E-Commerce · Agentic Operations Layer

The AI operations layer
for your online store.

Put a team of agents on yoursupport inbox.

Rach.Dev runs support, order assistance, returns and refunds, fraud screening, cart recovery and a live Fraud/Cart Sentinel across the systems you already use — with a human in the loop on every refund, chargeback and high-risk order, and a full audit trail on every action.

PCI DSS-aware by designWorks with Shopify, WooCommerce & BigCommerceAI disclosure on every chatEnglish & Spanish
The operating picture

What an agent team takes off your store

Most of an online store's load isn't strategy — it's answering the same order questions, processing returns, screening risky orders and chasing carts that almost converted. Here's where agents own the busywork, mapped to how your store actually runs.

In demo

Support Front Door

Every pre- and post-purchase question answered — 24/7, in English or Spanish, on chat and SMS.

  • Multi-channel intake (web chat, SMS, WhatsApp, email)
  • Order lookup & customer history linked instantly
  • Clear AI disclosure on first contact, every time
In demo

Order & Shipping

Where is my order? answered from live tracking — and delays flagged before the customer asks.

  • Real-time status from your 3PL / ShipStation
  • Address edits & holds staged for the warehouse
  • Carrier-delay and stockout alerts surfaced early
In demo

Returns & Refunds

Self-serve returns the customer finishes in one chat — refunds wait for a human to approve.

  • Eligibility checked against your return policy
  • Prepaid label generated; RMA opened automatically
  • Refund drafted and staged for human sign-off
In demo

Fraud & Risk Screen

Every order scored for risk — suspicious ones held and escalated, never auto-refunded.

  • Velocity, mismatch & chargeback-risk signals scored
  • High-risk orders held for manual review
  • Card data never stored — processor handles PCI scope
In demo

Conversion & Cart Recovery

The lost revenue almost everyone leaves on the table — re-engaged on autopilot.

  • Abandoned-cart nudges via Klaviyo, SMS and email
  • Sizing, comparison & in-stock answers pre-purchase
  • Back-in-stock and price-drop alerts to the right shopper
In demo

Reviews & Loyalty

Post-purchase follow-up, review requests and the loyalty nudges your team never sends.

  • Timed review requests after delivery
  • Sentiment triage on incoming reviews
  • Loyalty and win-back offers to lapsed customers
In your build

Catalog & Merchandising

Listing hygiene, enrichment and the merchandising chores no one wants to do.

  • Product-description drafts & spec enrichment
  • Duplicate / out-of-sync listing detection
  • Inventory-sync checks across channels
In demo

Knowledge Layer

A role-aware assistant grounded only in your policies, catalog and order data.

  • Answers shoppers, support and ops from approved sources
  • Every answer cites its source
  • Hard guardrails — never legal or payment-dispute advice

Watch the team run an order

Pick a case and press play. Watch the agent team run it end to end — a human approves every refund and every high-risk order.

Control Tower
Order #SF-20418 · Jessica HartmanWrong size hoodie — return + refund
Web chat · Business hours
Contact
Identify
Resolve
Risk Screen
Decision
Fulfill
Follow-up
Agents
AvaSupport Front Door
Idle
MarcusOrder & Risk Triage
Idle
NoraReturns & Case Drafter
Idle
OwenFulfillment Coordination
Idle
RileyPayments & Fraud Screen
Idle
IrisKnowledge Assistant
Idle
HopeFraud / Cart Sentinel
Idle
Decision trace0/7

Press Run the journey — or scroll in and watch it play automatically. Every clinical action waits for a clinician.

Every refund, credit and high-risk order waits for a human. Rach.Dev drafts, stages and routes — a person approves.

Seven specialists, one conductor

Each agent owns one job and hands the next a complete, structured context. Atlas routes the work, enforces the human-in-the-loop gates on refunds and risk, and writes every action to an audit log.

Atlas

Atlas is the orchestrator. It routes each contact and order to the right specialist, carries shared customer and order context between them, pauses for human approval on every refund and high-risk order, and records a complete, timestamped audit trail.

Live handoff pipeline
AvaSupport
MarcusTriage
NoraReturns
OwenFulfill
RileyPayments
IrisKnowledge
HopeSentinel

Click Ava below — the full support-to-fulfillment workflow plays out automatically.

MarcusOrder & Risk Triage

Order & risk triage. Matches the contact to the right order, checks return eligibility against policy, scores the order for risk, and escalates anything suspicious straight to the risk lead.

Context from Ava
Jessica HartmanReturn request#SF-20418
Order matched — delivered 6 days ago, in the 30-day window
Item eligible per return policy; no serial-return pattern
Risk path armed — escalates if fraud or chargeback signals appear
Order matchingPolicy & eligibilityRisk escalation
NoraReturns & Case Drafter

Returns & case drafter. Logs the reason, drafts the RMA, exchange and refund options, generates the prepaid label, and writes a clean case note — leaving a human to approve, not type.

Context from Marcus
Eligible returnReason: wrong sizeRefund $58.00
Return reason logged; exchange-vs-refund options drafted
Prepaid label generated and RMA staged for the customer
Refund drafted from the order — flagged for human sign-off

Draft only. The refund is issued solely after a human approves.

RMA draftingPrepaid labelsCase notes
OwenFulfillment Coordination

Fulfillment coordination. Pushes approved refunds to the processor, opens RMAs in the 3PL, stages address edits and holds, and keeps customers updated with tracking and reminders.

Context from Nora
Refund approvedRMA → 3PLTracking link
Refund pushed to the processor after human sign-off
RMA opened in the 3PL; return label and tracking shared
Exchange-shipment and delivery reminders scheduled
3PL orchestrationRefund routingTracking & reminders
RileyPayments & Fraud Screen

Payments & fraud screen. Scores every order for chargeback and fraud risk, checks refund value against policy, and never touches a raw card number — the processor keeps card data in PCI scope.

Context from Owen
Original card intactValue within policyNo prior chargebacks
Refund risk scan clean — original payment method, low value
AVS / velocity / device signals scored on the order
High-risk orders held and escalated — never auto-cleared

Card numbers are never stored. PCI scope stays with your payment processor.

Fraud scoringChargeback riskPCI-aware
IrisKnowledge Assistant

The role-aware knowledge assistant. Answers shoppers, support and ops from your policies, catalog and order data only — every answer cited, and never legal or payment-dispute advice.

Context from Riley
Shopper viewApproved sources onlyEN / ES
Answered shipping, sizing and policy questions from approved sources
Every answer carried its source citation
Chargeback / legal-dispute request → handed to a human, not advised

Iris informs. It never gives legal advice, disputes a charge, or issues a refund.

Role-awareCited answersNever legal advice
HopeFraud / Cart Sentinel

The Fraud / Cart Sentinel. Always on, reading the live order, payment and inventory stream for every store — catching fraudulent orders, abandoned carts, stockouts and shipping delays before they cost a customer or a chargeback, and staging the response for the team.

Context from the store
Peak trafficChargeback risk: elevatedInventory: tight
How Hope calibrates for this store
Baselines normal order, refund and cart-recovery patterns
Tightens fraud thresholds during sale events and high-value SKUs
Suppresses noisy alerts to cut alert fatigue for the team
What Hope watches in real time
Order velocityAVS / mismatchChargeback riskCart abandonmentStock levelsCarrier exceptions
Fraud — velocity spike with billing / shipping mismatch
Chargeback risk — repeat dispute pattern on a payment method
Cart abandonment — high-intent cart stalled at checkout
Stockout — bestseller crossing the low-inventory threshold
Shipping delay — carrier exception on an in-transit order
Price / listing error — a SKU mispriced against margin floor

Hope is advisory. It alerts and stages — a person decides and acts. Every alert is logged with the signal that triggered it.

Live · StoreAlways-on monitorFraud & cart watchAdvisory only

Ava captures patient intake by voice, then hands a structured summary to 6 downstream agents — triage, documentation, coordination, billing, knowledge and ICU monitoring — each acting on the shared context, with Atlas orchestrating and a clinician approving every clinical action.

Ava starts the workflow.

One layer over the systems you already run

Rach.Dev sits on top of your storefront, payment processor and 3PL — orchestrating agents, enforcing governance, and keeping a human in the loop on refunds and risk. No rip-and-replace.

Layer 4Human-in-the-loop

Every refund, store credit, chargeback response and high-risk order pauses for a person to approve. Agents draft and stage; people decide.

Approval gatesRefund & risk sign-offRole-based access
Layer 3Governance & audit

PCI-aware design with card data kept in the processor's scope, configurable GDPR / CCPA retention, AI disclosure on every chat, and a complete, timestamped audit trail.

PCI-awareFull audit logGDPR / CCPA controlsSource citations
Layer 2Agent orchestration

Atlas routes each contact and order to the right specialist, carries shared customer context between agents, and coordinates hand-offs end to end.

Routing & hand-offShared contextEscalation paths
Layer 1Integration

Bidirectional connectors to your storefront, payment processor, 3PL and marketing stack over the APIs and webhooks your systems already speak.

Storefront APIsProcessor & 3PLWebhooks & events
Works with the systems you already run
ShopifyWooCommerceBigCommerceStripePayPalShipStation3PL / WMSKlaviyoGorgiasTwilio

A knowledge layer that knows its limits

Iris answers shoppers, support and ops from your policies, catalog and order data only. Every answer is cited — and anything legal or a payment dispute is always handed to a human, never advised.

Try a question

IrisShopper view

Hi, I'm an AI assistant for the store. Ask me about orders, shipping, returns or sizing — I'll always cite my source.

Iris provides information only — never legal advice or guidance on disputing a charge. Refunds and payment disputes are routed to a human on the team.

A knowledge layer that knows its limits — questions and answers

Where is my order #SF-21755?
Your order shipped 3 days ago and is in transit, with a current ETA of Thursday, June 11. The carrier logged a brief delay in transit, which is why it's running a day behind the original estimate. You can follow it live with the tracking link I just sent, and I'll text you on the delivery day.
¿Puedo devolver este artículo y cuándo recibo mi reembolso?
Sí. Su artículo está dentro de la ventana de devolución de 30 días, así que puedo generarle una etiqueta prepagada ahora mismo. Una vez que el almacén reciba la devolución, el reembolso se procesa a su método de pago original en 3 a 5 días hábiles, después de la aprobación de un agente. ¿Prefiere un reembolso o un cambio de talla?
The charge looks wrong — should I dispute it with my bank?
I can't advise you on disputing a charge — that's a payment matter for a person, not an AI. I've pulled up your order and billing details and routed your question to our support team, who will review the charge with you directly and can issue a correction or refund if one is owed. Please hold off on a chargeback until they reach out so we can resolve it faster for you.

Governance, not just guardrails

The controls that let an online store actually deploy agents — enforced in the system, not promised on a slide.

Human in the loop on moneyEvery refund, store credit, chargeback response and high-risk order waits for a person to approve. No agent moves money or clears a risky order on its own.
Complete audit trailEvery action, hand-off and data access is logged with a timestamp, the agent, the source and the approver — exportable on demand.
PCI-aware, privacy by designCard numbers are never stored — PCI scope stays with your payment processor — and customer data follows configurable GDPR / CCPA retention and deletion policies.
Grounded with sourcesAnswers come only from your approved policies, catalog and order data and cite them. No open-web guessing, no ungrounded claims.
No vendor lock-inStandards-based connectors to your storefront, processor and 3PL, and your data stays yours. Turn an agent off and your store keeps running.

Live audit log

A sample of what every action looks like on the record.

23:18:04Inbound SMS (es) received & answered with AI disclosure — order #SF-21755Logged
10:02:41Return eligibility checked — within 30-day windowVerified
10:03:12Customer PII accessed: order + billing lookup (minimum necessary)PII
10:04:55Refund drafted — $58.00 to original card, awaiting sign-offPending
10:05:30Refund approved by Dana Whitfield, Support LeadApproved
14:22:09Fraud signals tripped (AVS + velocity) — order #SF-22090 held, risk lead pagedEscalated

What changes when agents own the busywork

Directional outcomes our pilots target — validated on your own data, never assumed.

24/7
Coverage, every channel
Chat, SMS, WhatsApp and email answered around the clock, in English and Spanish — no overnight backlog.
Minutes
From WISMO to answer
Order status, returns and sizing resolved in one chat instead of a multi-day email thread.
Fewer
Carts left behind
High-intent carts re-engaged and pre-purchase questions answered before the shopper bounces.
Caught
Risk before it ships
Fraud and chargeback signals scored and held before a high-risk order ever leaves the warehouse.

The problem, in numbers

  • Roughly 70% of online shopping carts are abandoned on average, leaving a large share of high-intent revenue on the table for recovery. Baymard Institute, Cart Abandonment Rate (averaged across 50 studies)
  • eCommerce merchant losses to online payment fraud exceeded about $48 billion globally in 2023, up from roughly $41 billion the year before. Juniper Research, 2022 (eCommerce online payment fraud forecast)
  • About 17.6% of merchandise purchased online was returned in 2023 — a markedly higher return rate than in-store. NRF & Appriss Retail, 2023 Consumer Returns in the Retail Industry

Figures above are external benchmarks and pilot targets, not guarantees — we validate every number on your own data before you rely on it. Monitoring agents are advisory: they alert and stage, a person acts.

Start with one workflow. Prove it. Then scale.

We stand up a single workflow on your existing storefront and 3PL, show the audit trail and the outcomes, and expand only once your team trusts it.

01
Scope one workflow

Pick the highest-pain workflow — support, returns, fraud screening or cart recovery — and we map it to your stack.

02
Pilot in your environment

Agents run on your storefront, processor and 3PL with a human in the loop and a full audit trail, in weeks not quarters.

03
Measure, then expand

Review the outcomes on your own data, then roll the agent team out workflow by workflow and channel by channel.