Hi everyone,
We’re the team behind Upsello, an AI sales assistant built for Shopify stores.
Most chatbots wait for shoppers to ask a question. But shoppers don’t always ask.
They compare several products, revisit shipping and return policies, pause at checkout, and sometimes leave without saying anything.
That’s the problem we’re trying to solve.
Upsello uses browsing and purchase signals to recognize moments when a shopper may need help. It can answer questions using your store’s product and policy information, recommend relevant products, and offer contextual assistance while purchase intent is still warm.
Here’s our honest situation: Upsello is live on the Shopify App Store, but we’re still early. We want to learn directly from merchants—not just from dashboards—where proactive assistance is genuinely useful and where it risks becoming another annoying popup.
The free plan includes 100 AI conversations per month, so merchants can test it with real shoppers before deciding whether it belongs in their stack.
If you try it, we’d especially value honest feedback on:
-
Whether its answers accurately reflect your products and policies
-
Whether the recommendations feel relevant
-
Whether it intervenes at appropriate moments
-
Which checkout concerns your shoppers raise most often
-
Anything that feels confusing, intrusive, or incomplete
Critical feedback is welcome. It will help shape what we improve next.
Shopify App Store: Upsello AI Chatbot & Live Chat - AI chat that guides shoppers, lifts AOV & recovers lost... | Shopify App Store
If you have questions about setup or whether Upsello fits your store, reply here or send us a message. We’re happy to help and will always be transparent that we’re part of the Upsello team.
It’s the most important app in AI era.
Deploying an AI sales assistant to reduce shopper hesitation is a great concept, but the biggest hurdle is that “silent catalog data corruption causes AI agents to confidently hallucinate wrong answers.”
Three catalog data failure modes that cause AI sales assistants to mislead shoppers:
-
Variant Image Dissociation on Option Switching: When a shopper asks the AI assistant “Show me the Sage Green in Queen size,” if the product catalog has dissociated variant images (due to blank secondary size rows in CSV imports), the AI renders the wrong color photo, increasing shopper hesitation.
-
Phantom Stock / Multi-Location Ingestion Gaps: If bulk inventory CSV updates overwrote location-specific inventory or ‘Variant Inventory Policy’ (‘deny’ vs ‘continue’) settings, the AI assistant will tell shoppers an item is “In Stock” when it’s physically backordered.
-
Missing Media Due to WebP Dropping: When product catalogs are imported referencing raw supplier .webp URLs, Shopify’s background importer often quietly drops image binaries, leaving your AI assistant to present blank media thumbnails.
Pre-normalizing catalog data and variant image mapping locally before import ensures your AI sales assistant reads 100% accurate product entities. Tools like EasyCatch (a client-side Chrome extension) convert supplier WebP images to static JPGs inside your browser sandbox and export Matrixify-compliant ZIPs with pre-mapped variant rows in 1 click. 100% Local-First so catalog data remains private.
The interesting question is not whether a shopper looks hesitant; it’s whether an intervention changes the next measurable transition.
I’d define a small signal set first: repeated PDP visits, variant changes, policy opens, backtracking, repeated checkout pauses, or Add to Cart attempts with no cart change. Then compare assistance versus a no-intervention holdout on product-view → ATC, ATC → checkout and checkout → purchase, while tracking dismissals and complaints.
“Chat started” and “time on page” are activity metrics. They can support a hypothesis, but they are not proof that the assistant helped. The best trigger is the one that improves a qualified journey without adding another interruption.
