All use cases

Convert

Cart-abandonment recovery outreach

A loaded cart goes quiet — the agent finds out why and finishes the sale, instead of defaulting to a discount.

Close the loopGet aheadBehavioral triggerEmailSMSWhatsAppPushChatE-commerce & Retail

What it does

A shopper builds a cart, hesitates, and leaves — across the industry, 70.2% of carts end this way. The standard answer is a templated coupon email, which trains customers to abandon on purpose and gives margin away to people who only had a question. The agent re-engages on the channel the shopper opted into and opens a conversation that closes the actual gap — shipping doubt, sizing doubt, a comparison left unresolved — and recovers the order, with every recovered dollar attributed.

How it works

  1. 1
    Trigger. a cart_abandoned signal fires after a timed threshold with the cart contents and session history attached.
  2. 2
    Decision. the agent weighs what's in the cart, how the session ended, and its persistent memory of the shopper, then checks consent and frequency caps; a judge gates every unprompted message for quality before anything is sent.
  3. 3
    Action. a short, specific message on the shopper's opted-in channel — email, SMS, WhatsApp, or push — that invites a reply rather than pushing a coupon; the conversation continues in chat, where the agent can answer, swap items, and send the checkout link.
  4. 4
    Follow-through. if the shopper completes the order before the send, the outreach is cancelled automatically; recovered orders are measured against a holdout; the client keeps owning the send channels throughout.
Example
AgentStill thinking about the trail shoes in your basket? If anything held you back — fit, delivery, the right model — just reply and I'll sort it out.
CustomerWasn't sure they're wide enough for me.
AgentFair question — they run narrow. The same model comes in a wide fit and it's in stock in your size. Want me to swap it in your basket and send you the checkout link?

Configuration

How the agent is wired for this use case.

Triggera cart_abandoned event from the e-commerce cart API, fired after a timed threshold with cart contents and session history attached.
Tools & actions
  • E-commerce cart API · read cart contents and session-exit context, swap items, generate the checkout link
  • Customer memory store · recall prior shopper history and infer the blocking gap
  • Messaging channel · send the opted-in outreach (email, SMS, WhatsApp, push) and continue the conversation in chat
  • Order system · detect order_completed, cancel stale sends, and write recovered-order attribution back
Autonomyjudge-gated outreach and in-conversation answers run unattended; discounts move only under explicit merchant pricing policy, never as the default, and conversion is attributed against a holdout.
Channelsemail · sms · whatsapp · push · chat
Escalationwhen the shopper's question falls outside what the agent can resolve in-conversation, it hands off to a human agent.

What you need

The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.

Signals

cart_abandoned events (timed threshold), order_completed for stale-send cancellation, session-exit context

Data

cart contents, browsing and purchase history, customer memory, per-channel consent state

Guardrails

explicit per-channel consent and frequency caps; judge gating on every unprompted message; automatic cancellation when the cart converts before send; discounts only under explicit merchant policy, never as the default move

Metrics it moves

  • cart-recoveryup: conversational recovery reaches 15-30% of abandoned carts, against single-digit norms for templated sends
  • recovered-revenueup and provable: every recovered order is attributed against a holdout
  • conversion-rateup: the information gap that blocked the first attempt gets closed instead of papered over with a coupon

See it on your own customer journey

Bring one drop-off, one churn cliff, or one silent segment. We will show you what a proactive agent with memory and judgment does with it.

Book a demo