How HOME decides what matters now
Every recommendation is based on clear rules, explainable signals, and measurable outcomes.
This page shows the complete logic behind the priority queue, risk detection, and customer suppression. No black box — every rule is auditable and can be challenged or tuned. Stage benchmarks are computed from the book itself and self-calibrate automatically.
Early warnings
Fire before anything goes wrong. These are time-sensitive alerts, not stalls — the client may still be fully engaged.
Pre-approval expiring soon
When:pre-approval issued 60 days ago is within 14 days of lapsing, and the case is still before final offer
Then:flag to lock the property or re-issue before it expires
Big payment due, client silent
When:offplan case with a payment milestone due within 30 days and no inbound reply in the last 10 days
Then:flag so the RM reaches out before the milestone lands
Handover approaching
When:offplan property with handover within 180 days
Then:schedule pre-arrangement outreach to line up the mortgage before completion
Valuation taking too long
When:valuation requested and still not completed after more than 5 working days
Then:flag the SLA breach and chase the valuer
Stall detection
Something already slowed. Only these flags can classify a case as stalled or paused.
Stuck longer than normal
When:dwell time in the current stage exceeds the typical (p75) for that stage — never on the funded 'disbursed' stage
Then:flag as stuck and surface how many days vs typical
Quiet during transfer wait
When:two-bank transfer, signed 45+ days ago, with no inbound reply in 14 days
Then:flag as process-blocked — chase the bank, do not nurture the client
Not responding to outreach
When:3 or more outbound messages with zero inbound over the last 10 days
Then:flag as gone quiet
Documents holding this up
When:2+ documents outstanding at application or valuation, with no document activity for 5+ days
Then:flag as docs stuck
Stall vs rational-pause classifier
Runs only on flagged cases, and only stall-detection flags count. This is the 'recover vs release' decision.
Stalled (recoverable)
When:a stall flag fired and the client replied within the last 14 days
Then:STALLED — still engaged, worth a call or nudge
Rational pause (release)
When:a stall flag fired, no reply for 14+ days, and stalled more than 21 days
Then:RATIONAL_PAUSE — customer-paused → rate-watch nurture; process-blocked (transfer) → chase the bank
Cross-sell triggers
Attach the right product at the right moment — and hold everything when the client is re-evaluating.
Conveyancing attach
When:at or after final offer, conveyancing not attached, 5+ days in stage
Then:offer conveyancing — it's already inside the closing costs
Life insurance gap
When:signed or later without life insurance attached
Then:flag the compliance gap — life insurance is mandatory with a UAE mortgage
Suppression (no friction)
When:a case is classified RATIONAL_PAUSE
Then:hold every cross-sell and active-outreach recommendation — shown greyed as 'held: client is re-evaluating' — and assign only the nurture track
Bank selection intelligence
Optimise funded revenue without ever pushing the customer to a worse rate.
Revenue-optimal switch
When:another candidate bank has higher expected funded revenue (loan × commission × approval probability) AND its customer rate is within 0.10% of the selected bank
Then:recommend the switch, stating the guardrail explicitly
DBR conflict
When:the case DBR exceeds the selected bank's limit but fits another candidate's limit
Then:surface the conflict and suggest switching before submission
Payout-event note
When:the selected bank pays commission at disbursal rather than approval
Then:note the longer revenue-at-risk window through the funding period
Assumption labelling
When:showing a bank's approval probability
Then:label it a static per-segment assumption, never a prediction
Priority score
How the action queue is ranked.
Priority = commission × stage × freshness
When:ranking each case
Then:expected commission × stage-close probability × staleness decay (1.0 if active within 7 days, down to 0.2 when cold over a month)
Assumptions & calibration
Fixed thresholds: Starting estimates tuned against HOME outcomes; stage benchmarks are computed from the book and self-calibrate.
Bank approval probabilities: Static per-segment assumptions, not a trained model. Inputs for revenue optimization, not predictions.
Rules-first approach: Transparent thresholds ship fast, build trust, and generate the labels future propensity models need.