It starts the conversation.
The agent reopens the conversation with whoever dropped off, cart ready and payment link attached. It runs campaigns, segments by your CRM or CDP and brings back the customers who already bought, at the right moment and on the channel where they actually reply.
What comes with it.
Abandoned cart
Reopens the conversation with the cart exactly as it was left. The customer replies and pays, rebuilding nothing. With multiple integrations and A/B testing on the approach.
- Multiple e-commerce integrations
- A/B testing
- Cart ready inside the message
Payment flows
Sales that were already won and stalled at the last metre: an expired Pix, a declined card, a charge left pending.
Campaigns and smart journeys
When the action is list-based rather than triggered, the broadcast goes out from here, with A/B testing and per-customer cadence respected. Cross-sell, upsell and repeat purchase run as journeys, with the AI suggesting what to test next.
- Broadcast campaigns
- Cross-sell
- Upsell
- Repeat purchase
- A/B testing
Segmentation by CRM or CDP
Who to approach, how long to wait and on which channel, from the signals in your CRM or CDP and the behaviour the agent already saw in the conversation.
From the first message to the result.
- 01
The signals are read
Abandoned cart, expired Pix, failed payment, order status, time since last purchase, through our CRM or an integration with yours.
- 02
The flow picks the moment
Who to approach, how long to wait and on which channel, respecting opt-in and that customer's quiet window.
- 03
The conversation reopens
An approved template on the official channel, carrying the real context of the interrupted purchase, with the Meta catalogue inside the message when it helps.
- 04
A/B decides what stays
Two approaches run side by side, the one that converts better becomes the default, and the AI suggests the next one to test.
On the channels you already have, plugged into what you already use.
In your operation, the next day.
messages sent by the agent, each one starting from a signal of that specific customer.
- The abandoned cart stops being a report and goes back to being a conversation.
- A sale that stalled at payment gets a second chance, and it is automatic.
- The approach arrives with the context of the purchase, not as a mass blast, and what stays live is whatever won the A/B.
The anatomy of the agent.
Marketing is one job of the same agent. It is not a prompt with access to your store: it is seven parts, and it is the presence of all of them that separates an agent from a chatbot.
Memory
What is already known about the customer: what comes from CRM and CDP integrations, and what they told you themselves in earlier conversations, on any channel.
Knowledge base
Your policies, your catalogue and your documents as the source of every answer. With no source, the agent does not answer: it escalates.
Tools
The actions it can take in your systems: look up an order, build a cart, open a return, issue a charge.
Skills
Your operation's procedures, written the way the agent must follow them, not as a generic support prompt.
Harness
The environment that runs the agent, with limits, permissions and a trail for every step. This is where governance actually happens.
Reasoning
Thinking before acting: understanding which case this is before deciding which tool to use.
Evals
The tests that say whether a change made the agent better or worse, run on every change, and not once at rollout.
And what you see afterwards.
AI Assistant
Ask about your operation in plain language and get the cut you need, without building a report.
AI Report
The report written by the AI: what changed in the period and what explains the change.
Revenue
Revenue attributed to each job and to the conversation that produced it.
Support Report
Support volume, resolution and escalation, broken down by reason.
FAQ diagnostics
What customers ask that your knowledge base still doesn't answer: the content backlog, found for you.
Questions that always come up.
- Is this mass messaging?
- Both exist, and most of it isn't. Flows start from a signal from that specific customer: the cart they left, the Pix that expired, the status that changed. List campaigns are possible too, with per-customer cadence and opt-out.
- Do I need an approved WhatsApp template?
- Yes, to start a conversation outside the 24-hour window, and we use the official WhatsApp Business API. Templates are built in WhatsApp Studio, including the abandoned-cart one with the native Meta catalogue inside.
- How do you test what works?
- A/B on the flows, on the campaigns and at every step of the journeys. The approach that converts better becomes the default, and the AI suggests the next hypothesis from what has already run.
- How is the result measured?
- By the conversation that produced the order: every sale is attributed to the approach that reopened it, and that shows up in Revenue in the console. The number for your operation comes from your own history during the pilot.
The same agent can also handle what comes after the sale.
Support is the next step: the agent that brought the shopper back also resolves the order, the delivery and the exchange.
See the agent bringing shoppers back in your operation.
A live demo with our team, driven by the context of your operation, not a generic deck.
- Marketing walkthrough with cases from your segment
- Design of your operation's priority use case
- Rollout plan and integration with your stack
The more the team knows about your operation, the more specific the demo gets. Takes a minute.
The team replies within 1 business day.