Skip to main content
Review volume is a queue management problem. A 3-star review with a shipping-delay complaint is a different reply than a 3-star with a product-quality complaint. Your VA takes 30 minutes to figure that out per review. Meanwhile the 24h-SLA window closes on Amazon and your health score drops.

Three unreplied reviews are hurting your health score.

Aspect-based sentiment. SLA-window replies. Negative-escalation alerts. Drafted for approval, never sent blind.

EXSEL's Review Management daemon monitors Amazon, Etsy, and eBay reviews in real time, scores sentiment per ABSA (aspect-based sentiment analysis), and drafts replies within seller-account SLA windows. It never posts autonomously unless the merchant has enabled 2FA-gated autopilot. Negative reviews trigger a root-cause draft that links to the inventory, shipping, or listing issue that likely caused the complaint. $49/mo; bundles with Account Health at 20% off.

  • Refusal contract: confidence threshold 0.35 enforced at the primitive level.
  • Causal-confirmed learning: only DoWhy-gated trajectories (p < 0.05 across 3 refuters + Cohen's |d| ≥ 0.2) train the adapter. No reward hacking.
  • Tenant isolation: per-tenant Qdrant namespace + per-tenant LoRA adapter.
  • Related ADRs: ADR-051 (D5), ADR-086, ADR-088.

What it does (concrete)

  • Aspect-based sentiment (ABSA) per review — product / shipping / service / pricing axes.
  • SLA-window reply drafting.
  • Negative-escalation alerts (specific merchant playbook per aspect).
  • Bulk draft → approve workflow.
  • Causal attribution on reply tone → future retention.

How it fits into the ClientConductor

Your ClientConductor learns your store's reply tone. After 500 confirmed replies, the adapter drafts in your voice — with attribution context merchants remember.

The broader architecture is at [/agents](/agents). The short version for this page: every action this capability takes flows through AgenticRAG (ADR-086 — confidence-gated retrieval, refuses if top-hit < 0.35) and ContinuousLearning (ADR-088 — only DoWhy-confirmed outcomes train the per-tenant LoRA).

---

Modes (autopilot is a toggle, not a tier)

  • Manual (default) — Conductor drafts, you approve.
  • 2FA-gated autonomous — Conductor executes; each action cleared through a 2FA pre-confirm.
  • Tamper-evident audit — autonomous + hash-chained audit log + weekly signed evidence bundle (Companion required).

Exit-intent drawer

> Before you go — see one output. > Run a sample on a public niche. 60 seconds. No email. > > [ your niche or SKU type ] > [ Show me one → ]

---

Close

3 numbered steps: 1. [ Start free ] — get your ClientConductor reflex-class on day one. 2. Try the capability in read-only / draft mode first. 3. Unlock when you've seen an output you'd have paid for.

---

CapabilityPriceIncluded quota
PriceIncluded quotaPlatform scope
**$49/mo**200 replies + 100 draftsAmazon, Shopify, TikTok Shop, Walmart, Etsy, eBay review surfaces

Annual prepay: −20%. Bundle 4+ active unlocks: −20%. Full catalog →