FS-DOC-03Operations
Process Runbook / SOP
FAQ & First Response Automation
[YourCompany.com] · Customer Support Department · Prepared by FullSpec · [Today's Date]
This runbook is the day-to-day operating guide for your support team. It explains exactly what the automation does on every inbound ticket, what your team is responsible for, how to handle exceptions, and who to call when something needs attention. FullSpec built, tested, and launched the automation end to end. Your team's role is limited to reviewing flagged tickets, keeping the Notion knowledge base current, and following the escalation contacts listed in Section 04.
01Process overview
This automation monitors your Zendesk inbox continuously. The moment a new support ticket or email arrives, three AI agents handle classification, reply drafting, ticket tagging, HubSpot logging, and Slack escalation alerting without any agent needing to act first. Tickets the system is not confident about are placed in a dedicated review queue so a human can step in. The result is that FAQ tickets receive an accurate, personalised reply within two minutes of arriving, urgent issues are flagged immediately to the right person, and every interaction is logged to HubSpot automatically. Your support team focuses only on the cases that genuinely need a human.
Process name
FAQ & First Response Automation
Trigger
A new ticket or email lands in the Zendesk inbox
Final output
Customer receives an accurate first response within 2 minutes; interaction is logged in HubSpot; urgent tickets generate a Slack alert to the support team
Agents running
Ticket Classification Agent, First Response Drafting Agent, Escalation and Logging Agent
Tools involved
Zendesk, Notion, HubSpot, Slack, Gmail
Weekly volume
~55 tickets/week (approximately 220/month)
Human checkpoint
Manual review of low-confidence tickets placed in the Zendesk flagged queue
Process owner
[Your name] — fill in once confirmed
Process Runbook / SOPPage 1 of 4
FS-DOC-03Operations
02Step-by-step: what happens and who acts
What you actually need to do: There is only one human step in this process. When the Ticket Classification Agent is not confident enough to auto-reply, it places the ticket in a dedicated Zendesk review queue and marks it with a low-confidence flag. Your support agent opens that queue, reads the ticket, and handles it manually — replying, tagging, and logging as needed. Everything else is handled automatically.
Step
What happens
Who acts
Type
1
A new ticket or email arrives in the Zendesk inbox. The automation platform detects it instantly via a Zendesk webhook, regardless of the time of day or current agent availability.
Automation
Automated
2
The Ticket Classification Agent reads the ticket body and subject line, identifies the intent category (FAQ, billing, technical, or escalation), and assigns a confidence score to that classification.
Ticket Classification Agent
Automated
3
The system checks whether the confidence score is above the agreed auto-reply threshold. If yes, the FAQ path begins. If no, the ticket is immediately flagged and moved to the low-confidence review queue in Zendesk.
Automation
Automated
4
For FAQ-category tickets above the threshold, the First Response Drafting Agent queries the Notion knowledge base and retrieves the matching help article content relevant to the customer's question.
First Response Drafting Agent
Automated
5
The First Response Drafting Agent combines the retrieved Notion content with the customer's name and specific ticket details to produce a clear, personalised first response ready to send.
First Response Drafting Agent
Automated
6
The drafted reply is sent to the customer through Zendesk. The correct category tags and priority level are applied and the ticket is routed to the appropriate agent queue. The full reply is delivered within two minutes of ticket receipt.
Automation via Zendesk
Automated
7
For low-confidence tickets that did not meet the auto-reply threshold, a support agent opens the Zendesk flagged queue, reads the ticket, writes a reply manually, applies tags, and logs the outcome. This is the only step your team initiates.
Support Agent
Human
8
Once a reply is sent (by automation or by a human agent), the Escalation and Logging Agent creates an activity note on the customer's HubSpot contact record. The note captures the ticket category, the reply type, and the timestamp so account managers have a full support history.
Escalation and Logging Agent
Automated
9
For any ticket classified as urgent or escalation, the Escalation and Logging Agent posts a structured alert in the designated Slack support channel. The alert includes the ticket summary, priority level, and a direct link to the Zendesk ticket, and tags the on-call agent.
Escalation and Logging Agent
Automated
Process Runbook / SOPPage 2 of 4
FS-DOC-03Operations
03Handling exceptions
The table below covers the most realistic exception scenarios for this process. For each situation, it describes what the automation does on its own and what, if anything, your team needs to do.
Situation
What the system does
What you do
Ticket arrives with missing or incomplete data (no email address, empty subject line, or unreadable body)
The Classification Agent assigns a low-confidence score and moves the ticket to the flagged review queue with a tag indicating missing data. No reply is sent and no HubSpot log is created until a human resolves the record.
Open the flagged queue in Zendesk, complete the missing information where possible (look up the customer in HubSpot by name or phone if available), and reply manually. If the ticket cannot be matched to a customer, reply asking for confirmation details before proceeding.
A duplicate ticket is submitted by the same customer for the same issue within a short window
Zendesk detects the duplicate based on the customer email and existing open ticket. The automation merges the new ticket into the existing thread and suppresses a second auto-reply to avoid confusing the customer.
No action is required unless the duplicate contains new information not in the original ticket. In that case, Zendesk will surface it in the merged thread for your review before the next reply goes out.
The Notion knowledge base does not contain an article matching the ticket topic (FAQ category with no matching content)
The First Response Drafting Agent fails to retrieve content and automatically downgrades the ticket to the flagged review queue rather than sending a blank or inaccurate reply. It tags the ticket with 'knowledge-gap' so the content gap is visible.
Reply to the customer manually. After doing so, note the topic in your knowledge base backlog so a Notion article can be added. Recurring knowledge-gap tags are a reliable signal that a new FAQ article is needed. Contact support@gofullspec.com if the gap tagging is not appearing correctly.
A flagged ticket sits in the low-confidence review queue and no agent acts on it within a set time window (default: 2 hours during business hours)
The system sends a reminder Slack alert to the support channel noting the ticket ID, how long it has been waiting, and the customer's name. A second reminder fires at 4 hours if the ticket remains open.
Open the flagged ticket and handle it. If the reminder alerts are not arriving as expected, check the Slack channel configuration and contact support@gofullspec.com.
A customer does not respond to the automated first reply within 48 hours (the ticket remains open and unresolved)
The automation does not send a follow-up automatically by default. The ticket stays in the relevant Zendesk queue at its assigned priority level and appears in the standard open-ticket view.
Your team's normal follow-up process applies. If your team wants automated follow-ups after 48 hours, contact support@gofullspec.com to discuss adding a follow-up step to the workflow.
Zendesk, Notion, HubSpot, or Slack is unavailable (tool outage or API error)
The automation platform detects the failed API call and retries up to three times with a short delay between each attempt. If all retries fail, the ticket is held in a queue and the workflow pauses rather than dropping the ticket. An error notification is sent to the designated support contact via email.
Check the status page of the affected tool (status.zendesk.com, notion.so/status, etc.) to confirm the outage. No manual action is needed for the queued tickets as the automation will resume and process them once the tool is back online. If the outage lasts more than one hour, work the Zendesk inbox manually during the downtime and contact support@gofullspec.com so the FullSpec team can monitor recovery.
Process Runbook / SOPPage 3 of 4
FS-DOC-03Operations
04Who to contact and when
Fill in the rows marked with brackets once your team is confirmed. The FullSpec rows are fixed. For any issue with the automation itself — incorrect classifications, failed API calls, missed Slack alerts, or unexpected behaviour — contact FullSpec support directly rather than trying to diagnose the workflow yourself.
Role
Name
How to reach them
Process owner (internal)
[Your name]
[Your email or phone]
Support Team Lead
[Rep name]
[Rep email]
Operations Manager
[Rep name]
[Rep email]
Zendesk admin (if separate from above)
[Rep name]
[Rep email]
Notion knowledge base editor
[Rep name]
[Rep email]
HubSpot admin (contact matching and deduplication)
[Rep name]
[Rep email]
FullSpec builder
FullSpec team
support@gofullspec.com
FullSpec support
FullSpec support
support@gofullspec.com — for automation errors, unexpected behaviour, threshold changes, and tool connection issues
05Ongoing maintenance
This automation is low-maintenance once live, but a small amount of regular attention keeps it accurate and effective. The table below sets out what to check, how often, and who is responsible.
Whenever a Notion article is added, edited, or removed
Update the relevant FAQ article in Notion using the same heading and tagging conventions established during the knowledge base audit. The First Response Drafting Agent pulls content based on article structure, so inconsistent formatting will reduce reply quality. After any major update, send a test ticket on the affected topic to confirm the agent still retrieves the right article.
Notion knowledge base editor
Whenever a reply template or tone needs to change
Contact support@gofullspec.com with the specific template or prompt change required. Do not edit agent prompt configurations directly unless the FullSpec team has given explicit guidance. Uncoordinated prompt edits are the most common cause of degraded reply quality.
Process owner, with FullSpec support
Monthly spot-check (first week of each month)
Pull the last 30 days of auto-resolved tickets from Zendesk and review a random sample of 10. Confirm the replies were accurate, on-brand, and matched the right Notion article. Note any categories where quality has dropped and flag them to FullSpec.
Support Team Lead
Monthly error-log review (first week of each month)
Review the automation error log (accessible via the workflow dashboard) for failed API calls, missed HubSpot logs, or Slack alerts that did not fire. Flag anything recurring to support@gofullspec.com. A single failed run is normal; the same error appearing more than three times in a month needs investigation.
Process owner
When a team member joins or leaves
Update the contact table in this runbook with the new team member's details. If the departing team member held Zendesk, Notion, HubSpot, or Slack admin access relevant to the automation, update API credentials and access permissions before their account is removed. Notify support@gofullspec.com if any connected credentials need to be rotated.
Operations Manager, with FullSpec support for credential rotation
Quarterly volume review (every 3 months)
Compare current monthly ticket volume against the baseline of 220 tickets/month used to calibrate the automation. If volume has increased by more than 30%, review whether the confidence threshold and Notion coverage are still appropriate for the new mix of ticket types. Contact FullSpec to discuss re-calibration if needed.
Operations Manager, with FullSpec support
The most common maintenance issue for this process is Notion knowledge base drift. When FAQ articles are updated in Notion without following the agreed heading and tagging structure, the First Response Drafting Agent fails to retrieve the correct content and either sends an inaccurate reply or downgrades the ticket to the flagged queue unnecessarily. Assign one named editor to own all Notion FAQ updates and brief them on the article format conventions before any content changes are made.
Process Runbook / SOPPage 4 of 4