Customer support · Production

One Erewhon employee builds a 39-step AI workflow that automates customer service across 10 stores

The problem

Erewhon's 10 Los Angeles stores each managed their own HelpScout inbox entirely manually, with no way to prioritize high-value members or maintain consistent response quality, leaving the team spending over 20 hours per week on repetitive ticket responses.

Workflow diagram · grounded in source
1
Ticket arrives in HelpScout
trigger
“When a ticket arrives, the system:”
2
Membership status lookup
integration
“Checks the customer's membership status against Erewhon's database”
3
Route by membership tier
routing
“Routes members to a dedicated AI with a warmer tone; non-members get standard responses”
4
Draft reply with institutional knowledge
ai_action
“Drafts a reply using institutional knowledge about membership policies”
5
Surface purchase history
integration
“Surfaces the customer's purchase history so store managers can prioritize high-spend members”
6
Deliver draft to customer service
output
“Puts the crafted response directly into the hands of customer service to respond quickly and effectively”
7
English Tutor quality grading
feedback_loop
“an "English Tutor" that grades each AI draft against the final human response, scoring how much the agent changed”
Reported outcome

The workflow achieved 70 percent of AI-drafted responses sent without human modification, saved 1,500 labor hours per year, and delivered approximately $40K per year in equivalent labor cost savings.

Reported metrics
AI-drafted responses sent without human modification70 percent
Labor hours saved per year on customer service1,500
Equivalent labor cost savings per year~$40K
Hours per week on repetitive ticket responses (baseline)20+ hours per week
Reported stack
Help ScoutChatGPTBigQueryvector store
Source
https://zapier.com/customer-stories/Erewhon
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The workflow achieved 70 percent of AI-drafted responses sent without human modification, saved 1,500 labor hours per year, and delivered approximately $40K per year in equivalent labor cost savings.

What tools did this team use?

Help Scout, ChatGPT, BigQuery, vector store.

What results were reported?

AI-drafted responses sent without human modification: 70 percent; Labor hours saved per year on customer service: 1,500; Equivalent labor cost savings per year: ~$40K; Hours per week on repetitive ticket responses (baseline): 20+ hours per week (source-reported, not independently verified).

How is this customer support AI workflow structured?

Ticket arrives in HelpScout → Membership status lookup → Route by membership tier → Draft reply with institutional knowledge → Surface purchase history → Deliver draft to customer service → English Tutor quality grading.

Related customer support cases