How Mercari resolves 120,000 support tickets a month with Zapier

Mercari handles roughly 120,000 customer support tickets every month. From returns and disputes to account access, payments, and Customer funds. The customer support work at Mercari is high-volume and layered with complexity.

Eric McNulty, Systems Manager for Customer Experience at Mercari US, had tested dedicated AI support tools. They worked for basic answers pulled from help center articles. They struggled with Mercari's more complex cases, where the system needed to follow specific rules, pull the right context, avoid unsafe answers, and know when to fall back to a human.

So Eric built Mercari's own AI support orchestration system on Zapier.

The system now handles first replies for about 97% of Mercari support tickets. About 47,000 tickets per month are handled entirely by Zapier automations, with no human agent response required. CSAT is up a full point on Mercari's 5-point scale.

Mercari handles roughly 120,000 support tickets a month, and many of them affect parts of the customer experience where mistakes are costly: account access, payments, disputes, returns, and customer funds. Keeping up with that volume meant either hiring and training more agents for repetitive work or finding a way to answer more tickets safely without lowering the bar for accuracy.

Eric wanted AI to help reduce the time his team spent on repetitive tickets, freeing up more time for humans to focus on high-touch cases. But generic AI could only handle the easy questions. He did not want a "free-range" agent deciding what to do with too much context and too little structure.

"If you don't implement it right, it can create more problems than it solves," Eric said.

The team also needed a system that the CX organization could operate without asking engineering to update every policy change, prompt, or routing rule.

Eric built a master Zap with roughly 90 steps. The Zap starts by classifying each ticket by category and subcategory using predefined definitions. It pulls live prompt data, category definitions, policy logic, and response content from Google Sheets.

The Sheets database acts as a live content management system. CX operations teams can update prompts, category language, or routing definitions in a spreadsheet instead of asking engineering to change the workflow. Zapier pulls those changes into the next run, so the people closest to the policy can adjust the automation as Mercari's support needs change.

Eric then layered a Google Apps Script UI on top of the spreadsheet, so non-technical team members can manage the system without opening Zapier.

"The spreadsheet's the database. The Zaps are the orchestrators, and they also take the actions," Eric said.

From there, sub-Zaps handle specific flows: FAQ responses, routing, response generation, and additional analysis. AI by Zapier acts as the classifier, analyzer, decision-maker, and response generator inside a flow Eric already defined.

Mercari now sends roughly 23,000 AI replies per week through Zapier. Nearly every ticket gets a first reply within minutes. Around 40% of total monthly ticket volume is fully handled by AI and automation. This provides Mercari customers with faster resolutions to their inquiries.

That equals more than 3,000 hours of support team work freed up each month, allowing customer service agents to assist customers who need more hands-on support.

The design also gives Mercari the guardrails it needs. If confidence is low or a workflow fails, the ticket reverts to the standard human support process. This process ensures that tickets don't disappear due to inaccurate resolutions and that customers don't receive unsafe replies.

Eric also stores ticket-level data in Dixa custom attributes, including AI response count, agent response count, category, subcategory, reasoning, and other fields generated by the Zapier workflow. That data feeds Looker reporting and spike alerts, helping Mercari see trends before they become larger CX problems.

"Implementing AI in customer service is not about replacing human agents. It's about giving customers the right answers as fast as possible so they get the resolution they need right away. This then frees up our human agents to focus more on our customers with complex cases."

Eric had used Zapier for about 10 years before building the support system. AI by Zapier changed what he could build.

Instead of asking an AI agent to find its own tools, search its own knowledge, and choose its own path, Eric uses Zapier to constrain the work.

"We give it exactly the right context," Eric said. "We give it the customer's message, scrubbed for PII to protect our customers. And then we give it exactly the right help center and knowledge base articles."

His simplest description of the pattern: "This is the data. Now do this."

That structure is why the system works at Mercari's scale. Zapier gives the workflow its structure, so AI is only making decisions inside the path Eric already designed. The CX team still owns the policies, the edge cases, and the moments where a human needs to step in.

Mercari’s success offers a portable pattern for any high-volume support organization:

1. Start small -- Prove value on narrow, internal workflows before moving to customer-facing replies.
2. Define the path -- Use automation to set the route and let AI handle the judgment within those bounds.
3. Provide exact context -- Don't give AI the whole database; give it only the data it needs for the specific step.
4. Empower the policy owners -- Build interfaces that allow CX experts to update rules without needing to write code.

As Eric puts it, "People were not meant to be stuck doing monotonous work all day, so automate what you can and let humans focus on what matters".