces one observation the moment a pattern is about to repeat.
The architecture behind every observation.
2.8M+
evaluations performed and compounding, 150,000+ from a single import
880
probability models per individual
60 sec
early warning before a pattern completes
Day 1
auditable from the first live session
02The pattern problem
Not analysis failure,
Behavioural failure. Repeating.
For each individual, the dominant cost driver is not the market. It is their own repeating behaviour - loss aversion sizing up after a loss, win rates decaying as a session extends, re-entries chasing the last trade.
Measured against their own baseline. Predicted before the pattern completes. Not a generic signal, their history, surfaced the moment it begins again.
Illustrative
SAMPLE SESSION
Session arc
−287ticks closed
From +94 ticks peak to -287 close.
The cascade began at 15:01 a pattern this individual has repeated across prior sessions.
The cascade began at 15:01 a pattern this individual has repeated across prior sessions.
Peak 15:01
Close 16:08
Size after loss
UP
+73% above their baseline
Their win rate,
by trade count
DECAY
by trade count
58% → 31%
!
PATTERN FORMING
14:36
Fourth same-direction re-entry forming. 78% probability it completes as a loss Calibrated to their own history.
03The Forecaster
Personalised probability models,
Calibrated against each individual's own history.
Every number the system surfaces is computed, not configured. Over 35,000 data points, probability models, and behavioural signals are evaluated before a single line is spoken.
L1 · L2 · L3
Thousands of signals. Fourteen words.
Statistical
PORTRAIT
59.8% win rate
Establishes the forecast band
from prior sessions
Best hour10–11 AM
VIX > 203.3× cost
Worst patternSame-direction re-entry
Behavioural
PATTERN
Trade 5decision decay
Detects the behavioural shift
as it develops in real time
CadenceAccelerates
Sizing↑ after losses
Warnings2× ignored · both red
Predictive
FORECAST
60sec early
Forecasts the next decision
before it occurs
Window45 sec post-loss
Forecast91% failure
ConfidenceHigh
Statistics describe. Behaviour explains. Prediction prevents.
Illustrative
Numbers shown are representative of system output. Not from a specific individual.
04Measured outcomes
Predict. Measure. Prove.
Each stage personalised per individual.
01Predict
Historical decisions analysed,
Patterns identified,
Cost and prediction logged.
The model builds a behavioural portrait from imported history. Patterns and their costs are known before the first live session.
02Measure
Live, every session,
Patterns detected,
Responses tracked.
The model runs alongside each individual in real time. Every detection, intervention, and response recorded as it happens.
03Prove
Predicted against actual,
Cost avoided measured,
Behavioural change tracked.
Unflagged decisions are consistently profitable; flagged behaviours explain the difference. Accuracy is auditable from day one.
Optimising decisions from their own behavioural data.
For the provider
Protect client capital.
Extend client tenure. Grow lifetime value.
Protecting client capital is not in tension with commercial performance, sustained accounts are the basis of recurring revenue.
Request a walkthrough05The intervention
Their data, One observation,
At the moment it counts.
The system doesn't advise. It observes how an individual behaves under pressure, and when a costly pattern begins to form again, surfaces one line of their own history.
Sample session
One session. Three moments.
Forecast
ELEVATED
Market volatility above your baseline
Pre-session briefing · 19 May
By tSheldon · 09:32
Elevated Conditions
· 412 prior sessions
- ·Market volatility above your baseline for the third time this month
- ·Your win rate drops from 58% → 38% in these conditions
- ·Position size increase after losses: 3 of last 4 sessions.
!
Behavioural alert
14:36
Session peaked at +$340. Three entries since the turn.
Last 6 sessions from here: average close -$120.
Post-decision observation · NQM5 long
By tSheldon · 14:48 · post-decision
Clean entry
+42 ticks · held 11m · pattern match #7
- ·Entry at structure, sized normally, no reactive follow-up
- ·Six of your last seven winners followed this same pattern
- ·Reinforcement: +2 confidence — clean entry, no chasing
Now
conversation · 14:36
Session pulse
14:36 · HIGH
"Past two o'clock. This is the stretch where your day usually turns.
More red sits in your final two hours than the rest of the session combined."
0:08
What's my P&L done after 2pm across the last month?
Eleven of your last fifteen afternoons opened green and closed red. The turn lands between 2 and 3 most days. Right now we're forty minutes into that window.
The profile sharpens every session, so observations grow more specific and interventions more accurate.
06The intelligence stack
Seven specialised layers,
Each building on what the others observe.
Together, they produce one observation at the moment it matters.
L1Sensing
Market Context
Reads the conditions surrounding every decision.
L2Sensing
Pattern Matching
Finds relevant history in the individual's own record.
L3Understanding
Live Monitoring
Detects behavioural shifts as they form, not after.
L4Understanding
Behavioural Profile
Deepens every session. Sharpens every decision.
L5Understanding
Forecaster
Personalised probability models. Calibrated against this individual's history.
L6Expression
Observation Delivery
One line. Their own data. Precisely timed.
L7Expression
Ambient Presence
Always there. Calm. Watching.
L1
Environmental Awareness
Reads the conditions, not the decisions.
L2
Pattern Recognition
Finds the closest moments in the individual's own history.
L3
Behavioural Detection
Sees patterns as they form, before they complete.
L4
The Living Portrait
Knows the individual, deeper with every session.
L5
The Predictive Layer
Estimates the next move — calibrated, logged, and scored.
L6
The Guardian Voice
Speaks rarely, and only with their own data.
L7
The Presence Layer
Always there. Calm. Watching.
The architecture behind every observation - logged, traceable, auditable.
07The desk manager view
The desk manager monitors every participant independently,
Risk surfaced across the desk as it forms.
Each participant carries their own portrait. The manager sees the whole desk at once,
without losing the individual underneath.
Scroll to walk through three active cases.
Hidden Labs
Desk
14 participants · 19 May · 10:42 ET
Independent at the individual. Aggregated at the desk.
08Individual journey
A living portrait of one individual,
Built over every session.
Patterns identified. Sub-patterns surface beneath them, Every session sharpens both the understanding and the evidence.
MC
M. Chen CASE-0041
Portrait: 1,204 decisions · 8-month behavioural arc
Sample arc
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Mar 14 · detect
Apr 3 · ignored
May 22 · forked from Giveback
Jun 11 · alert
Jun 12 · acted on
Aug 2 · detect
Today
Giveback12 events
Mar 14Detected
Apr 3Alert ignored
Forked into Revenge
Revenge7 events
May 22Forked from Giveback
Sizing drift1 event
Jun 11Alert raised
Jun 12Acted on
Settlement4 events
Aug 2Detected
Detection
Alert ignored
Acted on
Patterns identified
4
Interventions
24 sent
Acted on
7
Ignored
17
The system learns from what each individual acts on and what they ignore. Portrait sharpens.
09Supervisory surface
Behavioural risk becomes visible,
While it is still forming.
Detections become cases. Cases carry timestamped evidence. The layer that protects the individual is the same layer that demonstrates supervision.
Compliance Board
15 open
Sample board
Every detection logged · Every state change timestamped · Every owner assignment recorded
SLA · 24h response on flagged
Every detection, intervention, and acknowledgement is logged and immutable. The evidence layer is the system itself.
10Integration
It sits alongside existing systems,
Reads behaviour, Surfaces what matters.
Connected sources
Executions · context · sessions
Ingest throughput72%
Inputs
Reads your existing data
- Executions & positions
- Market context & session boundaries
- Standard data from your existing platforms
Observations surfaced
Last 30 days
1,204,860+8.4%
Outputs
Surfaces what matters
- Observations & risk states
- Supervisory evidence
- Accessible via API or embedded view
- Built for existing workflows
Observation only
Watches the desk
Design
Observation, not instruction
- Decisions stay with the desk
- Surfaces patterns and evidence
- Does not act, advise, or restrict
11 EVALUATE
Watch it work. Test it on your data. Run it with your participants.
A phased evaluation framework. Designed for independent assessment at every stage.
01
live walkthrough · screen share
01Platform walkthrough
Explore functionality
A live walkthrough covering real-time behavioural analysis, multi-participant risk detection, pattern-driven intervention, and audit trail generation. Your questions, answered.
02
Pattern analysis
Sample
02Your data, our analysis
Patterns, forecasts
A structured analysis run against a limited, anonymised sample of your activity. Pattern detection, cost attribution, and a mapped timeline of interventions.
03
Isolated tenant
Sandbox
Participant view
Desk Manager view
Compliance view
03Sandbox pilot
Controlled validation
Deployed alongside existing infrastructure - an isolated tenant where participants, desk managers, and compliance each engage with their own view of the platform.
12Next step