The top 10 ways support teams use AI workflows
Last Updated
Published On
Every workflow below is a template in Plain, built and run by real B2B support teams. Because Plain is open infrastructure, anything you can do by hand can run through a workflow. Install one in a click, then change it to fit how your team works.
We reviewed more than 1,500 support workflows built in Plain, with close to 2 million recorded executions, spanning everything from straightforward routing rules to an AI agent writing and scheduling a weekly report in Slack.
Eight of the ten most common workflows below use an AI condition, meaning the workflow acts on a question the AI answers by reading the thread instead of a field it matches. The other two match on a field and put an AI agent in the action. Both sit in the same builder as our agents and our API, so one workflow can read the message, decide what it is, act, and hand the thread to an agent.
These ten are where teams start, but they’re not the limit.
What teams build with AI conditions
From 1,500+ workflows built in Plain. The categories are broad on purpose. A status update covers closing a thread and other state changes, not closing alone.
What the workflow does | Share of workspaces using AI conditions |
|---|---|
Updates status, including closing threads | ~55% |
Applies a label | ~45% |
Assigns to a user, team, or agent | ~40% |
Changes priority | ~35% |
1. Hand the thread to an AI agent
The most common AI workflow in support is also the biggest change in how teams operate. Instead of using AI to help a human answer, teams let an agent own the conversation from the start. It handles the repetitive, high-volume questions and only brings in a person when necessary. For teams with enough volume, that clears a meaningful share of the queue before anyone sees it.
How it works in Plain: New thread → condition: channel is Slack → assign to Plain’s Customer Agent. Teams who want to be more selective add an AI condition first, for example “Is this a question our help center already answers?”
Plain
Hand the thread to an AI agent
2. Close the finished threads
A customer replies “thanks, that worked” and reopens a closed thread. Another writes “thanks, but I still can’t log in.” A keyword rule matches both the same way. An AI condition tells a closing acknowledgement apart from a reply that needs an answer.
How it works in Plain: New customer message → confirms the issue is resolved with no remaining question → set status to Done. If it does not match, the thread stays open.
3. Start an investigation before a human opens the thread
Support teams have built dozens of workflows that start a session with Plain’s Internal Agent, recording tens of thousands of workflow runs.
The agent searches your connected tools for related tickets, past incidents, engineering discussions, and documentation, then leaves a single note on the thread listing what it found. The first person to open the thread starts from that note instead of a blank page. If you want a human responsible for the reply, this is the most direct place to start.
How it works in Plain: New thread → start an Internal Agent investigation → the Internal Agent posts one note on the thread with what it found. No condition, so it runs on every new thread. Add a label or channel condition if you want.
Plain
Start an investigation before a human opens the thread
4. Catch the frustrated customer
Anger rarely announces itself in words a keyword filter can catch. It shows up as “this is the third time I’ve asked,” or a cancellation threat buried in an otherwise polite message. An AI condition recognizes those signals and gets the thread in front of a human. Tone identifies a thread worth reviewing, while customer impact determines the response.
How it works in Plain: New customer message → AI condition: “Does the customer express frustration about repeated attempts to get help?” → if yes, raise priority for human review.
5. Find customers who cannot use the product
Being mildly inconvenienced and being completely blocked are different problems, and the difference is usually stated in the message. Teams check for outages, access failures, and core functionality that has stopped working, then raise priority on what matches.
How it works in Plain: New thread → AI checks whether the customer cannot complete a core task → set priority to Urgent. Define the condition around impact and available workarounds.
Plain
Find customers who cannot use the product
6. Spot & route sales and expansion intent
Support inboxes catch a lot of revenue signals. Customers ask about adding seats, changing plans, or renewing. An AI condition identifies those and routes them to an owner. The distinction that matters is intent, because a customer asking to buy more needs a different next step from one disputing an invoice.
How it works in Plain: New customer message → AI condition: “Is the customer asking about account expansion?” → if yes, assign to the sales team.
7. Route to the specialist team
Channel and account data don’t always tell you who should handle a request. A general support message may contain a security disclosure, a privacy request, or a billing dispute. Use the content of the request to identify the owner, and set a fallback when the destination isn’t clear.
How it works in Plain: New thread → AI condition: “Does this request need a billing specialist, such as an invoice dispute or refund request?” → if yes, assign to the billing team.
Plain
Route to the specialist team
8. Label the thread by topic
Consistent topics make reporting, saved views, and further automation more useful.
They also show the team what customers keep asking about.
How it works in Plain: New thread → Apply multiple AI conditions to classify the request as Feature Request, Bug report, or Question → applies the matching label. Threads that don’t fit any category are left unlabelled.
9. Find customer champions and advocates
Support hears the best things anyone says about your product, and almost none of it leaves the ticket.
This one runs on a schedule rather than on a message. Every day, Sidekick, Plain’s Internal AI Agent, reads the tickets your team resolved in the last 24 hours and looks for a quote worth using, an unprompted comparison against a competitor, or real enthusiasm rather than a polite thank you. It posts what it finds to Slack, so marketing and sales can use it in real-time (if customer approved).
How it works in Plain: Daily schedule → Sidekick reviews threads resolved in the last 24 hours → posts what it found to Slack.
Plain
Find customer champions and advocates
10. Keep spam and noise out of the queue
Cold outreach and automated notifications compete with real requests for attention. An AI condition identifies messages that need no support response and mark them ignored.
The boundary matters, because a legitimate customer asking about a topic that shows up in unsolicited pitches should still reach the team.
How it works in Plain: New thread → AI condition: “Is this unsolicited promotional outreach with no genuine customer support request?” → if yes, set status to Ignored.
What the same builder does once you’re past the basics
A report that writes itself
Every Monday at 9am the workflow fires, the Internal Agent reviews the week’s open threads, and posts a summary in Slack before anyone’s online. Workflows run on a schedule, not only on a customer message.
A workflow that calls your own services
The HTTP request action means a workflow can push an update to a CRM record, kick off a build, or log to a dashboard. Anything you can do by hand in Plain also runs through the API.
Your agent, not only ours
Limio built a workflow that uses the Internal Agent to filter their documentation and return the docs that answer a customer’s question.
One place, not two systems to keep in sync
Other tools split routing rules from AI settings. In Plain you start an Internal Agent session or assign to the Customer Agent from inside the workflow, in the same place you configure the agents.
And if the workflow you want isn’t in the library, describe it in plain text and the Internal Agent will build it for you.
Start with the problem you can measure
Choose a recurring decision, then check whether the workflow improves it.
If your team struggles with
Start with
Review
Critical customer issues getting buried in the queue
AI detection of outages or access failures
Time to human response on critical issues, missed escalations, false alarms
Support team spending too long to piece together context before they can help
Start a Sidekick discussion when a technical request arrives
Time to a useful response and how often your team can use the findings
Customers repeating themselves as requests bounce between teams
AI detection of the expertise a request needs
Reassignments per thread and time to reach the right owner
Browse the workflow library
Pick one workflow above or browse our workflow library and build it in Plain: choose the trigger, define the condition, and connect the action.
Not using Plain yet? Start a free trial and install any of these on day one.
Every workflow below is a template in Plain, built and run by real B2B support teams. Because Plain is open infrastructure, anything you can do by hand can run through a workflow. Install one in a click, then change it to fit how your team works.
We reviewed more than 1,500 support workflows built in Plain, with close to 2 million recorded executions, spanning everything from straightforward routing rules to an AI agent writing and scheduling a weekly report in Slack.
Eight of the ten most common workflows below use an AI condition, meaning the workflow acts on a question the AI answers by reading the thread instead of a field it matches. The other two match on a field and put an AI agent in the action. Both sit in the same builder as our agents and our API, so one workflow can read the message, decide what it is, act, and hand the thread to an agent.
These ten are where teams start, but they’re not the limit.
What teams build with AI conditions
From 1,500+ workflows built in Plain. The categories are broad on purpose. A status update covers closing a thread and other state changes, not closing alone.
What the workflow does | Share of workspaces using AI conditions |
|---|---|
Updates status, including closing threads | ~55% |
Applies a label | ~45% |
Assigns to a user, team, or agent | ~40% |
Changes priority | ~35% |
1. Hand the thread to an AI agent
The most common AI workflow in support is also the biggest change in how teams operate. Instead of using AI to help a human answer, teams let an agent own the conversation from the start. It handles the repetitive, high-volume questions and only brings in a person when necessary. For teams with enough volume, that clears a meaningful share of the queue before anyone sees it.
How it works in Plain: New thread → condition: channel is Slack → assign to Plain’s Customer Agent. Teams who want to be more selective add an AI condition first, for example “Is this a question our help center already answers?”
Plain
Hand the thread to an AI agent
2. Close the finished threads
A customer replies “thanks, that worked” and reopens a closed thread. Another writes “thanks, but I still can’t log in.” A keyword rule matches both the same way. An AI condition tells a closing acknowledgement apart from a reply that needs an answer.
How it works in Plain: New customer message → confirms the issue is resolved with no remaining question → set status to Done. If it does not match, the thread stays open.
3. Start an investigation before a human opens the thread
Support teams have built dozens of workflows that start a session with Plain’s Internal Agent, recording tens of thousands of workflow runs.
The agent searches your connected tools for related tickets, past incidents, engineering discussions, and documentation, then leaves a single note on the thread listing what it found. The first person to open the thread starts from that note instead of a blank page. If you want a human responsible for the reply, this is the most direct place to start.
How it works in Plain: New thread → start an Internal Agent investigation → the Internal Agent posts one note on the thread with what it found. No condition, so it runs on every new thread. Add a label or channel condition if you want.
Plain
Start an investigation before a human opens the thread
4. Catch the frustrated customer
Anger rarely announces itself in words a keyword filter can catch. It shows up as “this is the third time I’ve asked,” or a cancellation threat buried in an otherwise polite message. An AI condition recognizes those signals and gets the thread in front of a human. Tone identifies a thread worth reviewing, while customer impact determines the response.
How it works in Plain: New customer message → AI condition: “Does the customer express frustration about repeated attempts to get help?” → if yes, raise priority for human review.
5. Find customers who cannot use the product
Being mildly inconvenienced and being completely blocked are different problems, and the difference is usually stated in the message. Teams check for outages, access failures, and core functionality that has stopped working, then raise priority on what matches.
How it works in Plain: New thread → AI checks whether the customer cannot complete a core task → set priority to Urgent. Define the condition around impact and available workarounds.
Plain
Find customers who cannot use the product
6. Spot & route sales and expansion intent
Support inboxes catch a lot of revenue signals. Customers ask about adding seats, changing plans, or renewing. An AI condition identifies those and routes them to an owner. The distinction that matters is intent, because a customer asking to buy more needs a different next step from one disputing an invoice.
How it works in Plain: New customer message → AI condition: “Is the customer asking about account expansion?” → if yes, assign to the sales team.
7. Route to the specialist team
Channel and account data don’t always tell you who should handle a request. A general support message may contain a security disclosure, a privacy request, or a billing dispute. Use the content of the request to identify the owner, and set a fallback when the destination isn’t clear.
How it works in Plain: New thread → AI condition: “Does this request need a billing specialist, such as an invoice dispute or refund request?” → if yes, assign to the billing team.
Plain
Route to the specialist team
8. Label the thread by topic
Consistent topics make reporting, saved views, and further automation more useful.
They also show the team what customers keep asking about.
How it works in Plain: New thread → Apply multiple AI conditions to classify the request as Feature Request, Bug report, or Question → applies the matching label. Threads that don’t fit any category are left unlabelled.
9. Find customer champions and advocates
Support hears the best things anyone says about your product, and almost none of it leaves the ticket.
This one runs on a schedule rather than on a message. Every day, Sidekick, Plain’s Internal AI Agent, reads the tickets your team resolved in the last 24 hours and looks for a quote worth using, an unprompted comparison against a competitor, or real enthusiasm rather than a polite thank you. It posts what it finds to Slack, so marketing and sales can use it in real-time (if customer approved).
How it works in Plain: Daily schedule → Sidekick reviews threads resolved in the last 24 hours → posts what it found to Slack.
Plain
Find customer champions and advocates
10. Keep spam and noise out of the queue
Cold outreach and automated notifications compete with real requests for attention. An AI condition identifies messages that need no support response and mark them ignored.
The boundary matters, because a legitimate customer asking about a topic that shows up in unsolicited pitches should still reach the team.
How it works in Plain: New thread → AI condition: “Is this unsolicited promotional outreach with no genuine customer support request?” → if yes, set status to Ignored.
What the same builder does once you’re past the basics
A report that writes itself
Every Monday at 9am the workflow fires, the Internal Agent reviews the week’s open threads, and posts a summary in Slack before anyone’s online. Workflows run on a schedule, not only on a customer message.
A workflow that calls your own services
The HTTP request action means a workflow can push an update to a CRM record, kick off a build, or log to a dashboard. Anything you can do by hand in Plain also runs through the API.
Your agent, not only ours
Limio built a workflow that uses the Internal Agent to filter their documentation and return the docs that answer a customer’s question.
One place, not two systems to keep in sync
Other tools split routing rules from AI settings. In Plain you start an Internal Agent session or assign to the Customer Agent from inside the workflow, in the same place you configure the agents.
And if the workflow you want isn’t in the library, describe it in plain text and the Internal Agent will build it for you.
Start with the problem you can measure
Choose a recurring decision, then check whether the workflow improves it.
If your team struggles with
Start with
Review
Critical customer issues getting buried in the queue
AI detection of outages or access failures
Time to human response on critical issues, missed escalations, false alarms
Support team spending too long to piece together context before they can help
Start a Sidekick discussion when a technical request arrives
Time to a useful response and how often your team can use the findings
Customers repeating themselves as requests bounce between teams
AI detection of the expertise a request needs
Reassignments per thread and time to reach the right owner
Browse the workflow library
Pick one workflow above or browse our workflow library and build it in Plain: choose the trigger, define the condition, and connect the action.
Not using Plain yet? Start a free trial and install any of these on day one.