AI Agents

14 commerce agents that don't just answer questions. They run the work.

Purpose-built for marketplaces, retail media, and D2C operations — every SellMetric agent is grounded in your governed data, follows your approval rules, and gets measured against the outcome it recommended.

14
specialized agents
5
work categories
4-step
accountable loop
Cited
grounding by default
How agents work

Every agent follows the same accountable loop.

From data to detection to action to measurement — no hidden steps, no unexplained recommendations, no black-box outputs.

01 · GROUND

Grounded in your data

Every agent operates on the Unified Commerce Graph — your governed, real-time source of truth. No external scraping, no model hallucinations.

02 · DETECT

Detect what changed

Agents watch every SKU, store, keyword, placement, and competitor 24×7. They surface what's drifting, not just what's broken.

03 · RECOMMEND

Recommend with citations

Each recommendation cites the exact data, the expected outcome, and the confidence band. You see why, not just what.

04 · LEARN

Measured & improving

Every recommendation is tracked against actual outcome. Agents tune their thresholds, signals, and confidence over time.

What agents optimize

One framework across every marketplace: the 4 Ps.

Whatever the agent, the work maps to the four levers that move commerce performance — Product, Price, Place, and Promotion — monitored and optimized continuously across channels.

Built for Performance Marketing Brand & Product Management Operations & Supply Chain
  • Smart reorder suggestions based on forecasted demand, inventory health, and shelf movement
  • Marketing automation & measurement to reduce wasted ad spend and lift ROAS
  • Workflow alerts for inventory, product, and operational issues — straight to your inbox
  • MAP and content-violation tracking to protect margin and brand integrity
P

Product

Listing quality, content health, catalog completeness, and review sentiment — so every detail page converts.

P

Price

Price drift, MAP compliance, and competitive pricing — flagged and corrected within your guardrails.

P

Place

Buy box ownership, on-shelf availability, and search visibility across every retailer and region.

P

Promotion

Bid optimization, ACoS and ROAS targets, and sponsored placement performance — automated and measured.

Agent catalog

14 agents, organized around the work they actually do.

Filter by category. Every agent ships with built-in approval rules, audit trails, and per-customer thresholds.

Detect

Buy Box Watch Agent

Monitors buy box ownership across Amazon, Walmart, and Target+ in near-real-time. Detects flips within minutes and classifies the root cause.

Amazon Walmart Target+
Detect

Availability Agent

Tracks SKU-by-SKU, store-by-store, region-by-region availability across marketplaces and hyperlocal commerce. Surfaces emerging OOS before sales are lost.

Instacart DoorDash Gopuff
Detect

Price Drift Agent

Watches your price against MAP, competitor moves, and retailer-side promo logic. Flags violations and unintended discounts in minutes.

MAP Promo logic Multi-retailer
Detect

Competitor Watch Agent

Tracks competitor pricing, share, listing updates, and ad placements across your category. Surfaces shifts that change your playbook.

Pricing Share Listings
Diagnose

Root-Cause Agent

When sales drop or ROAS dips, decomposes the change across price, traffic, conversion, mix, and availability to find what actually moved.

Sales drops ROAS dips Variance
Diagnose

Search Visibility Agent

Diagnoses lost organic and sponsored rank for every keyword that matters — and isolates whether the issue is listing, ads, content, or competition.

Organic rank Sponsored rank SOV
Diagnose

Review Sentiment Agent

Clusters reviews into themes, tracks sentiment trend per SKU, and surfaces emerging quality and content issues before they hit ratings.

NPS themes Trend Quality
Optimize

Bid Optimization Agent

Daily bid recommendations across Amazon Ads, Walmart Connect, and Target Roundel — grounded in incrementality, not just last-click ROAS.

Amazon Ads Walmart Connect iROAS
Optimize

Pricing Strategy Agent

Recommends price moves grounded in elasticity, competitor activity, channel mix, and margin floors — with simulated revenue and margin impact.

Elasticity Simulation Margin floors
Optimize

Channel Mix Agent

Allocates next-period spend and inventory across marketplaces, D2C, and hyperlocal commerce to balance growth, margin, and brand reach.

Allocation Growth × margin Multi-channel
Execute

Buy Box Recovery Agent

Once buy box is lost, executes the recovery playbook — repricing, FBA push, listing fix, or escalation — within your approval rules.

Repricing FBA Approval-gated
Execute

Replenishment Agent

Generates replenishment recommendations by store, region, and lead-time. Pushes approved orders to ERP and warehouse systems.

ERP push Lead-time aware Store-level
Report

Executive Briefing Agent

Drafts weekly and monthly business reviews — what moved, why, what to do next — grounded in your data and ready for board-ready edits.

WBR · MBR Narrative Cited
Report

Anomaly Digest Agent

Daily and weekly anomaly digests scoped to your role — what changed in your channels, why it matters, and what to act on first.

Role-scoped Daily · Weekly Slack · Email
Featured agent · in depth

The Buy Box Watch Agent in action.

When buy box ownership flips, every minute is lost revenue. This agent detects, classifies, and routes the fix — and the entire path is auditable end-to-end.

Sub-5-minute detection. Watches every ASIN in your portfolio against retailer feeds, marketplace APIs, and side-channel signals.
Reason classification. Identifies whether the loss is price-driven, stock-driven, FBA-driven, or content-driven — not just that it happened.
Approval-gated execution. Routes to the right team with the recommended action and expected impact. Auto-execution available with your rules.
Outcome-measured. Every recommendation is tracked against the result, so the agent's accuracy and impact are visible — not assumed.
Buy Box Watch Agent
Active · monitoring 4,128 ASINs
14:32:08 ET·DETECT
B08RTX-Hydra-1L lost buy box. Competitor at $4.79 (−4.2% vs SellMetric price $5.00). Estimated lost GMV in flight: $4,580/day.
14:32:41 ET·DIAGNOSE
Root cause: price-driven. Variant SKU has stock cover > 90 days. FBA in-network. Listing healthy. Repricing within MAP floor of $4.75 will recover.
14:33:02 ET·RECOMMEND
Recommend reprice to $4.77 (within MAP). Expected buy box recovery: ~92% confidence. Margin impact: −0.4pp on this SKU.
14:33:14 ET·EXECUTED
Approved by auto-rule: reprice-within-MAP-≤5%. Price updated. Buy box restored at 14:46:22 ET. Recovery time: 14 min 14 s.
Architecture

How an agent thinks — every time.

No black boxes. Every agent uses the same accountable stack: governed data, retrieval, reasoning, action, evaluation.

Layer 01 — Input signals
Marketplace APIs Retail media platforms ERP · WMS Reviews · sentiment Competitor signals First-party web
Streaming and batch sources land in the Data Foundation.
Layer 02 — Grounding
Unified Commerce Graph Semantic models SKU × channel × store × date
Agents read from a single governed graph — no parallel models, no untracked queries.
Layer 03 — Reasoning
Gemini / GPT-class LLM Domain reasoners Forecasting models Causal inference
Specialized reasoning per agent — not one model trying to do everything.
Layer 04 — Action
Approval rules Audit logs System-of-record write-back Slack · email · webhook
Every action is governed, logged, and reversible.
Layer 05 — Evaluation
Outcome tracking Confidence calibration Drift monitoring Human feedback loop
Every recommendation is tracked against its outcome — so the agent gets sharper over time.
Governance

Designed for teams who can't afford black boxes.

Every action explainable, every recommendation cited, every output auditable — at enterprise scale.

Cited grounding

Every output points back to the exact rows, dimensions, and time windows used. No claim without its source.

cite_sources() data_lineage

Approval rules

Agents act only within explicit per-customer rules. Auto-execute, route-for-approval, or never-execute — your call, per workflow.

RBAC guardrails

Audit trail

Every input, prompt, decision, and outbound action is logged. Replay any agent run end-to-end on demand.

replay() SOC 2

Outcome measurement

Every recommendation tracked against actual outcome. Accuracy, precision, and impact reported per agent, per workflow.

eval_metrics drift_alerts
Put agents to work

Bring 14 agents to your commerce operation.

Start with the agents that solve your most expensive problem first. Add the rest as your team scales.