How decisions get made. What's explainable. How to audit.
BHASM reads observable customer behaviour — purchase history, engagement signals, cycle position, world context — and produces three things per customer per cycle: a score, a recommended action, and the reasoning behind both.
It does not invent intent. It does not synthesize facts. Every score is derived from data the operator’s own systems already record.
Message text is generated with the assistance of AI, calibrated to the operator’s brand voice, and passed through automated content safeguards before any send. Scoring and timing decisions come from BHASM’s internal engine — no external model decides who gets contacted or when. Operators can review, edit, or stop any message from the terminal.
For any customer the operator can ask “why this score?” — and BHASM returns a list of contributing factors with relative significance.
45 days since last order · primary signal
1.5 cycles overdue · significant
last three orders declining · moderate
pre-festival window · moderate
steady, responsive · stable
Every decision is recorded with the world signals live at the time and the safeguard that let it through, and the record exports as a stream you can hand to an auditor. What you cannot do is re-run a past decision against today’s engine — a replay would answer with today’s model, not the one that decided.
Exact model coefficients, aggregation formulas, and proprietary calibration logic are protected intellectual property. This ensures the engine remains effective, continuously improving, and difficult to replicate. All decisions remain fully auditable for compliance and review needs.
Every outbound action passes through a safeguard stack before firing. The stack enforces:
Default silence — no action without an observable signal
Permanent opt-out respect — once a customer says stop, that decision never reverses
Frequency protection — no over-contact in any window
Sentiment safeguards — open complaints halt all outreach
Account scope — every query names the account it may read
BHASM records each outcome — returned, lost, or still inconclusive — and reports precision, recall and false-positive rate for the window you pick. Rows with no settled outcome are excluded rather than counted as wins. Below thirty settled outcomes the page reports nothing at all: it names the sample it has and waits, because an accuracy figure off twelve rows is a number, not evidence. Ask us for the methodology.
DPDP Act 2023 (India) — enforced
GDPR (EU) — enforced
CCPA (California) — enforced
Isolated Airlock encryption — every tenant under its own derived key
Data Processing Agreement available on request — write to hello@bhasm.ai
For audit log access, compliance documentation, or to ask a specific question about how the engine handled a specific decision — hello@bhasm.ai.