Shopping and sales

Product recommendationsCopy Anchor Link

When a customer asks things like "what's a good gift for a friend who loves cats", "anything under $40?" or "which one suits me", the AI picks up to 3 products from your store and recommends them, with product cards under the reply.

It picks in three steps:

1. Find candidates: search your catalogue by keywords and by meaning (a product can be found about 5 minutes after it syncs), plus the products you marked Featured.
2. Filter by rules: keep only products that are on sale, in stock and not marked Never. If the customer gave a budget, keep only those within it (same currency); if they named a size or colour, keep only those that have it. This step is done by code, not by the model.
3. Rank: judge each product against what the customer needs and recommend only good fits, at most 3. A Featured product goes first only when it fits about as well as another; it never beats a clearly better fit.

Prices and stock are read from your synced store data every time, never from an old copy. The reply still passes the pre-send check: a claim the product description doesn't support (say, whether it is dishwasher safe) is stopped and a person answers instead.

Recommendations are available in the website chat and the in-app page, and not while the AI is running a procedure.

Product cardsCopy Anchor Link

Each card shows the picture, name, price ("from" when variants differ), stock ("Low stock" when no variant has more than 5), and two buttons:

• View opens the product page.
• Add to cart appears only for Shopify stores when the product has exactly one variant in stock, and puts it straight into the cart.

Sold-out products get no card.

Card links pass through purser to count the click, then go to your store with utm_source=purser, utm_medium=chat, utm_campaign=ai_recommendation and the conversation id purser_c; the add-to-cart link also sets the cart attribute purser_conversation. Links work for 30 days.

Featured and NeverCopy Anchor Link

In Store data → Products, the Recommend column lets owners and admins mark a product:

• Featured: recommended first when fits are about equal.
• Never: the AI never recommends it on its own; when customers ask about it, the AI still answers truthfully.

Agents send product cardsCopy Anchor Link

In the inbox, click Products next to the reply box, search and pick 1–3 products; they go out as cards with the reply. Email, WhatsApp and other channels can't show cards, so each product becomes a line: "Name (price): link".

The AI knows which page the customer is onCopy Anchor Link

When a customer opens the chat, it carries the current page's title and address, so the AI knows what "this" or "it" means. Only the page the chat was opened on is sent, never the browsing before it.

Proactive outreachCopy Anchor Link

When a shopper hesitates, one line appears next to the chat button, for example "Looking at the tea set? Ask me anything about it — size, care, shipping." If they open it, that line is the first message of the conversation and the AI carries on. Set it up in AI → Outreach.

Needs the Growth plan or higher. Up to 10 rules per workspace, all off by default; only rules you turn on run.

Limits:

• At most once per visit (pages within 30 minutes of each other, new tabs included).
• At most twice a day.
• Never on checkout pages.
• Not when the customer already opened the chat on this page.

Triggers are checked in the shopper's browser; browsing is never uploaded. purser records only how often each rule was shown, opened and closed, and whether the customer replied.

A conversation the customer opened but didn't reply in never reaches the inbox; once they reply, the line and their reply appear there together.

Each rule card shows the last 30 days: shown, opened, replied, ordered.

Customers see your line exactly as you wrote it, so write it in their language, up to 200 characters.

How product pages are recognisedCopy Anchor Link

Shopify stores are recognised automatically. Other sites are recognised when the page has <meta property="og:type" content="product">; otherwise tell the chat window from the page:

<script>
  window.Purser = window.Purser || {};
  window.Purser.page = { type: "product", productId: "sku-123" }; // on checkout: { type: "checkout" }
  window.Purser.cart = { count: 2 }; // items in the cart, for "Leaving with items in the cart"
</script>

Shopify cart counts are read automatically.

Analytics → SalesCopy Anchor Link

The Sales view in Analytics covers the last 7, 30 or 90 days.

The page also lists the most recommended products, the best sellers after the AI helped, and the orders from each outreach rule.

An order is tied to a conversation by the first of these that matches:

1. The cart attribute purser_conversation, or purser_c on the page the customer landed on. Product cards set both.
2. The signed-in (verified) customer's id.
3. Their email.

Other rules:

• Each currency is counted on its own, never converted.
• Cancelled orders don't count.
• Each order counts once, for one conversation.
• Orders that were already more than 8 days old when they arrived (a backfill, say) are not counted.

Recommendations and sales attribution are on Starter and higher.

Where orders come fromCopy Anchor Link

• Shopify: automatic. New orders Shopify pushes in real time carry the cart attribute and the landing page; orders from a backfill carry the cart attribute.
• Your own store: when you write orders through the API, put the conversation id from the cart or landing page in the order's purserConversation field. Without it, orders still match by customer id and email.