Turn abandoned carts into revenue, on autopilot

The 4-week pilot is free. If it shows a real gain and you want to continue, you pay a share of the extra revenue measured against the held-back group, agreed in writing before the first paid month. No gain, no fee.

Your store does, which is why you set the cap and why every report subtracts discount cost before showing a gain. A discount the model picks can never go above your cap, and a shopper likely to come back anyway is not given one.

If the agent works, the held-back group gives up the extra recoveries on those carts. That is the price of a number you can trust, and it is why the held-back group is kept small. Without it, nobody can tell you which sales the messages actually caused.

No. Slipwork checks for an order when the cart arrives and again right before every send. If the shopper has already bought, the cart is skipped, no message goes out, and the reason is logged so it shows up in the report.

It cannot make an expensive one. Discount limits are applied in code before the model sees the cart, low-confidence decisions fall back to a plain reminder, and if the AI is unavailable the cart gets a template reminder.

Not yet. We are recruiting the first pilot stores now. When a pilot finishes, its numbers, held-back group included, go on this site with the store’s permission. Until then we would rather show you the method than a number we cannot back.

Not during the pilot. The pilot runs on email, where consent and deliverability are easiest to get right. SMS and WhatsApp come next, using the same per-cart decisions, the same discount limits and the same held-back group to measure them.

A holdout group is a random share of abandoned carts that gets no recovery message. Comparing its recovery rate with the messaged group shows how many orders the messages actually caused. Slipwork holds out one cart in five, chosen by a hash of the cart ID.

Shopify sends the same email to every abandoned checkout. Slipwork decides per cart whether a reminder, free shipping or a capped discount fits, writes in the shopper’s language, and measures the result against a held-back group so you see the real difference.

Your rules are applied in code first, so the AI only sees offers you allow. It then answers four questions about the cart, each with a probability. If its confidence is below the gate, the shopper gets a plain reminder instead.

Messages are written in the language the shopper checked out in. Before a pilot starts, we agree the languages your store sells in and check each one renders correctly in Gmail, Outlook and Apple Mail on desktop and phone.

You get a summary for each group every week. A fair answer usually needs the full four weeks and a few hundred abandoned carts in each group, because smaller samples move around by chance. The difference has to hold for the last two weeks to pass.