One profile drives discovery,
offers, and support answers.
This page replays the customer-side personalization pattern from our Revenue & CX track. A unified profile from the CDP feeds three journeys, every recommendation carries its rationale, and acceptance is tied to lift. The shopper below is synthetic; the pattern shape is what we ship.
A simulated run on the unified profile of Shopper 48121. Discovery: the query "waterproof trail runners under $150" returns three catalog matches with similarity scores 0.94, 0.91, and 0.89, each with a one-line rationale tied to the shopper's size history, trail-running affinity, or the price filter. Next-best-offer: the engine proposes 10% off a trail accessories add-on, then clears three guardrails: margin floor 35% versus a modeled 41%, discount frequency 1 of 2 used in 90 days, and personalization consent ON. The offer is approved and queued. Support: ticket 5512 asks where order A-77231 is and whether the jacket can be returned. The drafted answer cites the order record, in transit with delivery expected in two days, and return policy section 4.2, a 30-day window. Confidence 0.93 clears the 0.90 gate, so the reply auto-sends; below the gate it would route to an agent. As each journey runs, the profile traits it reads are highlighted, and consent is checked at read time.
Catalog items, scores, and thresholds above are representative. In an engagement the guardrails and the confidence gate are set with your merchandising and CX leads, then proven against an eval set you own.
Lift is the contract.
The numbers below are acceptance targets and design guarantees from the reference scope this demo is based on. Targets are agreed before kickoff and measured against your baseline during the pilot.
Support deflection counts only when the resolution rate holds: an auto-sent reply that bounces back to an agent scores as a miss, and the confidence gate is recalibrated.
Shared profile. Cited outputs. Gated sends.
Unified profile
Salesforce and CDP identities are stitched into one profile per customer. Consent flags are enforced at read time, so a journey without the right flag never sees the trait, and every read is logged.
Cited recommendations
Vector search over the catalog returns each match with a similarity score and a one-line rationale. Your merchandisers can read why an item ranked, and their corrections feed the eval set.
Gated automation
Offers clear a margin floor and a discount frequency cap before they ship. Support replies auto-send only above a CSAT-calibrated confidence threshold; everything below it routes to an agent with the draft attached.
Run this on your customer base.
Tell us your CRM and CDP, plus the lift number that would prove it: recommendation CTR, offer take rate, or ticket deflection. We respond in two business days with a scoped proposal.