Tool SOPs/Gorgias/Chatbot Setup

Gorgias Chatbot Setup SOP

AY
Anand Yadav · Founder, ReccordSOP
·Last reviewed June 1, 2026

A working chatbot deflects 20 to 40 percent of repetitive tickets, freeing your support team for complex cases. A bad chatbot creates customer frustration that ends up in your queue anyway, just angrier. This SOP separates the two outcomes.

Why this SOP matters

Gorgias chatbot setup looks deceptively simple, click a few intents and turn it on. The brands getting real deflection do four things differently: tight intent scope, clear handoff triggers, proactive escalation on emotion signals, and weekly performance review.

Step-by-step SOP

  1. 1

    Define the 5 highest-volume ticket intents

    Pull last 60 days of tickets, group by intent. Order status, refund request, return policy, sizing question, account access usually top the list.

  2. 2

    Train the chatbot on those 5 intents only

    Don't try to handle everything. Narrow scope wins. Each intent needs 30+ training examples.

  3. 3

    Build a clear handoff trigger

    Any negative sentiment word (frustrated, angry, manager, complaint) triggers immediate agent handoff with full context attached.

  4. 4

    Set fallback for unknown intents

    If the chatbot can't classify with 80% confidence, route to human. Never guess.

  5. 5

    Configure business-hours behavior

    Outside hours: chatbot responds with FAQ links, sets expectation for next-day response. Inside hours: handoff to human within 60 seconds of fallback trigger.

  6. 6

    Track deflection rate per intent

    Deflection rate = tickets resolved by chatbot / tickets attempted. Target 60+ percent on the 5 trained intents.

  7. 7

    Monitor CSAT on chatbot-handled tickets

    Customer satisfaction on bot-resolved tickets should be within 10 percent of agent-handled. Larger gap means deflection is too aggressive.

  8. 8

    Review weekly for the first 90 days

    Misclassified intents and bad responses surface in week 1 and 2. Fix fast or trust degrades.

Common mistakes to avoid

  • ×Training on too many intents at once (each one underperforms)
  • ×No clear sentiment-based handoff (customers get stuck arguing with a bot)
  • ×Treating chatbot as set-and-forget instead of weekly tuning
  • ×Measuring deflection rate without watching CSAT (you deflect tickets into churn)
  • ×Letting the bot answer queries it has low confidence on

Stop rewriting this SOP every quarter

Record your screen while performing the gorgias chatbot setup. ReccordSOP turns the recording into a SOP with timestamped screenshots. When your live setup changes, drift detection flags it.

Generate this SOP free

FAQ

What deflection rate is realistic?

20-40 percent of total tickets is good. 50+ percent is excellent and usually means you've trained on the right intents.

When should the chatbot hand off to a human?

Three triggers: low classification confidence (under 80%), negative sentiment keywords, customer explicit request for human.

Will the chatbot hurt CSAT?

Yes if you over-deflect. Target CSAT within 10 percent of agent-handled tickets. Wider gap means cut back.

AY
Anand YadavFounder, ReccordSOP

I built ReccordSOP after watching too many DTC ops teams lose months to undocumented workflows. These SOPs are battle-tested with Shopify operators running $1M to $50M brands.

Last reviewed June 1, 2026

More Gorgias SOPs