Customer Profiling
A single-client behavioral diagnostic: roughly 22 bias and trading-pattern signals, grouped into six construct families, each one gated by an evidence floor before it's allowed to speak.
A bias reads as a pattern, not an isolated flag — signals that overlap or could double-count the same evidence are deliberately grouped, not left to pile up independently.
Disposition effect, split into post-loss and post-win behavior trees rather than one blended score. Loss aversion, de-duplicated from disposition so the two never double-count the same trade. Breakeven anchoring — holding a loser specifically until it crosses back to cost basis.
Overconfidence and overtrading. Real attention-event detection from actual OHLCV price and volume data, gated so only genuinely excess activity counts — not every trade. Rank effect.
Recency bias. 52-week-high anchoring. Feedback trading — chasing whatever just won.
Lottery-stock preference. IPO first-day chasing. Repurchase reinforcement.
Underdiversification. Rebalancing inertia.
Personal-experience extrapolation — gated by a strict 270-day minimum ledger span, so it can never fire on a short, windowed data export. Only a genuinely long trading history earns this signal.
Every score groups into one of six construct families, so a review reads as a coherent pattern of behavior, not a list of unrelated flags.
Herding was evaluated and deliberately removed from the signal set after review — evidence the library is curated and pruned, not just accumulated.
Attention-event detection reads actual price and volume history, with an excess gate, so ordinary trading never registers as a behavioral flag.
A bias isn't a single number. It's a pattern across a family of related signals — or it doesn't get reported.
An internal audit surfaced a disposition-effect score reporting near-certain confidence from a ledger with no losses to disposition-effect over. The floor-and-dampening system exists specifically to close that gap.
The finding, in one sentence: a client with zero losing trades in their ledger scored 98.2% confidence on disposition effect — the exact bias the signal exists to catch, asserted with near-total certainty on evidence that couldn't support it. The fix wasn't a patch on that one signal. It was a floor and a dampening curve applied across all twenty-two.
Being honest about not knowing something yet is worth more than a confident-sounding number that isn't earned.
Customer Profiling runs on account and portfolio data the institution already holds, and masks customer names and identifiers by default everywhere its output surfaces.
Masking is the default posture, not an opt-in setting — every panel, export, and review surface shows the behavior pattern first. It's the same masking discipline the platform's Investor Insights feed uses, because both apps read the same underlying signal engine; they just surface it differently.
Masking isn't a redaction pass at export time. It's the default state the data starts in.
Customer Profiling and Customer Segmentation share the same signal engine and run on the same weekly-refreshed data. They answer different questions.
Segmentation scans the whole book — every investor, filtered by behavior, segment, bias, or holdings — for the advisor deciding where to look next. Profiling goes deep on the one client already in front of you: a review meeting, a complaint, a suitability case that needs a defensible answer today, not a ranked list.
Segmentation tells you where to look. Profiling tells you what you're looking at.
Short, specific answers — the same understated register as the rest of this platform.
Each signal has its own minimum evidence floor before it's allowed to report at all, and the whole profile is capped at a lower confidence tier when the overall ledger is short. Personal-experience extrapolation is the strictest case — a hard 270-day minimum ledger span, so it can never fire on a short, windowed export.
No. It surfaces a pattern and how confident the evidence behind it is — walking into the room with "here's what the ledger actually shows and how sure we are" instead of a single unexplained score. The conversation, and the decision, stay with the advisor.
Names and identifiers are masked by default everywhere Profiling's output surfaces — the advisor sees the behavior pattern, not exposed identity, unless a specific workflow requires it.
No — they're two views on the same engine. Segmentation scans the whole book; Profiling opens one client fully. Most reviews start in one and finish in the other.
It doesn't fire above its floor, and below full confidence it shows at a dampened, not-quite-certain strength. The platform found and fixed exactly this failure mode in an internal audit, and built the floor to close it.
Request a walkthrough on your own book — a real client's profile, the evidence floor in action, and the masking posture end to end.