From world-class support to world-class self-serviceFrom world-class support to world-class self-service

The goal

Help Shopify's merchants solve problems on their own and get back to running their business, without waiting on a support agent.

The approach

We started with the data. Working with Shopify's analytics team, we mapped the most common merchant problems and sorted them into two buckets: issues that could realistically be solved by the merchant on their own, and problems complicated enough that they were almost certainly going to end in a call to support. That distinction shaped everything we designed. From there, we drove alignment across support, marketing, security, and product leadership to set expectations and guidelines before anything shipped.

About three months in, LLMs hit the scene. The technology was new, full of gaps, and moving fast. We put our team to work learning what it could actually do, where it fell short, and anticipating what it would be capable of soon. We had to figure out how to build with it responsibly before anyone at Shopify had a playbook for that.

There was also a trust problem to solve. People had years of experience with useless bots on other sites, and Shopify's merchants were used to getting a real person on the line. Convincing them that an AI tool could actually help meant understanding these entrepreneurs deeply: what they were trying to do, how urgent it felt, and what would earn their confidence versus what would make them close the chat and pick up the phone.

Help on the go

Research surprised us on where help gets used. A lot of merchants reach for it on a phone, mid-sale: at a convention booth, a farmers market, a pop-up, wherever the store happens to be that day.

The mobile flow had to hold up to that. Ask anything from the home screen, get an answer with sources to check, follow up in the same thread, and land in the article with a way to keep asking. Nothing that requires a laptop, or time the merchant doesn't have.

Every AI answer shows its work. A merchant can open the sources behind a response and click through to the cited documentation, so the answer is a starting point they can verify, not a claim they have to take on faith.

From help to a human

Self-service has a ceiling. When an article and a follow-up don't get a merchant there, the handoff to a person has to be one motion, not a dead end that sends them hunting for a phone number.

From any article, a merchant can log in and open a chat without leaving the page. They see their place in the queue, talk with a support advisor who already has the context, and end with a transcript and ticket waiting in their Support Inbox. The help center and live support read as one system, because to the merchant they are.

Tuning the answers

An AI-backed help center is never finished. Answers drift, products change, and the only people who know when an article missed are the merchants reading it.

Feedback is built into every article, and it asks for more than a thumbs up. Merchants can flag an answer as irrelevant, inaccurate, unclear, or in need of small fixes, and leave suggestions in their own words. Each category routes to a different fix, so the signal tunes both the accuracy of the information and how efficiently it gets to the right merchant.

The results

We shipped a smarter search experience that felt familiar on the surface but worked completely differently underneath. It pulled from support documentation, community forums, Reddit, video content, and past agent transcripts to surface real answers, not just links. Every answer included citations merchants could click through for deeper context, helping build trust in a tool they had every reason to be skeptical of.

Later, we connected the experience directly to Shopify's admin, guiding merchants step by step to the exact settings they needed to fix their problem. Not just telling them the answer, but walking them there.

The impact went beyond self-service. Self-resolution rates went up. Calls to the help center went down. And the data flowing through the system started making live support agents better too, giving them richer insight into what problems merchants were hitting and how they were solving them on their own.

The AI standards we set for this project became the foundation for how Shopify approached AI across support. We defined what the technology could be trusted with, where a human needed to stay in the loop, and how to communicate AI-generated answers in a way that earned merchant confidence. Those standards outlived the project.

Taking advantage of new tech

• LLMs arrived three months in, and there was no playbook anywhere. We led the design team through the discovery, and spent as much time mediating the disasters nobody anticipated as shipping what worked.
• A merchant reads the help center mid-crisis. A confident wrong answer there costs real money, so every answer had to show its sources and let the merchant verify before acting.
• Support, marketing, security, and product leadership each had a different worry about AI in front of merchants. We pushed for alignment with each new discovery rather than holding release, because the best insights only come from real merchants using the thing.
• Nobody had defined what responsible looked like yet. The guidelines we set for this work became Shopify's first org-wide AI standards.